Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Will Bertrand
Citation for published version (APA):Houtum, van, G. J. J. A. N., Kok, de, A. G., & Ooijen, van, H. P. G. (editors) (2011). Will Bertrand: passievollogistiek ontwerper. Oisterwijk: Uitgeverij BOXPress.
Document status and date:Gepubliceerd: 01/01/2011
Document Version:Uitgevers PDF, ook bekend als Version of Record
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can beimportant differences between the submitted version and the official published version of record. Peopleinterested in the research are advised to contact the author for the final version of the publication, or visit theDOI to the publisher's website.• The final author version and the galley proof are versions of the publication after peer review.• The final published version features the final layout of the paper including the volume, issue and pagenumbers.Link to publication
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, pleasefollow below link for the End User Agreement:www.tue.nl/taverne
Take down policyIf you believe that this document breaches copyright please contact us at:[email protected] details and we will investigate your claim.
Download date: 22. Jun. 2020
WILL BERTRAND:
PASSIEVOL LOGISTIEK ONTWERPER
Liber Amicorum voor Will Bertrand
Redactie:
Ton de Kok Geert-Jan van Houtum Henny van Ooijen
INHOUD
VOORWOORD
A MATTER OF PERSISTENCE 1
Henk Zijm
BERTRAND, DE ONDERNEMENDE PROFESSOR 5
Paul Gosselink
WILL BERTRAND ALS DIRECTEUR VAN LMS 11
Twan Geenen
DE PROCESONTWIKKELING VAN DE KLM IN DE AFGELOPEN
12 JAAR 23
Peter Bos
A REFERENCE MODEL FOR THE DESIGN OF OPERATIONS
PLANNING AND CONTROL SYSTEMS IN THE FLOW PROCESS
INDUSTRY 39
Jan Fransoo
OPERATIONS MANAGEMENT IN HEALTH CARE 55
Jan Vissers en Guus de Vries
THE EUT MAINTENANCE RESEARCH 67
Geert-Jan van Houtum
A TYPICAL AND HIERARCHICAL WORKLOAD-ORIENTED
APPROACH IN EDUCATION 79
Corné Dirne
THE IMPACT OF NON-COMPLIANCE TO PRODUCION CONTROL
PRINCIPLES ON PERFORMANCE OF REAL-LIFE PRODUCTION
SYSTEMS 95
Joris Keizers
WHAT AND HOW OF PLANNING AND CONTROL OF OPERATIONAL
PROCESSES 105
Ton de Kok
IN THE GALLERY OF CELEBRITIES 119
Marc Lambrecht
EEN EXCURSIE DOOR EEN ORDER ACCEPTATIE MODEL EN HET
GEBRUIK DAARVAN TEN BEHOEVE VAN HET ONTWERPEN VAN
DE BEHEERSING VAN EEN ÉÉN-FASE PRODUCTIE-VOORRAAD-
SYSTEEM 127
Jacob Wijngaard
WILL BERTRAND AND DOUBLE MATCHING QUEUES 139
Onno Boxma
ON THE ONSET OF WEAK MONOTONICITY RESULTS ON LATTICE
FRAGMENTS 149
Andreea Dragut
PRODUCTIEBEHEERSING: GROEI NAAR VOLWASSENHEID 169
Will Bertrand
THE STRUCTURING OF PRODUCTION CONTROL SYSTEMS 187
Will Bertrand and Jacob Wijngaard
A NOTE ON THE DESIGN OF LOGISTIC CONTROL SYSTEMS 207
Will Bertrand
Lijst met promovendi met Bertrand als eerste promotor 221
Lijst met afgestudeerden met Bertrand als eerste begeleider 223
Lijst met ontwerpers met Bertrand als eerste begeleider 229
Auteursgegevens 233
VOORWOORD
Wim Monhemius schrijft ons naar aanleiding van onze uitnodiging om bij te dragen aan dit
Liber Amicorum het volgende:
Will Bertrand gaat in 2011 de faculteit verlaten; de faculteit waarvan ik in 1989 afscheid nam
en die toen de naam droeg: ,,Faculteit Technische Bedrijfskunde”. In de periode van 22 jaren
tussen toen en nu is er veel gebeurd, in die faculteit maar ook op het gebied van de
productiebeheersing. Welke gebeurtenissen dat waren en wat het belang ervan was heb ik om
allerlei redenen nauwelijks waargenomen en ik heb daarover geen enkel oordeel. Maar wel wil
ik terugblikken op het verschijnen in 1990 van twee boeken die in 1989 al ,, bijna rond” waren.
Dat waren,”Productiebeheersing en Material Management” en “Production control; a
structural and design oriented approach”. Vooral het verschijnen van dat laatste boek vond ik
zeer belangrijk. Naar mijn mening was daarmee namelijk vastgesteld dat productiebeheersing
waarlijk een ingenieursvak is, waarbij een wetenschappelijke benadering van groot belang is.
Systematisch ontwerpen, gesteund door multidisciplinaire analyses, leidend tot concrete
oplossingen voor concrete problemen.
Met het verschijnen van dit boek had mijns inziens Will Bertrand, samen met Hans Wortmann
en Jacob Wijngaard een stevig, voortreffelijk fundament gelegd. Deze auteurs waren indertijd
alle drie hoogleraar aan de Faculteit Technische Bedrijfskunde te Eindhoven. Productie-
beheersing is een ingenieursvak, goed passend in een opleiding tot bedrijfskundig ingenieur. De
disciplines die in de analyse van productiebeheersing een rol spelen, zijn vooral, zo blijkt uit het
boek van Bertrand e.a., toegepaste wiskunde en statistiek, organisatiekunde, bedrijfseconomie
en accountancy, bedrijfsinformatica en organisatiepsychologie. Het ontwerp moet leiden tot een
doelgericht en doelmatig beheersingssysteem. Gebaseerd op wetenschappelijk inzicht, gericht
op de praktijk.
Voor mij was daarmee overtuigend gedemonstreerd dat bedrijfskunde een technische studie-
richting op academisch niveau kan zijn.
Mijn eigen opleiding in de werktuigbouwkunde vond plaats tientallen jaren eerder aan de toen
geheten Technische Hogeschool te Delft. Daar had ik het belang geleerd van systematisch
ontwerpen, gebaseerd op en gebruik makend van wetenschappelijke methoden en resultaten. In
het bijzonder de colleges van prof. dr. ir. Biezeno en van prof. Ir. van Hasselt hadden op mij
grote indruk gemaakt.
Het proces van wetenschappelijk productief samen bezig zijn om het boek “Production
Control” te schrijven deed mij denken aan enkele regels van Rudyard Kipling:
If you can think and not make thoughts your aim, …. maar ook
If you can walk with kings nor lose the common touch, of
If you can talk with crowds and keep your virtue …..
Kortom, ik ben er trots op, indertijd met Will Bertrand te hebben samengewerkt.
W. Monhemius
Het leek ons als redactie van dit Liber Amicorum, ter gelegenheid van het afscheid van Will
Bertrand als hoogleraar aan de (sub) faculteit Technische Bedrijfskunde, gepast om als eerste de
nog in leven zijnde promotor van Will Bertrand, Prof. Em. W. Monhemius, aan het woord te
laten.
In 1964 zette Will Bertrand zijn eerste stappen op de Technische Universiteit Eindhoven (toen
nog Technische Hogeschool Eindhoven) als student werktuigbouwkunde. Toen in 1966 de
afdeling Technische Bedrijfskunde werd opgezet, waar studenten die hun propedeuse
Werktuigbouwkunde hadden konden instromen, stapte hij over naar deze in oprichting zijnde
afdeling. Hier studeerde hij in 1970 af met een opdracht bij metaalwarenfabriek N.V. Simmonds
Precision in Brummen. In 1981 promoveerde hij, samen met Hans Wortmann, op het
proefschrift “Production Control and Information Systems for Component Manufacturing
Shops”. In dit onderzoek, waarin een aantal principes voor het ontwerpen van productie-
beheersing- en informatie systemen worden ontwikkeld, kwam al duidelijk naar voren dat Will
“ontwerpen” in zijn hart had gesloten. Dit ontwerpen heeft hij nooit meer losgelaten.
Will is de Technische Universiteit Eindhoven altijd trouw gebleven en heeft alle rangen
doorlopen: van student tot hoogleraar. Daarnaast vervulde hij ook diverse functies in het
bedrijfsleven. Van 1984 tot 1985 was hij hoofd organisatieontwikkeling en logistiek van ASML
Lithopgraphy in Veldhoven en van 1985 tot 1988 hoofd Logistiek bij de Philips Machine
fabrieken in Eindhoven. Gedurende de jaren in het bedrijfsleven had Will echter toch altijd een
gedeeltelijke aanstelling bij de TUE. In 1988 werd hij aangesteld als voltijds hoogleraar aan de
Technische Hogeschool (nu Technische Universiteit) Eindhoven. In 1986 startte de Technische
Hogeschool met de 2-jarige nadoctorale ontwerpers opleiding Beheersing van Bedrijfs-
processen. Will was daar in het begin blokdocent en vooral betrokken bij de ontwerp-
componenten waarbij zijn hart met name lag bij het praktijkgedeelte. Daarin speelde het echte
ontwerpen een belangrijke rol en moet de cursist laten zien dat hij het vak beheerst; dit had
altijd zijn volle aandacht.
In de jaren ‘90 was hij nauw betrokken bij de opzet (het “ontwerp”) van de research school
BETA en hij was van 1996 tot 2003 de eerste wetenschappelijk directeur van deze research
school. Hiermee heeft hij een centrale rol gespeeld in de het omvormen van de faculteit
Technische Bedrijfskunde van een professioneel gerichte organisatie naar een wetenschappelijk
gerichte organisatie. Logistiek is uiteindelijk uitgegroeid tot één van de zogenaamde profilerings
gebieden van de TUE.
Will’s leidraad is altijd kwaliteit geweest en hij heeft altijd geopereerd op basis van een
inhoudelijke visie, een visie die gebaseerd is op de kernrol van de TUE: ontwerpend onderzoek.
De ingenieursaanpak is dan ook kenmerkend voor zijn, zowel theoretisch als ook toegepast,
onderzoek. In dit Liber Amicorum vinden we dit in alle bijdragen terug. We kunnen deze
bijdragen ruwweg in een drietal categorieën onderscheiden:
- Bouwen aan Bedrijfskunde, waar de bijdragen wat persoonlijker van aard zijn
- Toepassen van Bedrijfskunde
- Ontwikkelen van modellen
Een kenmerk van goede wetenschap is dat zij nooit haar actualiteit verliest. Wij menen dat
Will’s werk een voorbeeld is van goede wetenschap. Daarom hebben wij besloten een aantal
van Will’s meest veelomvattende conceptuele bijdragen in deze bundel op te nemen. Het
onderwerp van deze bijdragen betreft Will’s passie: het ontwerpen van logistieke beheersings-
systemen. De onderstaande drie bijdragen geven een beeld van de ontwikkeling van Will’s
denken met betrekking tot dit onderwerp:
- Will’s intreerede
- Het samen met Jacob Wijngaard in IJOPM gepubliceerde artikel over de structuur van
productiebeheersingssystemen
- Zijn notitie over het ontwerpen van logistieke beheersingssystemen, een notitie die de
basis is van het vak Design of Operational Planning and Control Systems. Deze notitie
is ook de aanleiding geweest voor de ontwikkeling (door Incontrol) van een software
pakket dat uitgaat van processen zoals gedefinieerd in deze notitie en ondersteunend is
voor het ontwerpen van logistieke beheersingssystemen.
De redactie
1
A MATTER OF PERSISTENCE
Henk Zijm
When I was asked to write a personal memory at the occasion of the retirement of our friend and
colleague Will Bertrand, my first feeling was one of hesitation. Not that I would find it hard to
respond, but merely since I was wondering whether Will would embrace the idea of a collection
of tributes. He has never been the kind of person who placed himself on the foreground, but
nevertheless has built a name not only in science, but merely (and I believe to his joy) in
industrial practice. No doubt Will has made significant contributions to the science of industrial
engineering, but I can’t help still seeing him in the first place as a bridge builder, between an
awakening science and the practice of operations management. It is this ability to strengthen ties
between various worlds that made him a key factor in expanding the position of Eindhoven
University of Technology, and to establish the name it currently holds in the field of Industrial
Engineering.
In this personal account, I will very briefly go back to the first encounters with my colleagues
from what is now named the School of Industrial Engineering, but which, in search for its
identity, went through a number of profiles. The first person I met there was not Will, but Jacob
Wijngaard, at that time an associate professor but also a former PhD of my supervisor Jaap
Wessels in the math department, where I did my PhD as well. Will at that time was working on
the theory and practice of workload control which would lead in 1981 to a joint dissertation with
Hans Wortmann who explored its IT ramifications. The book (it was also published as a
textbook) drew quite some attention from practice, where both Will and Hans got their primary
inspiration as well. Will, Hans and Jacob worked out these and other notions in their book on
production control, as well as in Dutch language books, which still are remarkable because they
did not just present theoretical foundations but, other than similar books, also described
systematic approaches on how to implement such systems in practice. They were not the only
ones; an early prophet of workload control in Germany was Hans-Peter Wiendahl (although not
very well known at that time), later Eliyahu Goldratt popularized the notion in The Goal, while
concepts like CONWIP and others were elaborated in books such as Factory Physics by Wally
Hopp and Mark Spearman. It should be said that not all the latter contributions sufficiently
acknowledged the pioneering role that the Dutch school, and in particular Will, had played in
developing these concepts and in bringing them to maturity. This experience certainly made
Will realize once more that, in order to gain full recognition, one simply has to play on two
2
different chess boards, that of practice and that of science. Which is what he did, although I
believe that, till this day, Will’s heart is basically at the side of practice, even more, of industrial
practice (despite his steps aside in the world of health care).
During my first period at the Eindhoven University of Technology, we did not have much
contact but that started to change when, after completing my PhD in early 1982 and
subsequently a short period back in Amsterdam, I joined Philips Electronics in Eindhoven in
1983, at walking distance of the university.
Natural contacts remained with my former supervisor Jaap Wessels but through him and Jacob
ties with Industrial Engineering were strengthened as well. Long before joining Philips, I had
become familiar with the two-volume book on the theory and practice of production and
inventory control, by Ruud van Hees and Wim Monhemius, the first one still at Philips when I
joined it, the latter one then a professor at the Eindhoven University of Technology, and one of
Will’s supervisors on his PhD research. Since I was working primarily in production logistics at
Philips at that time, I became naturally familiar with Will’s work, also because he practiced it at
the same time while being employed as head of the Logistics Division of Philips Machine
Factories during these years. In 1987, I was appointed as part time professor at the TU/e
Department of Mathematics and Computer Science, one year later Will received his
appointment as a full professor in Production Logistics at the Industrial Engineering
Department. So, although we still did not have many personal contacts until then, a future
collaboration seemed natural.
But things went different. In 1990, both Jacob and I left Eindhoven. Jacob accepted a position as
a full professor at the University of Groningen, and I left both Philips and the TU/e to become a
full professor in Production and Operations Management at the University of Twente. Ties with
Eindhoven remained, and were strengthened after a visit in 1994 of Will and Ton de Kok (who
was just appointed in Eindhoven at that time), together with some group members of Industrial
Engineering, to my group Production and Operations Management at the Mechanical
Engineering Faculty of the University of Twente. In the early nineties also the notion of national
PhD research schools came to life and we all felt this to be the chance to definitively establish
Operations Management and Industrial Engineering as a mature scientific discipline.
In what follows, I draw some bits and pieces from a speech that I held in June 2008 at a festive
meeting of the joint Eindhoven-Twente research school Beta. Will has been more than
instrumental in building Beta, he was the driving force behind its establishment and that was
both remarkable and necessary. Remarkable because Will was in the first place a man who
believed in transferring knowledge to society. And necessary because he was one of the few
people who had a clear vision on the multi-disciplinary nature of the field, and wanted to include
3
the full breadth of the disciplines. At the same time, however, he realized that, similar to
international scientific acknowledgements, a full recognition of the field of Operations
Management and Industrial Engineering by the scientific community in the Netherlands was
indispensible. This concerned in particular organisations such as the Royal Netherlands
Academy of Sciences (KNAW) and the Netherlands Scientific Research Council (NWO), the
former being responsible for the formal accreditation of Dutch Research Schools.
The first attempt was not an immediate success, not in the last place because established
scientific bodies in the Netherlands at that time (and still today) have difficulties in truly
valueing highly interdisciplinary research. The first version of Beta indeed became a school with
a very wide scope, so wide that some researchers were wondering what was the tie that held
them together.
That school was not accredited at a first attempt, partly for the reason already mentioned, but
partly also because there was quite some diversity in quality of the composing groups, and I
believe it was a wise and certainly a courageous decision to limit the scope of Beta somewhat
and to concentrate more on quality.
To lead such a process of rethinking scope and scale at such a still embryonic stage is a delicate
process. Establishing a research school in those days when people could not fully envision the
role and power of such schools is already hard in itself. Everyone wants to jump on the
bandwagon, hoping that once you’re in, the ships with money will sail in more or less
automatically. Excluding researchers who are at the same time friends and colleagues naturally
is quite a dramatic decision. And of course the person who is leading the process is the first one
to blame, not only by the excluded but even by the included, once they discover that the
perceived ship with money is sailing terribly slow. It takes a lot of patience, endurance and a
strong and sustainable believe in the outcome of the process to keep going, and I still feel
admiration for the way in which Will led the process to a successful end. Beta was established as
a research school in its current form in 1995, received accreditation and became highly
successful. Of course, there are many persons who deserve credits for that, but among them Will
is still the first one.
In 1999, I returned for one year to the Eindhoven University of Technology but family reasons
prevented me to move definitively. Beta as a research school already at that time played a
dominant role, as was also recognized by the Executive Board of Eindhoven University, who
granted Beta the right to award an honorary doctorate to John Buzacott, the first one ever in the
Netherlands in the field of Industrial Engineering. For Beta (and for me personally, acting as the
honorary supervisor), the ceremony but certainly also the seminar organized by Beta at that
occasion and the book resulting from it, marked a further step ahead in building a reputation.
4
The role Beta is currently playing in the Supply Chain Thought Leaders Forum (for which
Will’s successor as scientific director of Beta, Ton de Kok, deserves credits) is a further
indication of worldwide recognition. The recent foundation of the Dutch Institute of Advanced
Logistics (DINALOG) and the key role Beta has played in its establishment again contributes to
that recognition as a mature scientific institute.
The challenge to build on what has been established so far however remains big, and it is hard to
predict what will be the impact of current developments in Dutch politics with respect to
disciplines such as Industrial Engineering. History repeats itself; also the current debate is partly
on the delicate balance between scientific rigor and practical relevance, two concepts that should
go hand in hand but unfortunately too often are placed opposite of each other. In particular,
Industrial Engineering and Management Science have faced that challenge, far more than
traditional disciplines such as Physics or Chemistry with a rich tradition in industrial research as
well. But Industrial Engineering has made a significant step, and the role Will has played in
paving that road is undeniable.
What is too often forgotten is the price you pay for perseverance and endurance. What is
forgotten are the disappointments that we all encounter now and then and of which also Will has
received his share. What remains are the things to cherish: a sound scientific career, a great
feeling for industrial relevance, and above all a contribution that changed the landscape of
Industrial Engineering definitively.
What lies ahead is a future with hopefully some more time for all the things that had to be
postponed so long. Will, thanks for all your contributions, your inspiration and encouragement. I
wish you all the best.
5
BERTRAND, DE ONDERNEMENDE
PROFESSOR
P.B.A. Gosselink
Ik ben vereerd een kleine input te leveren aan het boek dat de periode van Will Bertrand aan de
Technische Universiteit Eindhoven beschrijft. Meer specifiek de faculteit Technische
Bedrijfskunde en de afdeling Logistiek. Met minimaal de volgende afkortingen: KBS, LBS,
OPC, OPAC. Een kleine bijdrage aan zijn afscheid kent wat mij betreft 2 fasen:
1. Mijn universiteitsperiode en zijn input (1983-1988)
2. ITP TUE-TNO (instituut gestart in 1986-2004 voor mij)
In die verschillende perioden hebben zijn en mijn (professionele) levens elkaar een aantal keer
gekruist! Van een beperkte interactie liep dat, in elk geval vanuit mij bezien, naar een
intensieve en leerzame samenwerking!!
1. Mijn universiteitsperiode en zijn input
Ik startte september 1983 met de opleiding Technische Bedrijfskunde. Wat me daar van vooral
nog is bijgebleven is dat Bertrand de (voor ons) moeilijke en beruchte vakken “Inleiding
Productiebeheersing” en “Beheersing dynamische processen in de bedrijfsvoering“ gaf. Nu
terugdenkend viel op dat hij worstelde met en ‘leed’ aan het lesgeven aan grote groepen waar
lang niet altijd het enthousiasme voor het Logistieke vak uit bleek. Veel rumoer en
ongeïnteresseerdheid van het publiek leidde tot een vermoeid gezicht bij hem en hij moest zich
er echt toe zetten ‘les te geven’. De punten van ons waren er ook naar! Daarna ben ik nog een
tijd student-assistent bij hem geweest om de homologatiefase voor de 2e fase
ontwerpersopleiding ‘Logistics Management Systems’ mede vorm te geven.
Ten slotte was hij mijn afstudeerbegeleider. Ik studeerde af bij het golfkarton productiebedrijf
De Zeeuw in Eerbeek. De opdracht voor mij was het ontwerpen van een geïntegreerde en
optimale productieplanning voor de Golfkartonmachine en de nabewerkingen. Wat was
Bertrand streng (doch ook rechtvaardig)! En wat heb ik geworsteld! Het was niet altijd een fijne
beleving om eens per 2-3 weken bij hem mijn tussenverslag te moeten bespreken. Ik vond het
6
moeilijk om de theorie in de praktijk te brengen. Hoe moest ik nou de juiste formules vinden,
toepassen en vertalen voor de planning van de Zeeuw. Na mijn afstudeervoordracht kreeg ik 2
typen commentaren. Mijn familie en vrienden hadden niet veel begrepen van de presentatie en
het rapport en hielden vooral wat termen vast: ‘Go - No Go decision’, ‘kick-off meeting’ en
‘orderdoorlooptijd’. Bertrand vond het maar zozo en gaf het een 6. Een kleine teleurstelling
voor mij.
Later hebben we het er nog wel eens over gehad en was hij milder! Hij zei dat het echt theorie in
de praktijk brengen was en iets (praktisch) had toegevoegd aan de theorie. Iets wat eind jaren
’90 en begin 2000 niet meer voorkwam bij afstuderen. Dat gaf mij achteraf weer een beter
gevoel!
2. ITP TUE-TNO
In 1986 werd het Instituut Informatie-Technologie voor Productieautomatisering TUE-TNO,
ofwel ITP, opgericht. Bertrand was daar één van de grondleggers van (naast bijv. prof.
Bemelmans, prof. van Hee en prof. Rijnsdorp). Het ITP was tot 1994 gevestigd in Gebouw O
(nu afgebroken), tegenover het Paviljoen. Het was als instituut opgenomen in de TNO-
organisatie en viel onder de Hoofdgroep Technisch-Wetenschappelijke Diensten (TWD), met
hoofdzetel aan de Schoemakerstraat 97 te Delft. Het ITP richtte zich op geavanceerde
industriële computertoepassingen in de productie, logistiek en productiemanagement. Doel van
de dit gezamenlijke initiatief van TUE en TNO was de bij beide instellingen beschikbare kennis
en kunde op dat gebied verder te ontwikkelen en efficiënt in de praktijk te brengen. ITP deed
strategisch en toepassingsgericht onderzoek en voerde projecten uit op het gebied van:
o De stuksgewijze productie, zowel kleine serie als in massaproductie;
o De procesindustrie;
o De logistieke beheersing van de goederenstroom en de ondersteuning van het
productiemanagement.
In deze lijnen is de hand van Bertrand duidelijk herkenbaar.
ITP had vaste medewerkers die naar behoefte met TUE medewerkers, studenten en promovendi
werd uitgebreid (detachering op tijdelijke of projectbasis). Bertrand zat in de Wetenschappelijke
Raad van ITP welke de directie terzijde stond bij het opstellen en uitvoeren van het
onderzoekprogramma.
7
Het paste goed bij een andere kant van Will: de ondernemende professor! Dat was mede zijn
uitgangspunt om zijn kennis en kunde via ITP in de praktijk te brengen (en er wat geld aan te
verdienen!). Actief speelde hij een rol bij de oprichting en inhoudelijke invulling van ITP.
Wat hem vooral aantrok, volgens mij, was de theorie van ‘Capaciteitsmanagement en doorloop-
tijdbeheersing’ bij complexe seriematig werkende productiebedrijven in de praktijk te brengen.
Vandaar dat hij ook Hoofd Logistieke organisatie van de Philips Machinefabrieken is geweest
en later inhoudelijk adviseur bij onder andere ASML en Besi/Fico in Duiven/Zevenaar en actief
betrokken is geweest bij het opzetten en implementeren van productiebeheersing.
Wat mij betreft heeft in onze samenwerking tijdens ITP toch vooral ‘het logistiek concept
volgens de 4 kranen’ en ‘de werklastbeheersing’ centraal gestaan. Dit uitte zich in:
o De 2-daagse cursus Capaciteitsmanagement en doorlooptijdbeheersing in produktie-
afdelingen. Jaap Jagtenberg, Bertrand en ik waren verantwoordelijk voor de organisatie
en inhoud hiervan. De cursus werd 2x per jaar gegeven, via open inschrijving en
meestal in de Rosep te Oisterwijk. Voor iedereen die verantwoordelijkheid draagt voor
de goederenstroom in een productiebedrijf. De cursus richtte zich met name op de
enkelstuks- en seriematige fabricage.
Opvallend was altijd weer de verrassing bij de planners dat Werklastbeheersing tot
doorlooptijdverkorting en verhoging leverbetrouwbaarheid leidde, met een lagere
werklast. Dat laatste was meestal het grootste discussiepunt! Het ging zo tegen het
gevoel in om weinig werk op de werkvloer te hebben.
Tijdens het PC-spel ‘BeWe’ en in de avond bij de behandeling van de praktijksituaties
had Bertrand vaak hele interessante en vurige discussies met de planners. Daar voelde
hij zich als een vis in het water en genoot hij volop!
Later hebben we er nog een module rond organisatie (taken, verantwoordelijkheden,
bevoegdheden en rekenschap) aan toegevoegd om de toepassing in de praktijk te
vergemakkelijken.
Eind jaren ’90 was Werklastbeheersing volgens Bertrand wel ‘klaar’ qua theorie. Even
later zijn we ook gestopt met de cursussen en is de toepassing van Werklastbeheersing
in de praktijk stil komen te staan.
o Het Europese CRAFT-project Shopfloor Planning And Registration / SPAR. Hier
hebben we een prototype voor de PC ontwikkeld van een systeem voor shopfloor
planning en controle. Dit was gebaseerd op de principes van Werklastbeheersing en de
4 kranen. Een aantal bedrijven in Nederland, België en Portugal hebben het
8
daadwerkelijk gebruikt. Limis uit Enschede (spin-off van de Universiteit Twente) was
verantwoordelijk voor de computertoepassing en is erin geslaagd het gedachtegoed en
het softwaresysteem te vercommercialiseren. Meerdere systemen zijn er sinds 1994
verkocht en onderhouden. Het bedrijf bestaat nog steeds en heeft de functionaliteit flink
uitgebouwd.
o Advisering en toepassing Werklastbeheersing bij bedrijven. Enerzijds paste ik het zelf
toe bij vooral toeleveranciers in de metaalindustrie. Voorbeelden zijn tandwielen-
fabrikant Hankamp uit Enschede, fijnmetaalbedrijf Buhl uit Hilversum, fijnemechanisch
productiebedrijf Norma uit Hengelo, fabrikant van badkameraccessoires Haceka uit
Amersfoort, Philips Machinefabriek in Acht en metaalbedrijf Janssen uit Heijen.
Anderzijds schakelde Bertrand mij soms in bij bedrijven waar hij kwam. Leverancier
van equipement voor de halfgeleiderindustrie Fico uit Zevenaar was de meest
intensieve. Naast de Trim&Form fabriek uit Zevenaar kwam ik later ook nog in hun
Tooling fabriek in Brunssum en had ik samen met Bertrand enkele management sessies
met de directie van de groep over de goede toepassing van het Logistieke concept.
o Studentenbegeleiding (medio jaren 90 en rond 2005 enkele jaren); ik was enige tijd de
flexibele kracht voor OPAC afdeling op de TUE om afstudeerders te begeleiden. Een
leuke tijd om betrokken te blijven bij nieuwe theoretische ontwikkelingen en om jonge
mensen te laten zien hoe de theorie toe te passen in de praktijk. Wat een herkenning van
mijn eigen afstudeertijd!
ITP werd uiteindelijk niet het succes wat Bertrand (de professoren van de TU/e in het
algemeen?) ervan verwacht had. De eerste jaren ging het als een speer, binnen 4 jaar waren er
50 mensen werkzaam voor ITP. In 1990 liet TNO ITP echter fuseren met het Metaalinstituut
van TNO uit Apeldoorn tot IPL (Instituut voor Productie en Logistiek). Samen meer dan 200
mensen, op 2 locaties met 2 afdelingen: Logistiek Management en Productiesystemen. Het werd
echter geen eenheid. Herstellen van de oude situatie zat er vanuit TNO niet meer in. Een ADL-
onderzoek van TNO centraal leidde tot een filosofie waarin een researchclub van 50 personen
niet meer paste! De afdeling Productiesystemen werd overgeplaatst naar de Technisch
Physische Dienst van TNO in Delft. Logistiek Management werd in april ’92 verzelfstandig als
BV onder TNO Management BV’s, met nog 12 mensen. Eind ’94 verhuisde IPL van gebouw O
op de TUE campus naar De Run in Eindhoven (achter de 1e gebouwen van ASML).
In 2000 werd IPL helemaal verzelfstandigd van TNO via een management buy-out. Ik verliet
het IPL als werknemer en aandeelhouder in 2004 (voor de BOM). Tegenwoordig zit IPL met
circa 10 mensen in het Beta gebouw op de High Tech Campus in Eindhoven. Bertrand heeft er
volgens mij al jaren geen betrokkenheid meer mee.
9
Tot slot
De laatste jaren hebben Bertrand en ik elkaar niet meer gezien. Ik ben veel meer werkzaam in de
Nieuwe Energie en groene grondstoffen industrie. Niet zozeer zijn gebied. Tevens was hij
begonnen met de afbouw van zijn loopbaan als hoogleraar en ging hij meer en meer genieten
van zijn privé leven: hardlopen en zijn dierbaren om hem heen.
Ik hoop dat hij echt blijft genieten van …. en dat hij met een goed gevoel de Universiteit achter
zich kan laten..
Ik kijk terug op een interessante, leerrijke en plezierige samenwerking met hem!!
10
11
WILL BERTRAND ALS DIRECTEUR VAN LMS
Twan Geenen
In dit hoofdstuk wil ik graag de rol belichten die Will heeft gespeeld als directeur van de
Ontwerpersopleiding Logistics Management Systems van 20 juni 1991 tot 2 juni 1999 en van 15
december 2000 tot 1 maart 2002. Ik zal dat doen door een stukje van de geschiedenis van deze
opleiding te beschrijven en hoe ik mijn periode als programmamanager (van oktober 1995 tot
november 2007), met Will als directeur, heb ervaren. Dit hoofdstuk pretendeert niet de volledige
geschiedenis van de opleiding te beschrijven, het beschrijft de naar mijn mening belangrijkste
punten. Daarvoor heb ik uiteraard gebruik gemaakt van het archief van de opleiding en daarbij
kwamen ook zaken naar voren die ik nog niet wist.
1. Een stukje geschiedenis over hoe we elkaar leerden kennen
Het is misschien, als achtergrondinformatie, aardig om te weten dat ik Will al ken sinds eind
jaren zestig. Will was net afgestudeerd en als wetenschappelijk medewerker verbonden aan de
faculteit Technische Bedrijfskunde i.o. De faculteit was in oprichting en had nog geen eigen
propedeuse. Nadat ik in 1968 mijn propedeuse werktuigbouwkunde had behaald ging ik verder
met bedrijfskunde en kwam daar Will, toen nog getooid met een woeste baard passend bij de
tijdgeest, regelmatig tegen als docent. In 1972 werd hij begeleider bij mijn afstudeerproject.
Nadat ik 2 jaar als consultant en 15 jaar in managementfuncties in de gezondheidszorg had
gewerkt kwam ik medio 1989 weer terug bij de faculteit bij de vakgroep Organisatiekunde en
vanaf dat moment zagen we elkaar weer regelmatig. Na enige tijd ging ik over naar de vakgroep
Internationale en Distributielogistiek die later weer fuseerde met de capaciteitsgroep Operations,
Planning, and Control waarvan Will de voorzitter was.
12
2. Korte geschiedenis van de totstandkoming van de Ontwerpers-
opleiding ‘International Program in Logistics Management
Systems’ (LMS in het kort) De aanleiding tot het opzetten van ontwerpersopleidingen was gelegen in het in 1982
terugbrengen van 5 naar 4 jaar van bijna alle universitaire opleidingen waaronder ook de
ingenieursopleidingen. Het bedrijfsleven was bang dat de ingenieursopleidingen theoretischer
van aard zouden worden en dat het ontwerpkarakter verloren zou gaan. Onder druk van het
bedrijfsleven werden ontwerpersopleidingen opgezet en deze kwamen vooral van de grond in
Eindhoven en veel minder in Delft en Enschede. Eind 1985 nam het plan van de faculteiten
Technische Bedrijfskunde, Werktuigbouwkunde en Wiskunde & Informatica concrete vormen
aan om in september 1986 gezamenlijk te starten met de 2-jarige nadoctorale
ontwerpersopleiding Beheersing van Bedrijfsprocessen. Daar waren enkele rapporten op hoger
niveau aan vooraf gegaan.
Een voorbereidingswerkgroep bestaande uit een aantal hoogleraren uit de deelnemende
faculteiten en aangevuld met een aantal mensen uit het bedrijfsleven (Philips, DAF en Volmac)
legt de contouren en de eindtermen vast en besluit tot de volgende naam: Ontwerpen van
Logistieke Besturingssystemen. In die contouren wordt ook vastgelegd dat het eerste halfjaar
van de opleiding besteed wordt aan ‘gewone’ vakken uit de predoctorale fase(s) van de
opleiding(en) gevolgd door een jaar lang speciale nadoctorale werkcolleges, vaardigheids-
trainingen, groepsprojecten en als afsluiting een onderzoeksproject (het ware natuurlijk beter
geweest om dit een ontwerpproject te noemen) van een half jaar. Medio 1986 wordt de
voorbereidingswerkgroep opgeheven en wordt een opleidingswerkgroep ingesteld waarin de
bedrijfsmensen geen zitting meer hebben. Deze laatsten komen in een Adviesraad. Uit de
opleidingswerkgroep zien we de Docentenraad ontstaan waarin de docenten zitting hebben die
de verschillende ‘blokken’ van de opleiding gaan coördineren. We komen dan de naam van Will
Bertrand tegen als blokdocent van blok G waarover later meer.
De opleiding krijgt nu de volgende structuur: een half jaar ‘homologatiefase’, daarna een jaar
‘cursorische fase’ en ter afronding een ontwerpopdracht van een half jaar.
De naam voor het eerste half jaar is eigenlijk niet erg toepasselijk. In die periode worden vakken
gevolgd uit de gewone predoctorale fase(s) van de opleiding(en). Het betreft eigenlijk een
‘homogeniseringsfase’. Het doel is om bij de brede en diverse instroom (vanuit verschillende
vooropleidingen zoals technische bedrijfskunde, werktuigbouwkunde, wiskunde, informatica,
natuurkunde, econometrie, landbouwkunde, scheikunde etc. ) een gemeenschappelijk fundament
en begrippenkader aan te brengen waarop in het daarop volgende jaar (de cursorische fase
genoemd) kan worden voortgebouwd. Dit eerste halfjaar heeft altijd tot veel onbegrip geleid bij
13
het overkoepelende Stan Ackermans Instituut omdat onze opleiding eigenlijk de enige was met
zo’n diverse instroom. De andere ontwerpersopleidingen hadden daardoor veel minder of geen
behoefte aan een ‘homogeniseringsfase’.
In tegenstelling tot het eerste halfjaar worden in de ‘cursorische fase’ vakken gedoceerd die
speciaal zijn ontworpen voor de opleiding.
2.1 De werkelijke start De Ontwerpersopleiding draagt bij de start in september 1986, wanneer de eerste cursisten
starten met hun opleiding, de naam Ontwerpersopleiding Logistieke Besturingssystemen en is
dan één van de circa 10 ontwerpersopleidingen van de Technische Universiteit Eindhoven.
Informeel werd deze ontwerpersopleiding nog tot ver in de jaren ’90 en zelfs nog daarna,
aangeduid met de benaming NADO. Dit laatste stond voor NADOctorale opleiding maar de
naam was buiten de eigen faculteit natuurlijk niet onderscheidend. Alle ontwerpersopleidingen
waren immers NADO’s. Ze waren organisatorisch ondergebracht bij het IVO (Instituut
Vervolgopleidingen), het latere Stan Ackermans Instituut. Zij werden echter gehuisvest binnen
de ‘penvoerende’ faculteit.
In september/oktober 1986 beginnen de eerste twee cursisten aan hun opleiding. Uit het
sollicitantenoverzicht van oktober 1986 blijkt dat er redelijk veel HBO-opgeleiden solliciteren
naar een opleidingsplaats maar dat uiteindelijk slechts een enkeling wordt toegelaten. In latere
jaren worden ze niet meer toegelaten omdat de kans van slagen minimaal blijkt te zijn.
In februari 1987 wordt Andries Veldkamp de eerste cursuscoördinator. In de archiefstukken lees
ik dat hij al in juni van datzelfde jaar aan de docentenraad zijn ‘kijk op de ‘NADO-
LOGISTIEK’ presenteert. Aan de hand van een Ishikawa (visgraat)-diagram, gebaseerd op de
industrie voor professionele producten, maakt hij de relatie duidelijk tussen een aantal factoren
zoals het product, het initiële proces, het productieproces, organisatie cultuur & ondersteuning
en tenslotte externe invloeden en de als gevolg daarvan geleverde logistieke prestatie. Een
aantal leden van de raad herkent en deelt de gepresenteerde relaties en de conseqenties die dit
kan hebben voor de wijze waarop een logistiek probleem dan ‘moet worden aangepakt’.
De gedachtengang is dat de meeste ‘logistieke ontwerpopdrachten’ in feite ‘herontwerp-
opdrachten’ zijn met als doel een gegeven situatie (prestatie) te verbeteren.
Veldkamp doet het voorstel om dergelijke Ishikawa-diagrammen niet alleen te ontwikkelen
voor de industrie voor professionele producten maar ook voor consumentengoederen, de
procesindustrie, de confectie, vervoersorganisaties en het grootwinkelbedrijf en daarop dan een
soort weging toe te passen om gevoel te krijgen voor de mate waarin de factoren invloed kunnen
hebben op de uiteindelijke logistieke prestatie. In feite probeert hij daarmede ook aan te geven
dat er naast de productie en het vervoer nog andere sectoren in de industrie zijn waar logistieke
processen een rol spelen. Veldkamp neemt de ‘dienstverlening’ mee maar is nog niet zover om
14
ook administratieve dienstverlening (zoals bijvoorbeeld uitkeringsorganisaties, salaris admini-
straties en verzekeraars) te zien als organisaties waarin logistieke processen een belangrijke rol
spelen (later worden daarvoor de termen ‘kantoorlogistiek’ en ‘Workflow Management’
gehanteerd). Met de ontwikkeling van de bestuurlijke informatieverwerking neemt het belang
daarvan in latere jaren zeer sterk toe.
Met het benoemen van de ‘prestatiebeïnvloedende’ factoren in zijn Ishikawa diagram probeert
hij ook de begrenzing aan te geven van hetgeen tot het aandachtsgebied van de logistiek moet
worden gerekend.
Tijdens de vergadering van de docentenraad op 3 mei 1988 meldt Veldkamp dat hij het
Ishikawa-diagram wat heeft aangepast naar aanleiding van een aantal gesprekken die hij
gevoerd heeft met ‘verschillende personen uit het bedrijfsleven’. De door hem gewenste
‘weging’ kan niet worden gemaakt maar een aantal geïnterviewden gaf wel aan ‘dat ze het
Ishikawa-diagram in zijn kwalitatieve opzet een goed in kaart gebrachte inventarisatie vonden
van kennis die eigenlijk allemaal vereist is voor het goed doen functioneren van de logistieke
functie’. De docentenraad acht het diagram dan ook prima bruikbaar voor de vormgeving van de
opleiding NADO-logistiek. We praten dan dus niet over het diagram als schematechniek maar
over het door Veldkamp ‘gevulde’ diagram.
Tijdens diezelfde vergadering presenteert Veldkamp een notitie met zijn gedachten over de
invoering van Quick-Scans in de opleiding. Hij ziet dat als een oefening in het snel
diagnostiseren van een bedrijfsprobleem. Het beantwoorden dus van de vraag: “Wat is de
oorzaak van een gesignaleerd probleem (waardoor is de prestatie van het systeem niet conform
de verwachting)? ” Hij trekt de parallel met de opleiding tot arts. Eerst een wetenschappelijke
opleiding tot arts maar voordat je je mag vestigen als ‘behandelend’ arts moet je ook leren een
diagnose te stellen en een effectieve therapie te kiezen. Het uitvoeren van Quick-Scans bij
bedrijven door cursisten zou een goed middel kunnen zijn om die laatste vaardigheden te
trainen. De logistiek manager van de toekomst is volgens hem meer gebaat bij het vermogen tot
snelle diagnose dan ‘het onderbouwd hebben van een felle ontwerpdrift’.
Verder stelt hij dat, indien het lukt dit instrumentarium uit te bouwen, het in de toekomst
mogelijk moet zijn om de Quick-Scan uit te bouwen tot een product dat op de markt gevraagd
wordt en dan kan helpen een dekking te leveren voor de kosten van de opleiding. Hij beschikt
over een vooruitziende blik want het concept Quick-Scan blijkt in de jaren daarna zeer
succesvol zowel voor onze cursisten als voor bedrijven. Latere verfijningen in het concept
dragen daaraan zeker bij. De docentenraad besluit om iedere cursist minstens twee Quick-Scans
te laten uitvoeren.
Jammer genoeg legt Veldkamp hier zelf niet de link tussen de Quick-Scan en het door hem
opgestelde Ishikawa-diagram maar het is voor mij in ieder geval duidelijk dat dat diagram van
grote waarde had kunnen zijn bij het diagnostiseren van problemen. Ik schrijf hier bewust ‘had’
15
omdat ik later, als programmamanager, het Ishikawa-diagram niet meer terugzag als instrument
bij de Quick-Scan. Het zogenaamde ‘cause-effect-schema’, dat in de kern ook beoogt om een
probleem gestructureerd in kaart te brengen, wordt daarin wel gehanteerd maar dat is enkel een
(lege) schematechniek terwijl in het door Veldkamp voorgestelde diagram heel veel
gecondenseerde kennis ligt opgeslagen.
Indien ik kennis had gehad van deze voorgeschiedenis toen ik in oktober 1995 programma-
manager werd, dan zou ik binnen de opleiding waarschijnlijk meer gebruik hebben gemaakt van
het ‘gevulde’Ishikawa-diagram in relatie tot het diagnostiseren van logistieke problemen.
2.2 De rol van Will als blokdocent binnen de ontwerpersopleiding Uit de opleidingswerkgroep zien we de Docentenraad ontstaan waarin de docenten zitting
hebben die de verschillende ‘blokken’ van de opleiding coördineren. We komen dan de naam
van Will Bertrand tegen als blokdocent van blok G. In september 1986 komt blok G , ook wel
het ‘integratieblok’ genoemd, voor het eerst mondeling en inhoudelijk aan de orde in de
Opleidingswerkgroep, schriftelijke uitwerking moet dan nog volgen. Het gaat dan om cases,
stages, groeps- en de ontwerpopdracht. In zijn brief van 1 oktober 1986 zet Will uiteen hoe hij
het betreffende blok G wil opzetten. Belangrijkste elementen zijn: gemeenschappelijke
begrippen afspreken (o.a. gebaseerd op de APICS-dictionary), verwijzing naar dezelfde
praktijkproblemen en het ontwikkelen van een gemeenschappelijke visie op de structuur van
logistieke beheersingssystemen bij de verschillende docenten. Veel later hoor ik dezelfde
aspecten weer terugkomen in discussies binnen de capaciteitsgroep als het over het steeds
terugkerende onderwerp ‘ontwerpen’ gaat.
Het gaat te ver om hier de gehele structuur van de ontwerpersopleiding te beschrijven maar ik
maak een uitzondering voor het gedeelte dat altijd de warme belangstelling van Will had en
waar volgens mij ook zijn hart lag. Dat was het praktijkgedeelte waarin het echte ontwerpen een
belangrijke rol speelt en waarin de cursist moet laten zien dat hij het vak beheerst (het hiervoor
genoemde blok G). Een blok dat zeer gewaardeerd wordt door de cursisten omdat ze het gevoel
hebben dat het daar uiteindelijk allemaal om draait. Binnen dit blok moeten ze logistieke
problemen binnen bedrijven oplossen en dus kunnen ze daar het ontwerpen oefenen en in de
praktijk brengen. Het blok bestaat dan uit een aantal componenten die cursisten in groeps-
verband uitvoeren en een individuele component. In groepsverband wordt tweemaal een Quick-
Scan uitgevoerd bij een bedrijf met een logistieke vraagstelling. De daarbij gebruikte Methode
Quick-Scan geniet ruime bekendheid en met twee groepen wordt parallel en in competitie aan
een project gewerkt. Daarbij gaat het er altijd fanatiek aan toe en wordt ‘bedrijfsspionnage’ niet
geschuwd. Het project wordt in deeltijd uitgevoerd. Beide groepen presenteren hun ‘oplossing’
aan de opdrachtgever en er wordt een groep aangewezen die het ‘het beste’ heeft gedaan. Het
‘beste’ betekent in dit verband de ‘beste’ oplossing passend bij dit specifieke bedrijf en gemeten
16
naar de stand van de wetenschap. Ook de presentatie, rapportage en overtuigingskracht van de
groep speelt een belangrijke rol.
In 1993 worden de eisen die gesteld moeten worden aan het uitvoeren van de ‘ontwerpopdracht’
nog eens onder de loep genomen. De indruk bestaat dat cursisten te weinig van de methoden en
technieken, die ze geleerd hebben tijdens hun opleiding, gebruiken bij het (her)ontwerpen. De
‘wetenschappelijke’ benadering bij het ontwerpen komt te weinig aan bod. Op 29 november
1994 presenteert Will als blokdocent de checklist “Methoden en technieken NADO-Logistiek”,
welke checklist vanaf dat moment een belangrijke rol speelt bij alle praktijkcomponenten.
In de jaren 1996 en daarna komt als ‘groepscomponent’ het Global Supply Chain Project erbij
onder leiding van Jan Fransoo. Ook dit project wordt in deeltijd uitgevoerd in samenwerking
met studenten en prof. Hau Lee van de Stanford University in de Verenigde Staten en heeft een
doorlooptijd van ongeveer 3 maanden. Daarbij komt een groep studenten van Stanford een week
naar Eindhoven om te werken aan een probleemstelling bij een bedrijf. Daarna gaan zij terug en
wordt binnen de eigen universiteiten en eventueel bij bedrijven in de Verenigde Staten en
Europa gedurende ruim 2 maanden gewerkt aan de vraagstelling. Ter afronding gaan de
cursisten uit Eindhoven een week naar Stanford University om samen met de studenten daar het
project te presenteren en af te ronden. Meestal kunnen ze er dan nog een weekje vakantie aan
vastknopen. Al met al een zeer populair onderdeel van de opleiding waarbij de cursisten leren
om met Amerikaanse studenten samen te werken die een werkbenadering hebben die je zou
kunnen karakteriseren als deadline driven in tegenstelling tot de planning driven benadering die
in Eindhoven meer gebruikelijk is. Daarnaast leren ze om met moderne communicatiemiddelen
om te gaan waaronder uiteraard e-mail maar ook met video-conferencing dat op dat moment
nog niet zo gebruikelijk is. Ook de tijdsverschillen leiden tot een aparte dynamiek en heeft tot
gevolg dat er binnen de opleiding op tijden gewerkt wordt die niet ‘gebruikelijk’ zijn binnen de
faculteit.
De ‘individuele’ component bestaat dan uit een Logistics Design Project (eindproject) van
ongeveer 6 maanden waarbij de cursist, binnen een bedrijf, door het diagnostiseren en oplossen
van een logistiek probleem (door het maken van een geschikt herontwerp) moet aantonen dat hij
het logistiek ontwerpen beheerst. Het bedrijf moet hiervoor een substantieel bedrag betalen aan
de opleiding. De cursist wordt gedurende die periode fulltime gedetacheerd bij het bedrijf. Het
voordeel van het laten betalen door het bedrijf is dat daardoor echte problemen worden
gedefinieerd, dat het bedrijf er een zekere urgentie aan toekent en dat er een resultaats-
verplichting ontstaat voor de opleiding (zoals bij consultancy). Dit stelt daardoor ook zware
eisen aan de begeleiders van deze projecten. De begeleiders komen dan ook uit een
geselecteerde groep docenten met ontwerpervaring.
Zoals ik al schreef ligt Will´s hart bij deze ontwerpcomponenten en hij houdt als blokdocent de
touwtjes hierbij stevig in handen om de kwaliteit te kunnen bewaken. Dit doet hij o.a. door alle
17
opdrachtomschrijvingen zelf te beoordelen op voldoende ontwerpgerichtheid en -inhoud.
Daarnaast door de docenten te selecteren die gekwalificeerd worden geacht om eindprojecten te
kunnen begeleiden. In de loop der jaren wordt dat wel een steeds groter probleem. Er zijn steeds
minder docenten (medewerkers) die ontwerpgericht zijn en/of ontwerpervaring hebben. Selectie
van medewerkers door de universiteit geschiedt gaandeweg steeds meer op onderzoeks-
kwaliteiten.
3. De rol van Will als directeur van de ontwerpersopleiding Van 20 juni 1991 tot 2 juni 1999 en van 15 december 2000 tot 1 maart 2002 was Will directeur
van de opleiding. Als programmamanager van oktober 1995 tot november 2007 werkte ik
gedurende die periode nauw met hem samen.
In de eerste jaren van de opleiding neemt de instroom geleidelijk toe. Een top wordt bereikt in
1993 wanneer een kleine 40 nieuwe cursisten instromen. Dit betekent dat er ultimo 1993
ongeveer 70 cursisten in totaal zijn waarvan er ongeveer 15 in bedrijven bezig zijn met hun
eindopdracht. De cursisten spreken hun zorg uit over de gevolgen voor de kwaliteit van hun
opleiding bij dit grote aantal. Er wordt beweerd dat het grote aantal zou samenhangen met de
geplande opschorting van de dienstplicht per 1 mei 1997. Vanaf augustus 1996 zouden al geen
dienstplichtigen meer worden opgeroepen. Door na hun universitaire opleiding nog een
ontwerpersopleiding van 2 jaar te volgen zouden cursisten kunnen voorkomen dat ze de
dienstplicht moeten vervullen. Vanaf 1994 neemt de instroom weer geleidelijk af. We gaan dan
richting hoogconjunctuur met een afnemende instroom van Nederlandse cursisten en wat later
een toename van cursisten uit Oost-Europa en het Verre-Oosten.
In oktober 1995 word ik programmamanager van de opleiding. Will laat er als directeur geen
gras over groeien en op 2 januari 1996 schrijft hij mij een brief waarin hij stelt dat het ‘tijd is om
eens na te denken over een aangepaste opzet van de NADO-Logistiek opleiding’. Hij wil
daarvoor een werkgroep opstarten die zich bezighoudt met de volgende 2 aspecten.
De studieduren van technische opleidingen zijn weer op het niveau van 5 jaar gebracht en het is
de vraag hoe de aansluiting van de doctorale op de postdoctorale opleiding het beste vorm kan
worden gegeven (bijv. een gedeeltelijke ‘indaling’ van de ontwerpersopleiding in de initiële
(ingenieurs)opleiding. Door deze ‘indaling’ zou mogelijk tijdwinst te behalen zijn. Misschien
zijn daarover afspraken te maken met de belangrijkste toeleverende faculteiten. Als gevolg van
het feit dat in de jaren daarna de Nederlandse instroom voor een groot deel ‘opdroogt’ en de
instroom dus vooral uit het buitenland komt, heeft het weinig zin meer om die ideeën nader uit
te werken.
18
Een tweede reden voor een mogelijke herprogrammering zijn de resultaten van een evaluatie die
gehouden is onder afgestudeerden en aanwijzingen gegeven door de Raad van Advies.
De herprogrammeringscommissie komt begin 1997, samen met de docentenraad, tot de
conclusie dat geen grote veranderingen nodig zijn en dat alleen wat kleine wijzigingen en
aanpassingen zullen worden doorgevoerd. Zo zal onderzocht worden of er meer ontwerp-
mogelijkheden in de opleiding kunnen komen en wat meer technologie. In de vergadering van
de docentenraad van 12 juni 1997 brengt de commissie onder leiding van Will haar eindrapport
uit. Door de commissie is het gehele programma geherwaardeerd door middel van een zo goed
mogelijke schatting van belastingsuren. Daarbij wordt ook de ‘homogeniseringsfase’ terug-
gebracht ten gunste van de ‘cursorische fase’. Vervolgens worden een aantal kleine
aanpassingen voorgesteld voor een aantal vakken maar de grote aanpassingen vinden plaats bij
het ‘ontwerponderwijs’. Na een aanpassing naar aanleiding van opmerkingen van de Adviesraad
wordt het reeds eerder genoemde Global Supply Chain Project officieel ingevoerd met een
studiebelasting van 150 uren. De belasting voor de twee Quick-Scans wordt verhoogd van 200
uren tot 300 uren totaal en het Logistics Design Project wordt van 800 uren uitgebreid naar 900
uren. Met genoemde verschuivingen en uitbreidingen kan worden tegemoetgekomen aan reeds
eerder geuite wensen van het Stan Ackermans Instituut en van de Certificatiecommissie van het
KIVI. Naast deze uitbreidingen wordt ook een vak ingevoerd voor het ontwikkelen van sociale
en communicatieve vaardigheden (professional development). Met name door de toename van
buitenlandse cursisten ontstaat hieraan steeds meer behoefte. Deze cursisten blijken op dit
gebied toch minder vaardig dan de Nederlandse cursisten en ze moeten toch worden voorbereid
op een arbeidsplaats in een Nederlands of ander westers bedrijf waar op dit punt hoge eisen aan
hen zullen worden gesteld.
Medio 1998 wordt door de docentenraad, mede op verzoek van een voor de opleiding relevant
deel van het bedrijfsleven, besloten om de opleiding vanaf 1 december daaropvolgend geheel in
de Engelse taal te gaan aanbieden. Een bijkomend voordeel is dat dan ook buitenlandse
cursisten gemakkelijker kunnen instromen. Tot dat moment moeten zij eerst de |Nederlandse
taal machtig worden om in staat te zijn de Nederlandstalige colleges te volgen en te kunnen
participeren in de projecten. Uiteraard heeft dit besluit grote gevolgen. Zo moet al het materiaal
vertaald worden en kan voor het ‘homologatieprogramma’ geen gebruik meer worden gemaakt
van vakken uit de predoctorale fase(s) van de opleiding(en). Extra kosten zijn dus het gevolg.
Halverwege 1998 is de instroom van Nederlandse cursisten zodanig teruggelopen dat wordt
besloten om een communicatiebureau in te schakelen om een marktonderzoek te doen en de
wervingsactiviteiten te verbeteren. Tot dat moment geschiedt de werving centraal door het
overkoepelende Stan Ackermans Instituut en we vragen ons af of we de werving niet beter zelf
ter hand kunnen nemen. Bij de presentatie aan de docentenraad van de derde fase van dit project
op 2 juni 1999, komt ook de naam van de opleiding ter sprake. Men besluit om de naam
19
‘Ontwerpersopleiding Logistieke Besturingssystemen’ te wijzigen in de volgens sommigen meer
aansprekende naam ‘International Program in Logistics Management Systems’. Dit verloopt
niet zonder discussie want de voorgestelde benaming ‘International Logistics Management’
vindt geen genade in de ogen van voorzitter Will. De opleiding gaat immers niet over
international logistics en het is ook geen opleiding op het gebied van logistics management. De
nieuwe naam moet dus met ‘aandacht’ worden gelezen.
Aan het einde van de vergadering van 2 juni 1999 draagt Will de voorzittershamer die hij 8 jaar
heeft gehanteerd, over aan Peter van Laarhoven. Ongeveer 1,5 jaar later krijgt hij hem weer
terug voor een interimperiode van ruim 1 jaar.
Vlak voordat Will de voorzittershamer weer opneemt wordt nog een belangrijke structurele
wijziging in de organisatie van het programma doorgevoerd. Tot medio 2000 heeft de opleiding
een jaarindeling met 3 trimesters en een zomerperiode waarin ook cursussen worden gepland.
Door de daarin vallende zomervakantie is die studieperiode veel korter. Bij 3 instroomtijd-
stippen per jaar leidt dat tot veel planningtechnische problemen. Besloten wordt om het jaar in
te delen in 4 periodes die inclusief vakantiedagen, een deel van de ADV-dagen en de feest-
dagen, evenveel werkdagen bevatten zodat een gelijkmatige werkverdeling ontstaat. De
overgang van het ene naar het andere systeem brengt uiteraard de nodige problemen met zich
mee, maar na invoering ontstaat veel meer rust in het programma.
In december 2000 neemt Will de voorzittershamer weer op. Door de lagere instroom zijn ook
begrotingsproblemen ontstaan en tegelijkertijd speelt de discussie of en hoe de opleiding moet
reageren op de invoering van de Bachelor/Master-structuur in de eerste fase-opleidingen. In
eerste instantie wordt door het management van de opleiding gedacht aan het voortzetten van de
ontwerpersopleiding als een masteropleiding in de logistiek voor excellente studenten. De
overheid heeft in haar notitie van 13 november 2000: “NAAR EEN OPEN HOGER
ONDERWIJS” die mogelijkheid geopend. Groot voordeel zou zijn dat de financiering via de
eerste geldstroom gaat en studenten onder het stelsel van studiefinanciering vallen waardoor
geen salaris meer betaald hoeft te worden. Selectie van studenten blijft mogelijk. Het College
van Bestuur blijkt nog niet toe te zijn aan masteropleidingen voor excellente studenten. Men is
bang voor een devaluatie van de ‘gewone’ masteropleidingen.
Als uitvloeisel van de discussie met het faculteitsbestuur hierover blijkt dat de gedachte leeft dat
er misschien behoefte is aan een postdoctorale (parttime) opleiding op het gebied van Logistics
Management Systems (MBA-achtig) gericht op academici en HBO-opgeleiden met een
technische of bedrijfseconomische achtergrond en met een aantal jaren relevante bedrijfs-
ervaring. Het opleidingsmanagement werkt gedachten hierover verder uit. Deze opleiding zou
dan buiten de eerste fase-opleiding en buiten de ontwerpersopleiding gestalte moeten krijgen.
Doordat het faculteitsbestuur geen duidelijk standpunt inneemt over deze voorstellen komt ook
dat niet van de grond.
20
Eind 2001 wordt het echter spannend voor de ontwerpersopleidingen en voor LMS in het
bijzonder. De TU/e is financieel gezien in zwaar weer terechtgekomen. De LMS-begroting voor
2002 wordt door het overkoepelende SAI niet goedgekeurd met als voornaamste reden de
dalende instroom en het daardoor optredende financiële tekort. Ook wordt bekend dat het
College van Bestuur op een eerdere toewijzing van gelden aan het SAI nog 1 miljoen gulden
gekort heeft (2001 is het laatste jaar van deze vertrouwde munteenheid). Ook heeft het College
aan een commissie opdracht gegeven om een advies aan haar uit te brengen over de gewenste
toekomst van het SAI en de ontwerpersopleidingen. In een gesprek op 14 januari 2002 tussen
directie SAI, directie LMS, afvaardiging faculteitsbestuur en afvaardiging College van Bestuur
proberen we duidelijk te maken dat LMS nog steeds een goede toekomst heeft en ook in een
behoefte voorziet. LMS moet zich dan wel duidelijker gaan richten op een buitenlandse
instroom (Oost-Europa en Azië). Dit betekent wel een andere werving en selectie, een
herstructurering van het programma en intensievere begeleiding gezien de optredende
cultuurverschillen. Op 15 januari vindt een overleg plaats tussen de Rector Magnificus, directie
SAI en leden van de commissie herstructurering. Uit informatie blijkt dat men conform het
advies van de commissie het aantal ontwerpersopleidingen van het SAI van 10 naar 5 wil
terugbrengen. De opleiding LMS is daar niet bij. Het dreigt fout af te lopen voor LMS. Echter
door goede interne contacten en een informeel overleg tussen faculteitsbestuur en Rector
Magnificus wordt besloten dat LMS doorgaat met daarbij een herprogrammering en dat de
opleiding zich gaat richten op een grotere instroom vanuit het buitenland. Op 11 februari, twee
weken voordat Will de voorzittershamer voor de tweede keer overdraagt, horen we dat de
doorstart ook is gehonoreerd door de directie SAI, de commissie herstructurering en het
faculteitsbestuur. Dit laatste is nodig omdat onderdeel van de plannen ook blijkt te zijn dat de
ontwerpersopleidingen ondergebracht worden bij de penvoerende faculteiten waarbij de Rector
Magnificus beleidsmatig verantwoordelijk blijft. Dit houdt echter in dat de financiële en
personele verantwoordelijkheid voor de opleidingen geheel bij de desbetreffende penvoerende
faculteit komt te liggen. Het SAI wordt als aparte beheerseenheid binnen de TU/e opgeheven en
wordt omgevormd tot een klein staf- en dienstenbureau voor het College van Bestuur.
Het onderbrengen van de opleidingen bij de desbetreffende faculteiten zal, naar later blijkt,
grote consequenties hebben. De centrale regie over de opleidingen komt te vervallen en de
faculteiten kunnen nu zelfstandig beslissingen nemen over het eventuele voortbestaan van een
opleiding. Verder kunnen zij eisen stellen aan de wijze waarop aan de inhoud wordt
vormgegeven.
Ik denk dat het een gelukkig moment voor Will was om de voorzittershamer op 1 maart 2002
over te dragen aan Nico Dellaert. De richting die daarna met de opleiding wordt ingezet onder
invloed van eisen gesteld door het faculteitsbestuur zullen hem niet met vreugde vervullen. Om
tegemoet te komen aan de financiële eisen van de faculteit moet steeds meer gebruik gemaakt
21
worden van vakken uit de reguliere masteropleiding en het specifieke karakter gaat dus een
beetje verloren.
Ik wil daar niet verder op ingaan want dit verhaal gaat over de periode Will Bertrand en die
eindigt hier. Het vervolg van het verhaal mag iemand anders schrijven in een volgend Liber
Amicorum.
Nogmaals benadruk ik dat ik niet heb geprobeerd de volledige geschiedenis van LMS in deze
periode te beschrijven. Ik heb de onderwerpen gekozen waarvan ik dacht dat ze belangrijk
waren en waarbij Will een leidende en bepalende rol heeft gespeeld.
4. Enkele persoonlijke herinneringen Zoals in de eerste paragraaf beschreven ‘ken’ ik Will inmiddels zo’n 40 jaren, waarvan de
laatste 21 jaren een stukje beter als collega. Zowel als lid van zijn capaciteitsgroep als in mijn
rol als programmamanager van de ontwerpersopleiding heb ik met veel plezier met hem
samengewerkt. We hadden vaak dezelfde mening over bepaalde onderwerpen maar soms ook
niet. Dat leverde altijd aardige discussies op tijdens de bekende koffie- en theepauzes. Het was
mijn gewoonte om hem en andere leden van de groep ’s morgens en ’s middags uit hun kamer te
halen voor de gezamenlijke koffie of thee. Enerzijds omdat ik het belangrijk vond dat mensen
zich niet de gehele dag in hun kamer achter de computer opsloten maar er was ook een meer
persoonlijke reden. Ik zat in een andere gang tussen buitenlandse cursisten en moest daar de
gehele dag Engels (of Dunglish zo u wilt) spreken en soms onnavolgbaar Engels aanhoren en
dan had ik af en toe behoefte om weer even gewoon Nederlands te kunnen spreken met
collega’s en over onderwerpen te kunnen praten die me interesseerden. In die gesprekken met
Will en de andere collega’s bespraken we ‘de toestand in de wereld’ en trachtten we ook
oplossingen te bedenken voor de wereldproblemen. De laatste jaren kregen wij als duo de
bijnaam ‘grumpy old men’ omdat we nogal eens discussieerden over de achteruitgang van het
onderwijs of misschien beter gezegd het dalende niveau van de instromende studenten aan de
universiteit en uiteraard ook omdat we tot de oudere generatie behoorden. Het doet me deugd
om te zien dat allerlei belangengroeperingen en ook de overheid tot ditzelfde inzicht is gekomen
en probeert daaraan iets te veranderen. Een zaak van lange adem.
Onze samenwerking binnen de ontwerpersopleiding was ook erg plezierig. Samen met Stance
van Woensel als secretaresse kon ik in grote zelfstandigheid de opleiding managen. We
vormden de ‘vader en moeder’ van de opleiding. Opbloeiende liefdes, liefdesverdriet,
huwelijken en ruzies, dat kwam allemaal voor. Ook werden we geconfronteerd met heimwee,
huisvestingsproblemen enz. Stance was fantastisch in het oplossen van deze zaken en ze hoorde
22
ook altijd als er iets aan de hand was. Wat betreft de opleiding zelf probeerden we zaken zoveel
mogelijk zelf te regelen en we betrokken Will er pas bij als het echt niet anders kon. Hij kon dat
zeker waarderen.
Samen met Will en meestal ook nog een derde persoon heb ik vele sollicitatiegesprekken
gevoerd met mogelijke cursisten. Daarbij keken we naar vooropleiding, intellectuele
capaciteiten maar we probeerden ook in te schatten hoe die persoon het later mogelijk zou doen
in het bedrijfsleven. Will en ik waren het daarbij opvallend vaak met elkaar eens. Ik denk wel
eens dat we dat helemaal niet slecht deden hoewel we ook vergissingen hebben gemaakt. Veel
van onze afgestudeerden kwamen snel aan een baan en ook vrij snel op hoge posities in het
bedrijfsleven terecht.
Terugkijkend kan ik zeggen dat ik met heel veel plezier met Will heb samengewerkt, zowel in
de capaciteitsgroep als binnen de ontwerpersopleiding.
Één uitspraak van Will zal ik nooit vergeten toen ik me weer eens druk maakte over iets dat niet
goed ging maar waaraan ik toch niets kon veranderen. Hij zei toen: “Aan een dood paard moet
je niet trekken”. En zo is het maar net.
Hartelijk bedankt Will voor de jarenlange fijne samenwerking en vooral voor je troostende
schouder toen ik 4 jaar geleden een persoonlijk drama meemaakte. Je toonde je persoonlijke
betrokkenheid en dat heeft me ontzettend veel goed gedaan.
Ik hoop dat de komende periode voor jou heel fijn zal zijn en dat je er in goede gezondheid van
mag genieten. ’ Er zijn nog genoeg paden om te bewandelen’.
Het ga je goed.
Twan Geenen.
23
DE PROCESONTWIKKELING VAN DE KLM
IN DE AFGELOPEN 12 JAAR
P.N. Bos
Inleiding
Twaalf jaar geleden kreeg ik van onze huidige president directeur, drs. ing. Hartman, de unieke
kans het hoofdproces van de KLM met alle operationele en commerciële afdelingen opnieuw te
structureren (sectie 1) en startte de samenwerking met prof. Bertrand. We konden de
gezamenlijke taal en werkmethodiek opnieuw opzetten en ontwikkelen, ten einde zo de
voortdurende ontwikkeling van de airline te faciliteren (sectie 2). Nu, ruim tien jaar later, is dit
nog steeds de methodiek, maar ook die vraagt voortdurend om onderhoud. Met mijn vertrek
naar de divisie grondafhandeling kregen we de gelegenheid de visie daar te herzien in een die
aansloot bij de net op ondernemingsniveau ontwikkelde visie (sectie 3) en in de jaren die
volgden hebben we de noeste arbeid verricht om de werkelijkheid en het model van de aircraft
flow naar elkaar toe te brengen en te vervolmaken (sectie 4). In dat proces hebben we een aantal
enthousiaste medewerkers en studenten aan onze zijde gekregen. Voor alle betrokken
uitvoerende afdelingen viel het niet altijd mee want de IT ontwikkeling ging met horten en
stoten. Maar eind goed al goed: het komende half jaar volgt de afronding.
Met de verdere ontwikkelingen is er nog een perspectief voor jaren: de completering van de
aircraft flow, het concept van grondtijdmanagement, waar supporters voor zijn, en in de
”verbreding” het concept van “real time hub”, dat in het MT van de Hub is ontwikkeld (sectie
5). Ten slotte komen we in sectie 6 terug op de diverse rollen die prof. Bertrand in de afgelopen
jaren bij de KLM heeft gespeeld.
1. Het Programma Operationele Integriteit
1.1. Het programma In het programma Hub in Transfer werd het belangrijke besluit genomen om een apart Hub
controle Centrum ( HCC) te starten. Dit besluit was een uitgangspunt voor de oprichting van het
24
OCC. De besturing van de hub bleef geografisch en organisatorisch gescheiden van de besturing
van het netwerk.
Aan de oprichting van het Operationeel Controle Centrum (OCC) in de periode 1997 tot 1999,
ging een lange voorbereiding vooraf. Alle functies/bloedgroepen die een besturende rol hadden
op de dag van uitvoering in het wereldwijde netwerk gingen bij elkaar in een gebouw. Dit
gebouw werd er speciaal voor neergezet en ontworpen. Perioden van gezamenlijke training
namen veel tijd in beslag voordat de fysieke inhuizing plaatsvond.
Om het OCC goed te verankeren in het voorbereidingsproces van de operatie, werd er besloten
tot de start van het Programma Operationele Integriteit.
Dit moest zekerstellen dat de uitvoering startte met een haalbaar plan dat gebaseerd was op een
goed doordacht netwerkontwerp en een haalbaar serviceproces.
Deze twee, bij een airline voortdurend veranderende processen, gaan steeds door, maar moeten
twee maal per jaar een uitvoerbaar plan opleveren met de laatste goedgekeurde ideeën op het
gebied van netwerk en service. Het besluitvormingsproces voor beide processen was gescheiden
en een integraal besluitvormend circuit was hard nodig om de spanning tussen plan en
uitvoering weg te nemen en de benodigde veranderingen in netwerkontwikkeling en
serviceontwerp op gang te houden.
Dit herontwerp heeft voor een deel parallel met en voor een deel na de oprichting van het OCC
plaatsgevonden.
De oplossing bestond uit de verdeling van het hoofdproces van de onderneming in een aantal
fasen: strategisch, tactisch en operationeel(zie Fig.1). In de strategische fase, met een horizon
tussen 10 en 1 jaar, maakt de directie een match tussen wat zij in commercieel opzicht nodig
acht en wat operationeel en financieel haalbaar is.
In de tactische fase bepalen netwerk en service hun doelen en maken de match met operationele
haalbaarheid. Deze fase loopt van een jaar vooruit tot drie maanden voor de dag van uitvoering.
In de operationele fase, van drie maanden voor tot kort na de dag van uitvoering, zorgt het OCC
ten slotte dat de fit tussen wat commercie wilde en wat operationeel kon, in takt blijft.
In deze operationele fase blijven bescheiden bijstellingen van de plannen onderdeel uitmaken
van de werkelijkheid. De zomer en de winterdienstregeling werden beiden in tweeën gedeeld
om wijzigingen te vergemakkelijken en werden Operationele Plannen (OP) genoemd.
25
Figuur 1.
Bovendien werd er een scheiding gemaakt tussen “Wat” er moest gebeuren in opdracht van de
verschillende commerciële afdelingen, de functionele specificatie, en “Hoe” het uitgevoerd
moest worden door een groot aantal operationele afdelingen, de technische specificatie. Dit
leidde tot een heldere scheiding der verantwoordelijkheden en schiep ruimte aan de operationele
kant om de mogelijkheid van kostenverlaging te vergroten.
Er ontbrak in deze veranderingen nog een mechanisme om de tactische fase en operationele fase
vorm te geven. Om bij te dragen aan het ontwerp hiervan legde Prof. Bertrand zijn auditrol en
bespiegelende rol, die hij tot nu toe in het redesign had gehad, af en ging een actieve rol spelen
in het herontwerp zelf.
1.2. De ontwerpen In de marge van bovengenoemde veranderingen werkten wij ook aan praktisch redesign werk.
Demand
Matching
Supply
Delivery Evaluation
Integrate Supply
Schedule planning Adjustments Execution Evaluation
Development / Acceptance
Planning / Preparation
Process Control
Matching
Staying Matched
Specify Demand
Integrate Demand
Specify Supply
Demand Management
Ope
ratio
nal P
lan
Con
trac
t
Operational Plan Management & Control (including flexibility)
Perf
orm
ance
Mea
sure
men
t / E
valu
atio
n Market
Strategy
Operations
Strategy
Matching
Strategic Operational Tactical
3 months
1 year
10 years
Day-of-delivery
26
1.2.1. Tijdens het onderzoek naar de oorzaken van het slecht functioneren van het Europese
vliegnetwerk, bleek er geen verband te zijn tussen de gehanteerde vliegtijdnormen en de
gerealiseerde vliegtijden.
De afdelingen OCC en Netwerk namen samen een LMS student in dienst, om begeleid door
Prof. Bertrand te komen tot een elegant Ontwerp van een planningsmethodiek voor de KLM
dienstregeling. Dit heeft, als ik mij goed herinner, ook de tweede prijs van het Stan Ackermans
instituut (ontwerpersopleidingen) gewonnen.
1.2.2. In het OCC was er bovendien een een probleem met het bepalen van de hoeveelheid
reservecapaciteit. Een andere LMS student heeft daarom gewerkt aan een methodiek voor de
bepaling van de tijdbuffers en reservecapaciteit die aanwezig moeten zijn om onder normale
omstandigheden verstoringen op te kunnen vangen, zodat deze niet blijvend doorwerken in de
rest van de dienstregeling.
De twee ontwikkelde tools zijn zowel bruikbaar bij het ontwerp, de acceptatie, als de bijstelling
van de dienstregeling.
2. Het Building Block project
De aanleiding voor het project was de introductie van duidelijke verantwoordelijkheden bij de
uitvoerende en commerciële partijen en een verzakelijking op het raakvlak tussen beide hier
bovengenoemde partijen. Door de uitwerking van de wat/hoe scheiding moest de aansturing van
de commerciële afdelingen naar de uitvoerende divisies eenduidiger worden. Daarnaast werden
de processen als te complex en niet transparant ervaren en was er geen sprake van
procesverantwoordelijkheden en, niet onbelangrijk, waren de kosten onvoldoende beheerst.
In de zomer van 1999 is het Building Block en Compositie Regel concept geïntroduceerd en in
het najaar is een feasibility study uitgevoerd. Eind 1999 is een projectgroep uit alle diensten van
de KLM gestart. Ik was daarin projectleider en prof. Bertrand zat in een adviesrol en steunde het
programma met zijn autoriteit.
Zomer 2000 is de introductie ervan in de uitvoering geweest, terwijl in de tussenliggende
periode een aantal workshops zijn geweest met het MTR (Management Team Rederij). Het
MTR trad op als opdrachtgever.
27
Figuur 2
Het uitvoerend proces van de KLM is in 7 stukken geknipt om daar de bouwstenen (Building
Blocks) van te maken voor een dienstregeling (MPS) van de KLM. Deze opdeling is te vinden
in Figuur 2 (PAX = passagierstroom; BAX = bagagestroom).
Blok één representeert links de aankomende vlucht en rechts de vertrekkende vlucht.
Blok twee en vier staan voor het leegmaken van het vliegtuig na aankomst resp. klaarmaken van
het vliegtuig voor vertrek. Blok drie staat voor toestel staat in onderhoud,staat reserve of wacht
op een volgende taak ergens op het vliegveld.
Vijf, zes en zeven zijn de stromen passagiers en koffers gerelateerd aan resp. overstappen,
arriveren of vertrekken. De overstappende stroom is bij de KLM 70% van de passagiers. Met
deze zeven blokken (BB) en een aantal compositieregels (CR) kan er een zogenaamd Master
Productie Schedule gemaakt worden.
Van belang daarbij is uiteraard de mogelijkheid om snel en efficiënt te komen tot de vaststelling
van het optimale MPS (zie Fig. 3) vanuit een catalogus van integraal en gestructureerd
ontwikkelde BB’s en CR’s, waarbij het bijbehorend OP betrouwbaar en efficiënt kan worden
gerealiseerd door de CSP’s (uitvoerende divisies).
1 = flight
2 = arriving aircraft
3 la o er aircraft
= pax/bax
= aircraft
1 1 2
3
4
5
6 7
= pax/bax
= aircraft
1 1 2
3
4
5
6 7
28
Figuur 3
Teneinde de business control te verbeteren, kent de tactische fase zes doelstellingen:
Kostenbeheersing
Heldere interface tussen commercie en operatie
Een verantwoordelijke per processtap
Een transparant en efficiënt produktontwikkelingsproces
Een helder en efficiënt opgezet MPS/OP
Een integer en reproduceerbaar uitvoeringsproces.
Deze doelen zijn in een leerproces met de hele organisatie langzaam op zijn plaats gekomen en
er is een hele organisatie opgebouwd om er dagelijks mee om te gaan. Er zijn eigenaren van de
procesbouwstenen (Building Blocks), die tevens voorzitter zijn van een assemblageteam waarin
de ontwikkelingen in de BB werden afgestemd. Er zijn performance meetings voor de
dagelijkse operationele prestaties.
In deze fase heeft prof. Bertrand de relatie gelegd met de NS en die hebben we twee keer
bezocht om ervaringen uit te wisselen.
3. De hub Schiphol
3.1. Visie op de hub Schiphol Ik ben begin 2001 naar de hub, de divisie grondafhandeling, gegaan en heb het inleerproces van
het BBCR concept overgedragen aan mijn opvolger en Prof. Bertrand heeft ook hem begeleid
BB ontwikkeling
CR ontwikkeling
catalogus
MPS ontwikkeling
OP
Regie MTR
Service
Network
T ffi flow
BB/CR worden vastgesteld BB/CR staan vast
3 maanden voor dag van uitvoering
1 jaar voor dag van uitvoering
Operationele Fase
Strategiche Fase
e
Tactishe Fase
29
en geadviseerd. In de jaren erna hebben we het model voor de hub ontwikkeld, vervolmaakt en
gedetailleerd.
Om de juiste blik op de ontwikkeling van de hub te krijgen en daarmee de context voor mijn
eigen bedrijfsonderdeel Aircraft Services helder te krijgen, zijn we in overleg met de
divisiedirectie met een werkgroep gestart samen met de centrale besturende afdeling op de hub,
het hub controle centrum. De conclusie van die werkgroep was dat het managen van al die
stromen passagiers en koffers vroeg om ketenmanagement van alle deelprocesstappen om die
stromen maximaal te ondersteunen. Voor het bagageproces kwamen er in die tijd revolutionaire
conclusies uit, en wel dat de afdeling die de koffers in en uit het vliegtuig haalde en de afdeling
die de koffers sorteerde onder een en dezelfde baas zouden moeten vallen om de stroom koffers
maximaal te ondersteunen en de toen heftige aansluitings problemen tussen die afdelingen op te
lossen. Door de stromen bagage en passagiers zoveel mogelijk te synchroniseren, werd
voorkomen dat koffers kwijt raakten (irrate) of dat we de passagiers gingen missen.
Figuur 4
De passagiersstroom vroeg bij aankomst oppakken en passagiers de goede weg wijzen
voorkwam problemen later in het traject. Het gaat pas echt mis als de stromen tot stilstand
komen, hetgeen we met name zien we bij de chaos die ontstaat bij slecht weer. Dus hou de
flows stromende !
Synchronisatie met betrekking tot het vliegtuig bleek gecompliceerder. Dit werd vooral
veroorzaakt door het feit dat er zeer veel verschillende partijen aan het vliegtuig werkten en er
niet gecoördineerd werd. Bovendien wordt er ook nog eens met het vliegtuig geschoven op
Schiphol: van gate naar buffer naar nog een buffer, dan naar onderhoud en daarna naar de gate
voor vertrek. Schiphol is uniek in het aantal verplaatsingen van vliegtuigen. Dit leidde tot een
enorme complexiteit aangaande de besturing van de vliegtuigstroom, en om de synchronisatie
tussen vliegtuig enerzijds en pax/baxstromen anderzijds goed te laten verlopen. De centrale
besturing werd verantwoordelijk voor het synchronisatie-mechanisme (zie Fig. 4).
aankomst transfer/lokaal vertrek
BAX Flow
A/C FlowCentrale
Besturing
PAX Flow
aankomst transfer/lokaal vertrek
BAX Flow
A/C FlowCentrale
Besturing
PAX Flow
30
3.2. De ontwerpen Ondertussen deden we ook de eerste studies naar mogelijkheden om de kosten te verlagen en de
prestaties te verbeteren. In die periode hadden we geen duidelijke opvattingen over de
organisatie van het bedrijfsonderdeel. Dit bestond uit een aantal functionele afdelingen: de
sleepdienst voor het verplaatsen van vliegtuigen, de push back service voor het achteruit
drukken van vliegtuigen voor het vertrek, beiden samen een afdeling. Waterservice, toiletservice
en het rijden van los equipment rond het vliegtuig waren gecombineerd in een afdeling aqua. De
de-icing afdeling zat op dezelfde plek als aqua. De tankdienst, in bezit van een airline, en als
laatste de schoonmaak- en de cateringdiensten. Deze afdelingen gingen samen ongeveer 1,25
milj. keer per jaar naar het vliegtuig. Deels met eigen mensen en middelen, deels met mensen en
middelen uitbesteed zoals de schoonmaak maar dan wel voorzien van contractmanagers.
3.2.1. Houben (ref. 3) deed de eerste verkennende studie naar mogelijkheden voor verbetering.
Hij adviseerde, in een vroeg stadium, beter te controleren of de dienstregeling leidde tot een
voor de afdeling haalbare werkbelasting. Hij legde de eerste verbanden tussen de afdelingen en
de mogelijkheden om prioriteiten te stellen. Hij stelde voor om board supply als een functie
erbij te nemen en te combineren met schoonmaak om zo de prestaties overall te verbeteren.
Naast het beter aansturen van de push back deed hij ook suggesties voor herverdeling van de
besturing en aanbrengen van een gelaagdheid in de organisatiestructuur.
4. De Aircraft Flow
4.1. Het proces: Gereedstelling, Verplaatsing en Vluchtondersteuning In de zoektocht naar een structuur in de vliegtuig flow concentreerden we ons op de tien
processen binnen het bedrijfsonderdeel Aircraft Services (AS). Het eerste hoofdproces dat we
vonden binnen de aircraft flow was het gereedstellingsproces. Om de performance te verbeteren
werd volgens het idee van Houben het Board Supply proces van een andere KLM afdeling
overgenomen en gecombineerd met het schoonmaakproces. Daarnaast gingen we op zoek naar
nieuwe schoonmaakpartners en we betaalden hen beter. De combinatie van acties deed de
performance met een sprong stijgen. We voegden daar het halen en brengen van catering aan toe
en completeerden het met water, toilet en brandstof service en het plaatsen van los benodigd
equipment rond het vliegtuig. Het is complex maar daarmee is het toestel weer klaar voor een
volgend vertrek en de ontvangst van Pax en Bax. Om de afstemming van push-back en slepen
beter te laten verlopen werd de regie in het HCC naast die van de vliegtuigopstelplaatsen gezet.
Als volgende verbetering werd de de-icing en de push back beter aan de regie van het
31
vluchtproces gekoppeld; dit werd vluchtondersteuning genoemd. Later is de Bruggendienst
overgenomen en in vluchtondersteuning geplaatst, evenals de Security aan boord welke een
onderdeel van gereedstelling is geworden. Daarmee viel gereedstelling in building block 2 en 4.
Vluchtondersteuning viel in building block 1 en was slepen het verband tussen 2, 3 en 4.
Op deze wijze geanalyseerd bestaat de Aircraft flow binnen AS uit de drie bovengenoemde
activiteiten (zie Fig. 5).
Naast AS hebben ook Passage en Bagage beiden een activiteit in de aircraft flow. Bovendien
heeft het onderhoud een activiteit en als laatste heeft het OCC het beheer van lege vliegtuigen.
Het geheel is dus een complex geheel van veel partijen.
Figuur 5
4.2. De besturing van de Aircraft flow Traditioneel wordt bij de KLM al het platform werk geregisseerd door groepen ervaren ex-
uitvoerenden die met portofoon als verbindingsmiddel en allerlei varianten van een handmatig
planbord proberen de uitvoerenden van hun functionele afdeling op het platform van vliegtuig
naar vliegtuig te sturen. Met het bovenstaande in gedachten hebben we het ambitieuze plan
opgevat om het geheel in drie fasen te automatiseren. Eerst de planborden in digitale vorm om
zetten, daarna al de teams en voertuigen eraan koppelen met hand held terminals die gestuurd
worden door de planborden, en deze na ontvangst van hun werkopdracht in verschillende
stappen de status van hun werk terug te laten melden en als derde en laatste stap de planborden
met een algoritme de indeling te laten automatiseren (zie Fig. 6). Een volgende stap zou zijn om
de digitale planborden van gereedstelling aan elkaar te koppelen en als laatste gereedstelling,
verplaatsing en vluchtondersteuning aan elkaar te koppelen. Zo zien we binnen de flow drie
besturingsniveaus om te komen tot integrale ketenaansturing, plus een niveau voor de hub als
geheel. We hebben dus vier niveaus van besturing voor de hele hub.
vlucht vluchtde-boarden, lossen boarden, laden
gereedstellen,maintenance,…
gereedstellen,maintenance,
…
gereed-stellen,mainte-nance,
…
vluchton-dersteunen
aankomst-
service
verplaatsen
verplaatsen
doors open doors closed
vlucht vluchtde-boarden, lossen boarden, laden
gereedstellen,maintenance,…
gereedstellen,maintenance,
…
gereed-stellen,mainte-nance,
…
vluchton-dersteunen
aankomst-
service
verplaatsen
verplaatsen
doors open doors closed
32
Figuur 6
Eind 2004 heb ik prof. Bertrand en prof. Kroon (Erasmus in Rotterdam) gevraagd als auditor de
plannen te evalueren op haalbaarheid. Hun conclusie was positief, met de kanttekening dat het
een uiterst ambitieus plan was. Het project met al zijn organisatie en besturingsveranderingen en
de IT, zal met zijn afronding begin 2011 zes jaar geduurd hebben.
Er zijn nog twee verdere ontwikkelingen uit deze plannen gekomen en die zullen in hoofdstuk 5
aan de orde komen.
4.3. De ontwerpen
Het bovenstaande proces- en besturingsmodel werd ontwikkeld door een team binnen het
bedrijfsonderdeel ondersteund door prof. Bertrand en een zevental studenten. Daarbij werden
tevens de fundamenten gelegd voor een geavanceerd geautomatiseerd besturingsmodel dat ook
kort beschreven is. De zeven projecten waren:
4.3.1. Een vervolgstudie naar de opzet van de besturing in meerdere lagen, bij de gereedstelling,
de zeven sub processen, en bij gereedstelling en verplaatsing én het bovenliggende coördinatie
mechanisme (M. Schiebaan). Door de leiding is gekozen om slepen tot uitgangspunt te nemen
en alleen wijziging van sleeptijden aan te vragen als het schema onhaalbaar wordt.
Gedigitaliseerde regie (per AS proces) - CHIP elektronisch planbord - Mobiele datacommunicatie (MDC)
Geautomatiseerde indeling (per AS proces) - CHIP met ‘indeelautomaten’
Centrale regie voor de hoofdprocessen Verplaatsing, Gereedstelling en Vlucht-ondersteuning
- Centrale CHIP functionaliteit
Ni i i i Veranderings
Doelstellingen
• Het inzicht in de status van de processen vergroten ten behoeve van centrale besluitvorming
• Een efficiëntere regie realiseren: automatisering en integrale ketenaansturing
• Een efficiëntere uitvoering realiseren: wachttijden-vermindering
33
4.3.2. Een onderzoek naar de inzet van de resources in de processen en hoe dat optimaal kan
gebeuren gegeven de onderlinge beïnvloedingsmogelijkheden tussen de processen (R. Graste).
Ook daar komt de invloedrijke rol van de sleepdienst als procesverstoorder voor catering en
tanken naar voren .
4.3.3. Onderzoek naar de relatie tussen slepen /push back en de gereedstelling en hoe een goed
schema voor de eerste ruimte biedt voot de tweede (M. Plonka). Binnen gereedstelling is
gezocht naar de maximale ruimte voor het tankproces door vroege berichtgeving en de bouw
van een simulatiemodel om het een en ander te toetsen.
4.3.4. Bouw, gebruikmakend van het Fleischmann algorithme, van een optimalisatie tool, met
vier sleepstrategien en zes planninghorizonten (Koray Hakan). Een halvering van het aantal
vertragingen lijkt mogelijk. Daarnaast lijkt de EBA-R de beste planning horizon.
4.3.5. Een buiten model probleem: Wat gaat er mis in de cabine en leidt al langdurig tot onvrede
bij de passagier ? (Karolina Omachel). Dit betreft zowel schoonmaak van als onderhoud aan het
interieur. In een lange zoektocht n.a.v. dat onderzoek hebben we die onvrede door een mix van
cabine onderhoud en schoonmaak voor het eerst sinds lang kunnen oplossen.
4.3.6. Ideeën voor een nieuw planning en control framework en een optimalisatie tool voor de
bouw van onze automatisering (Tony Tany). Daarnaast leidde dit ook tot een voorstel voor een
nieuwe strategy.
4.3.7. Afronding van het onderzoek voor dit deel met een implementatie ontwerp gebaseerd op
alle voorafgaande kennis (J. Perdaen). Dit leidde tot een hele mix van adviezen om een optimaal
draaiend geheel te krijgen.
34
Figuur 7
5. Verdere ontwikkelingen
5.1. Grondtijdmanagement De in hoofdstuk 4 geschetste ontwikkelingen geven een eerste inzicht in de complexiteit van de
aircraft flow. Andere spelers in het spel dan de daar genoemde zijn: passage en bagage (met het
lossen en laden van passagiers en koffers). De technische dienst met het lijn onderhouden het
OCC met de bewaking van de lange wachttijden van de vloot. Dit moet allemaal gecoördineerd
plaats vinden. Dit alles bij voorkeur parallel 30 keer tegelijk en efficiënt uitgevoerd. Er bestaat
hier geen integraal coördinatiemechanisme voor, het proces vindt nu zijn weg door een lange
historie en met veel intuïtie (zie Fig.7).
Het bedrijf heeft de ambitie om het kostbaarste productiemiddel, het vliegtuig, maximaal in de
lucht te houden en te benutten.
Dit legt een extra claim op de beperkte grondtijd en op de strijd voor kostbare minuten op de
grond. Een omdraai per dag extra maken is het verschil tussen winst en verlies in Europa. Deze
grondtijd (Fig.7) managen, met deze complexiteit,en onder de gegeven beperkingen is de
uitdaging waarvoor het bedrijf staat. Maar in een verzuilde structuur is het best lastig om dit
voldoende snel voor elkaar te krijgen. De uitgewerkte stukken en geïmplementeerde ideeën zijn
er om te helpen dit succes in afzienbare tijd te halen, als de fakkel maar opgepakt wordt.
ATA ATD
Coordination point Actual groundtime
PLANNED GROUNDTIME ACTUAL GROUNDTIME
AS PS BS E&M
35
5.2. Real time Hub Het concept van de proces keten met intelligente sturing uit hoofdstuk 4, uitgebreid tot grondtijd
management uit 5.1, is goed toe te passen op de procesflow van met name bagage, maar ook bij
passage is het een goed bruikbaar concept.
Is de vliegtuigketen opgebouwd uit volgtijdelijke en parallelle deelprocessen, de bagageketen is
ideaal sequentieel opgebouwd uit ongeveer tien stappen. Door alle stappen van de
bovengenoemde IT te voorzien en voorwaart te koppelen, ontstaan sturingsmogelijkheden die
we nu niet hebben. Iets vergelijkbaars is bedacht voor de passage keten.
Door deze drie ketens te koppelen en actief te synchroniseren zouden grote stappen voorwaarts
te maken moeten zijn in de prestaties, de coördinatie, maar ook met betrekking tot de efficiency.
Het HCC heeft deze coördinatiemodule al laten bouwen en wacht op aansluiting van de
onderliggende flows (Fig. 8)
Dit leidde tot een herdefinitie van de besturingslagen en die konden we weer vereenvoudigen.
Ook dit concept is beschreven, en na afronding van de financiën kan gestart worden met deze
lange termijn ontwikkeling voor zeker de komende vijf jaar.
5.3. Het onderzoek 5.3.1. In mei 2009 is een bijzonder goed onderzoek gedaan naar grondtijdmanagement en de
benefits (Theresia Yunita). Zij komt op ongewoon heldere wijze tot twee scenario’s: een
gematigd en een offensief scenario waarin bij de laatste optie alle mogelijkheden gerealiseerd
kunnen worden. Meer kwantitatief onderzoek kan dit in kaart brengen (ref 11)
Figuur 8
36
5.3.2. Ter afsluiting van deze reeks onderzoeken heeft een verkenning naar een nieuwe horizon
plaatsgevonden (K. Dekkers). Niet de dag van uitvoering, maar het voorbereidende traject:
plannen, roosteren en indelen van activiteiten. Dit is de start van een zoektocht naar nieuwe
mogelijkheden voor verbetering en besparing. (ref 12).
6. Rollen prof. Bertrand
De afgelopen twaalf jaar is in samenwerking met prof. Bertrand, in een aantal diverse rollen,
een bijdrage geleverd aan de ontwikkeling van het totale hoofdproces van de KLM .
Ter uniformering en voor het bijspijkeren van de kennis op logistiek gebied trad hij voor al het
hoger kader in de uitvoering van de KLM op als opleider in de hoofdprincipes van de logistiek.
Door de jaren heen gaf hij masterclasses aan hoger kader en daarnaast gaf hij incidenteel, op
verzoek, opleidingen en trainingen aan een groot aantal afdelingen.
Zijn andere rol was als auditor tijdens het programma Operationele Integriteit. In deze rol trad
hij met senior adviseur Bas Classen op als team. Dit team kwam iedere drie maanden langs om
te zien of er in het programma Operationele Integriteit voldoende voortgang gemaakt werd bij
de KLM, om ons de spiegel voor te houden, en ook om na te gaan of ook de koers werd
vastgehouden in overeenstemming met de plannen.
Hij begeleidde een twaalftal studenten van de postdoctorale opleiding Logistieke Besturings
Systemen, het latere LMS dat valt onder het Stan Ackermans Instituut. Zodoende heeft hij op
ieder niveau bijgedragen aan de systematische ontwikkeling van de beheersing van het
uitvoerende proces van de KLM.
Een andere rol was die van absolute autoriteit op het gebied van de ontwikkeling van de
Logistiek, zowel vanuit de theorie als vanuit de praktijk en op ieder niveau: vanaf de raad van
bestuur tot het uitvoerend kader. Dit was mede mogelijk door zijn brede en diepgaande kennis
in zijn vakgebied, een nuchtere persoonlijkheid en de interesse voor de mensen waarmee hij
werkt. Bovendien verbindt hij op zinvolle wijze en met groot gemak de theorie met de praktijk,
dit in de context van zijn absolute fascinatie voor het doorgronden van de soms zeer complexe
praktijk. Als laatste wil ik ook noemen de combinatie van betrokkenheid en het gemak waarmee
hij tegelijkertijd afstand kan nemen en overzicht kan houden.
Twaalf jaar teamwork tussen ons heeft bovenbeschreven schat aan kennis en onderzoek
opgeleverd.
37
Referenties
1. Slagt, A.W., mei 1999, Ontwerp van een planningsmethodiek voor de KLM dienstregeling
2. Moltzer, M., maart 2000, Planningsmethodiek reservecapaciteit vloot
3. Houben, N., februari 2002, A logistic redesign of Schiphol Brede Services to improve the
punctuality of the KLM
4. Schiebaan. H., sept 2002, Integral fleet service control
5. Graste, R., oktober 2004, Planning and control at Aircraft services
6. Plonka, M., aug. 2005, Design of an off-line schedule for aircraft towing and push back to
maximize the slack for gereedstelling processes
7. Hakan,,K., februari 2006, Online scheduling of towing process to facilitate on time
completion of gereedstelling processes
8. Omachel, K., februari 2006, Design of a management control model for cabin Quality at
KLM
9. Tany, T.W., september 2006, On line scheduling of towing and preparation processes to
increase the departure punctuality
10. Perdae, J.A.G., november 2006, Implementation design for better coördination between
towing and preparation processes
11. Yunita, T., mei 2009, Managing effective and efficient operations
12. Dekkers, K., juli 2010, A design or KLM aircraft services’ planning rostering indeling
chain
38
39
A REFERENCE MODEL FOR THE DESIGN
OF OPERATIONS PLANNING AND
CONTROL SYSTEMS IN THE FLOW
PROCESS INDUSTRIES
Jan C. Fransoo
Abstract I present a reference model for the production planning and control of a process industries
production unit. The reference model is aimed to serve as a basis for designing decision rules
for an actual production unit like this, and fits within the design methodology for operations
planning and control systems developed by Bertrand.
0. Preface
While the main scientific contribution of Will Bertrand in the academic literature is related to
workload control and related concepts such as due date setting, in fact Bertrand’s actual main
contribution has been to introduce a more systematic way of engineering (“designing” would be
the word used by Bertrand) production planning systems, or more general, systems for
operations planning and control. Many engineers with an Eindhoven degree have been trained in
this systematic way, in a graduate course that subsequently has been named Voortgezette
Productie- en Goederenstroombeheersing in the Technische Bedrijfskunde Ingenieurspro-
gramma and Design of Operations Planning and Control Systems in the Master Program
Operations Management & Logistics. The design methodology developed by Bertrand had
strong characteristics of an art, despite being systematic. The ambition of the research group
surrounding Bertrand around 2000, was to further develop this art into a method. An extensive
collaborative project took place, taking the CD production at EMI in Uden (Netherlands) as an
example. The idea was to more systematically develop the design rules, building further on the
notions presented in Bertrand et al. (1990). These efforts have never been completed and hence
published. The main results however can be summarized as follows:
40
• Primary processes are the start of the design process. The primary processes should be
carefully documented. In 2003, this lead to the development of the RMO framework,
implying that processes could be characterized by three types of units: Resources, Materials,
and Orders. Later, Bertrand further developed this into a separation between resources and
processes, where process is a more general description of the conversion, independent from
the specific resource.
• Reference models as a basis to design decision procedures. In order to develop actual
decision rules, within the designed control structure, engineers do not start blank, with just a
set of OR tools. In fact, virtually all production units can be reduced as belonging to a
certain class of units, depending on their material and resource complexity. For each class, a
basic model exists that describes the best way to control such a production unit class. These
basic models were denoted as reference models.
Basic version of reference models were developed at the time for job shops, assembly lines, and
process industry shops. Again, these were never finished and published. Over the past weeks I
have started to complete a document on the reference model for the flow process industries, a
core area of research of mine in the 1990s, and also the topic of my PhD dissertation supervised
by Will Bertrand. The current chapter in this Liber Amicorum is hence the first actual completed
reference model, and based on work that I did at the time jointly with Simme Douwe Flapper1
.
1. Introduction The process industries cover a wide variety of industries. In this Chapter, I limit myself to the
so-called flow process industries (Taylor et al. 1981, Wallace 1984, Fransoo & Rutten 1994).
The flow process industries are business that manufacture (semi)continuously a small number of
products on a facility which can be regarded in modeling terms as a single machine, i.e., in-
process buffers are not controlled as inventory points.
The APICS definition (Wallace, 1984) of process industries characterizes their general
characteristics as follows:
Process industries are businesses that add value to materials by mixing, separating, forming, or
chemical reactions. Processes may be either continuous or batch and generally require rigid
process control and high capital investment. 1 Especially the contributions in Section 5 on maintenance, co- and by-products, and rework are due to Flapper.
41
The definition indicates that the type of manufacturing process is one of the most important
characteristics. Mixing, separating, forming and chemical reactions are operations that usually
involve non-discrete products and materials. In order to efficiently conduct these processes,
companies usually need large installations, which tend to be very expensive. If demand is high,
this justifies continuous production (thus higher investment). If demand is low, the investment
into a large installation is not worthwhile, and production takes place in batches. Processes like
the ones include in the APICS definition are difficult to control which often results in typical
symptoms as variable yield and returning flows of material.
The literature lists many characteristics as “typical” of process industries, such as divergent bills
of material, unstable products, and storage constraints. Though these characteristics can be
found in process industries, they are not general in the sense that virtually all process industries
are characterized by these properties.
On the other hand, they are discriminating in the sense that they are found predominantly in
process industries and hardly in discrete industries. I refer to Fransoo and Rutten (1994) for a
review of these characteristics.
Building on a classification introduced by Taylor et al. (1981), who extended the work of Hayes
and Wheelwright (1979), and Wallace (1984), Fransoo and Rutten (1994) further specify the
distinction between batch process industries and flow process industries. In this Chapter, I limit
my analysis to flow process industries. Process/flow is defined by Wallace (1984) as:
A manufacturer who produces with minimal interruptions in any one production run or between
production runs of products which exhibit process characteristics such as liquids, fibres,
powders, gases.
In process/flow businesses, the lead time is mainly determined by the cycle time, i.e. the time
between two consecutive runs of the same product.
The actual processing time per unit is very small, but due to high change-over times and high
production speed, production orders are large. The number of different products is not only
limited, but there is also relatively little variety between the products. Little variety, low product
complexity and a small number of production steps cause all products to have the same routing.
Since the total market demand for the relatively small number of products is high, investments
in specialized single-purpose equipment are economically justifiable. The use of single-purpose
equipment simplifies the determination of available capacity: usually the installations are used
continuously (round-the clock production). The added value in general is quite low. Since the
production speed is very high, the material costs usually account for 60-70 per cent of the cost
price.” (Fransoo and Rutten, 1994).
42
In the next section, I will discuss the characteristics of flow process industries in detail, using
the Resource/Material/Order (RMO) framework. In Section 3, I describe the principles of
production control in flow process industries, while Section 4 contains a formal mathematical
description of the decision model. Where Sections 3 and 4 focus on an idealized situation, in the
sense that extreme values of the RMO framework are considered, Section 5 discusses
relaxations of these characteristics. I conclude in Section 6.
2. Characteristics
Characteristics of actual processes in reality vary widely, and every situation is very specific. In
this chapter, I will limit my analysis initially to an idealized situation. Relaxations of this
idealized situation are discussed in Section 5. In this Section, I describe the characteristics of my
idealized situation. It is based on the so-called “Elementary Flow Process Industry Production
System” developed in my PhD dissertation (Fransoo, 1993). The characteristics are however
described more explicitly and extensively than in my dissertation. Furthermore, I group the
characteristics using the R/M/O framework. Note that I have added one dimension to this
framework (namely E for economics), since particularly this economic characteristic is very
essential and drives the principles of production control in this situation.
2.1 R(esource) Single machine
I limit myself to the situation with one production resource. Obviously, if multiple parallel
machines exist, my model can be applied straightforwardly if a fixed allocation of products to
machines exists. Especially in situations with large set-up times, this fixed allocation is a
reasonable assumption, since products tend to be grouped by product similarity on machines
anyhow. In food and pharmaceuticals production, exchange of products across machines may be
limited by legal restrictions due to food safety authorities or so-called Good Manufacturing
Practices.
The notion of a single machine is a notion from a production control perspective. In reality,
there can be multiple machines in a line, without any buffering opportunity, and with high
production speeds. Effectively, this then means that I model the line as a single machine. The
high production speed generally also disallows the creation of intermediary buffers, since these
would consume a lot of space. Apart from this constraint from a space or economical
43
perspectives, there can also be a physical limitation on the buffer existence, since product may
need to be processed further without delay.
No uncertainty in processing time.
Products are produced in large volumes and in long production runs. This enables a lot of
learning to take place while producing. This means that in general the time required for the
processes is more or less given, and can assumed to be deterministic. In some industries, such as
pharmaceuticals, it is even required by law that the processes are completely under control and
reproducible.
Large setup times independent of sequence
Setup times need to be included in the model explicitly and cannot be included into the
processing times (as may be done in job shops where single order production takes place). Setup
times often also include start-up time, because it takes time to realize a certain constant working
environment, like a constant temperature and pressure all over a big tank. Moreover, quite often
cleaning is required and related to this there is often some time required to find out whether or
not a pipeline or tank is sufficiently clean. The reason why setup times often cannot be included
in the processing times is their importance for lot size decisions.
Independence from other resource decisions
I assume that other resource decisions, including maintenance planning and workforce planning,
do not affect the planning and control decisions. This means that I only consider the machine
capacity, which is assumed to be available 100% of the time.
2.1 M(aterial) Limited number of products
The number of products I consider is limited but greater than 1. In reality, the actual number of
products that is typically produced on one machine varies across industries, but in most cases is
more than five and less than a couple of dozen.
Supply of raw materials is not considered an issue
The material supply problem is not modeled as part of this production control problem. This is a
fair assumption, since generally the number of materials is very limited (so very easy to plan).
Furthermore products are either easy to acquire (commodities) with short lead times so they do
not require extensive looking ahead and being taken into account as constraints, or they are
more difficult to acquire but in that case so essential that for strategic reasons stocks are held
that are so high that from a production control perspective it is also not a problem (always
available). Strategic stocks can also be kept if materials are only available during a part of the
year (seasonal materials from agricultural sources) or if price fluctuations exist due to temporary
44
scarcities of the material. In the latter case, the level of inventory is rather determined by long-
term trading conditions than by (short-term) production control considerations.
2.2 O(rder) Low variance in demand volume
I assume there is low variance in demand. This is a reasonable assumption and can be explained
by a number of causes. Many flow process industries produce commodities. Commodities are
sold on a general anonymous market that also includes many traders. Other flow process
industries produce products that are maybe less common, but their customers tend to also be
very capacity-oriented so in fact the amount that they can process is limited by their resources.
It should be noted that this low variance in this latter case pertains mainly to variance in total
demand volume. There may be variance in mix.
2.3 O(rder)-R(esource) Production orders are different from customer orders.
This holds when business economic or technically determined production lot sizes are much
bigger than the orders from a number of customers, e.g. due to long setup times. Note that base
materials are usually ordered by a number of customers. This applies e.g. for the production of
standard beer.
2.6 E(conomic) Small added value related to (large) investment in resource
It can be observed in the far majority of flow process industries that margins are low. Products
that are sold are generally more commodity than specialty, and are sold in a business-to-
business setting. The cost of capital is very high, and hence equipment needs to be operated at a
very high level of utilization.
3. Principles of production control There are a number of important characteristics and dynamics to realize when designing the
production control system for this environment.
The first characteristic that is very important is based on the economics of the plant. Due to the
low margins and the high capital investment, the plant can only operate at profitability if the
utilization rate is very high. The utilization rate is a direct consequence of the selection of the
batch size, which determines the ratio between productive capacity (i.e., capacity used for
45
producing products) and setup capacity (i.e., capacity used for changing over from one product
to another).
The lot size selection is essentially the so-called economic lot-scheduling problem (ELSP),
which has been widely studied in the literature; usually under deterministic demand assumptions
(Elmaghraby, 1978). A way of dealing with the ELSP under capacity constraints has been
described well by Silver et al. (1998, Section 11.6), who summarize earlier research results in a
clear policy. They restrict themselves to pure rotation cycles, i.e., the cycle time for each of the
products is identical.
If the demand, set-up time, or cost parameters strongly differ per product, a pure rotation cycle
is not close to optimal, and a cycle time needs to be determined for each product individually. In
this case, the challenge is to construct the cycle times such that the resulting batches can
actually be incorporated into a schedule. This problem is NP-hard, but there are a number of
solution approaches that give reasonable solutions at the expense of introducing slack into the
system. I will abstain here from discussing those, but refer to some key publications, notably
Doll and Whybark (1973), Elmaghraby (1978), and Roundy (1989).
All of these approaches assume deterministic demand. Although demand variability tends to be
very low in the flow process industries, my reference model should be able to cope with a small
amount of variability. There are roughly a few ways of dealing with this demand uncertainty:
• Stochastic Lot Scheduling Problem formulation
Under this policy, the demand variability is explicitly included into the ELSP model
formulation. A schedule is created in advance and after a disruption an algorithm is applied
to return as soon as possible to the original schedule without losing capacity in the long run.
Federgruen and Katalan (1996) present a rather complex stochastic procedure, while
Gallego (1990) presents a procedure, based on control theory. Winands et al. (2011) have
provided a recent review.
• Fixed Cycle Times and Variable Lot Sizes
Under this policy, target cycle times are determined for each of the products assuming
deterministic demand (where the expected demand level is taken as the deterministic
demand). This decision is taken at a higher planning level, where also safety stock levels are
set. At the lower planning level, orders are accepted as long as there is inventory available,
and lots are produced such that the target cycles are maintained. This policy has been
proposed by Bertrand et al. (1990) and further researched and developed in my dissertation
(Fransoo, 1993) and will be described in further detail below.
46
• Fixed Lot Sizes and Variable Cycle Times
Under this policy, at the higher level the lot sizes are determined based on an ELSP
formulation, while at the lower level production takes place in exactly these lots, but
changing sequence if certain products run out of stock.
• Variables Cycle Times and Variable Lot Sizes
Under this policy, target cycles are determined at the higher level in a similar way to my
approach and executed at the lower level, with the exception that once a product runs out of
stock, it is allowed to (temporarily) reduce the cycle length (Leachman and Gascon, 1988). I
showed (Fransoo, 1995) that this policy operates well under lower levels of utilization, as
may be present in Consumer Packaged Goods industries. It does not perform well under the
high levels of utilization I consider in this Chapter and that are typical of flow process
industries.
In this reference model I use the fixed cycle times policy (Bertrand et al. 1990, Fransoo 1993,
Fransoo et al.)1995, in which the control is two-tiered:
• At the higher level, target lot sizes are determined using, e.g., the procedure described by
Silver et al. (1998) (for the pure rotation cycle) or developed by Fransoo (1993) (based on
Doll & Whybark, 1973) (for coordinated yet different cycle times per product). These target
lot sizes lead effectively to target cycles
• At the detailed level, two functions are being executed:
o order acceptance, through which it is ensured that the aggregate inventory remained
remains in balance with the cycle time
o actual lot-sizing, which could be based on a base stock level determined based on
the target lot sizes and the safety stock
Effectively, this leads to a cycle time being kept stable at the operational level due to order
acceptance and lot-sizing being executed in-sync.
In the next Section, I will describe this model formally.
4. Formal model for production control The model consists of two levels, namely a top level and a base level. At the top level, the target
cycle times are determined, while at the detailed level the operational lot sizes are determined.
As indicated above, I will illustrate this only for the pure rotation cycle, but using other
47
deterministic models (See, e.g. Fransoo 1993, Doll and Whybark 1973, and Roundy 1989) the
same principles would apply in the hierarchical design of the control structure.
At the top level, the pure rotation cycle length is determined in two steps. In the first step, the
cost function is minimized, trading off inventory and setup cost. In the second step, the
minimum feasible cycle time is determined. We use the notation and model provided by Silver
et al. (1998).
Step 1. Determination of the optimal pure rotation cycle time, assuming no capacity
restrictions exist.
This can easily be determined as an extension to an EOQ type formulation. The total cost for a
product i that can be influenced by this decision, is the sum of setup and holding cost:
−+=
i
iii
ii p
dTdhTuTC 1
21)(
The total cost for all items then is
∑=
=n
i
i TCTC1
)()(
I now take the derivative of C(T) and set it equal to zero, rendering
∑
∑
∑
=
=
=
−
=⇔
=
−+
−=
n
i i
iii
n
ii
u
n
i i
iii
i
pddh
u
T
pddh
TuTC
1
1*
12
1
2
0121)('
Note that this T* is unconstrained, hence the notation Tu*.
Step 2. Determine the minimal required length of the cycle time to meet all demand.
The optimal T is constrained by the available capacity (unlike Tu*), namely, the length of the
cycle needs to be greater than or equal to the sum of setup and production time:
48
∑
∑
∑
=
=
=
−
≥⇔
+≥
n
i i
i
n
i
i
c
n
i i
ciic
pd
s
T
pTdsT
1
1*
1
**
1
It is clear that the optimal capacitated cycle time then equals the maximum of Tu* and Tc*. This
T* we consider the target cycle time.
At the bottom level, the operational lot sizes are determined. At the moment a product is due for
production, a quantity is being produced up to a defined base stock level of the product.
Essentially this implies that the quantity that is being produced is the quantity that was sold
since the last time production had started.
Note that this simple base level rule can only operate feasibly if at the aggregate level the order
acceptance function is managed such that the total capacity-equivalent demand sold during the
length of a cycle is not more than the total capacity-equivalent production during one cycle.
5. Exceptions and extensions to the basic model The principles and model described above have been designed for the idealized situation in flow
process industries as defined in Section 2. As such, this model can serve as a reference model
when designing a production planning and control system for a particular situation in real life.
In real life, however, there will be particular characteristics that are not covered in the reference
model. In this Section, I will briefly discuss a number of those characteristics.
For some of these characteristics, the reference model can be easily extended: the basic
characteristics of the model remain the same, while maybe some extra constraints may need to
be added or an extra decision function inserted to deal with an issue prior to applying the
proposed reference model. I will discuss these characteristics and a brief comment on the
changes to the reference model in Section 5.1 (extensions). For some other specific
characteristics, the emphasis of production control may change so dramatically, that this can no
49
longer be considered an application or extension of this reference model. These exceptions will
be discussed in Section 0.
5.1 Extensions Inventory constraints
Constraints may exist on the maximum amount of inventory that can be stocked. Also in the
flow process industries these constraints may concern inventories of input materials, products-
in-process and end products.
Restrictions on storage capacity for input materials play an important role when companies are
forced to accept and process all what is delivered. This form supply driven processing is no
uncommon in the process industries. For instance, in dairy cooperatives the factories usually
have an obligation to accept all of the milk supplied. Also in the petrochemical industry,
installations like refineries and crackers are never stopped and downstream operations have to
process all material that is supplied. In dairy, this is usually dealt with by creating overcapacity
in one of the downstream operations that can covert the product into a product with very long
shelf life, for instance in the production of milk powder. In the petrochemical industry, this
“valve” consists of extra sales of the superfluous products on the commodity market, usually at
lower prices than desired. In other cases, there may be other options such as subcontracting
production or storage capacity.
Restrictions on output storage capacity can be included and dealt with as indicated in step 2 in
Section 4. In this case, rather than a minimum cycle length, additionally a maximum cycle
length is imposed. The constraint could be a consequence at the individual product level (for
instance if each product uses a designated tank) or at the aggregate level (if they use a common
warehouse).
Product deterioration / perishable products
Product quality may deteriorate over time. This can pertain to both raw materials and finished
products. In the finished products case, this can simply be added as an upper bound on the target
cycle time in the model used at the higher level of decision making. For raw materials, this can
also be an effective solution, assuming that the moment of arrival of raw materials can be tuned
to the production schedule.
Deteriorating processes
In a number of situations, the production process itself may deteriorate. Sometimes this
deterioration can be automatically controlled and corrected. In other situations this requires
special activities by the operators or special people, for instance through rework.
50
In case of structurally deteriorating processes that cannot or only partly be corrected, it is
important to take this into account as an upper bound on the cycle time.
Co- and by-products
The distinction that is usually made between co- and by-products is that co-products are
considered to be marketable with a profit, while by-products are considered “waste” and have a
cost associated with them. The main issue in co- and by-products is to align the sales of the
products with the main product that they are produced with. If this can be modeled such that the
main product is leading, then the existence of co- and by-products in fact only leads to extra
inventory costs (for both co- and by-products) , extra sales (in case of co-products) or disposal
cost (in case of by-products), which can be incorporated into the cost of the main product
(Fandel, 1987). If it is essential that the production of the co- and by-products are taken into the
trade-off explicitly, then this should be treated as an exception. This holds for instance for
situations where one or more of the co-products that are produced can also be produced
separately and are as such included in the production cycle. Then the generation of the above
co-products as a result of the production of other products should explicitly be taken into
account when deciding on how much to produce of these co-products separately.
Parallel machines (without a fixed allocation)
It has not been researched whether this reference control model can still be applied when
parallel machines exist. under the reference model presented in Section 3. It can be questioned
whether such a complete flexibility is useful as we know from the flexibility literature ( Jordan
and Graves 1995) that only a very limited flexibility, if smartly designed through the chaining
concept, would reach almost the same benefits as full flexibility. This would entail that the
model described above could be used effectively, where a part of the production cycle is then
allocated to the group of products that has flexible allocation. Research needs to be conducted to
confirm this hypothesis.
Sequence dependent setup times
Research in my dissertation (Fransoo 1993) has demonstrated that allowing the sequence of
products to vary instead of using a fixed sequence only brings more than marginal benefits
when the number of products is large. In reality, when the number of products is large, the
length of the setups tends to be shorter.
Consequently, it can be assumed that long sequence dependent setup times virtually only occur
in cases with a relatively small number of products. The model suggested above can be easily
extended to cope with this by:
51
• modifying the higher level model to include determining a sequence. Many models are
available, generally based on a TSP formulation.
• modifying the lower level procedure to not allow any sequence changes.
If it turns out in reality that the number of products is very large and there are significant
sequence dependent setup times, while the products cannot be grouped into product groups, then
the model needs a comprehensive extension and it is not clear if that can be coped with using
the decomposition approach suggested in Section 4.
Many products
The models for determining cycle times discussed in the previous Section have no explicit
assumption about the number of products, so theoretically this number can be arbitrarily large.
However, in reality the number of different cycle times should be limited and the length of the
rotation cycle may also not be too long. This means that the number of products considered
should be limited to not more than a couple of dozen products. If more products exist, then
aggregation of products into product groups is necessary.
Non-stationary demand
In case of seasonal demand, an aggregate inventory or outsourcing plan needs to be made at a
planning level that is positioned above the decision levels discussed above.
Maintenance
Because facilities should be working 24/24 7/7, the planning of (preventive) maintenance
becomes much more important than in situations where not 7 days a week and not 24 hours a
day is produced, when preventive maintenance can be scheduled during the time that the facility
is not required for production like in the weekends. The maintenance activities usually require a
lot of time, partly because of the required cooling down and starting up of facilities before or
after maintenance. Preventive and corrective maintenance can be taken into account at a higher
level by a factor representing the effective availability of a facility and by safety stocks.
Preventive maintenance can be included in the production cycle as the production of an
imaginary product.
5.2 Exceptions Alternative recipes
In some industries, for instance cattle feed, the optimization of the use of raw materials is very
essential and determines largely the margin that can be obtained. This is due to price variations
of (mostly natural) raw materials. In this case, the actual usage of the resource is a secondary
52
problem subject to the use of raw materials. In the real life examples I know, setup times (the
main reason for producing cyclically) are very small. The reference model presented here hence
should not be used as a starting point for designing the production control system. We refer to
Bertrand and Rutten (1999) for resolving this issue.
Rework/Reprocessing
An essential difference between rework/reprocessing on the same facility as used for initial
production when compared with usual multi-product production on the same facility, is that the
“required inputs” for the rework/reprocessing are the outputs of production activities on the
same facility. This limits the possibilities for rework/reprocessing. The rework/reprocessing can
be included in the production cycle of the company taking into account the above “input
materials “ restriction in a number of ways:
• Directly after the production of a certain item in the cycle has finished (extension
production time in cycle (variable production time)),
• After a certain minimum quantity is available for rework
• After a number of cycles.
Note that rework may mean the mixing of batches with different qualities, which makes the
inclusion of rework more complex.
Note that the planning and control of rework/reprocessing may have features in common with
the planning and control of alternative recipes and co- and by-products. See further Flapper et
al. (2002).
6. Conclusions
In this Chapter, I have illustrated the use of reference models for the conceptual and detailed
design production control systems in the process industries. The model proposed is very much
in the tradition of the production control methods developed by Bertrand in the following sense:
• The model proposed is hierarchical, and clearly separates the setting of the key
parameters that impact performance from the operational control
• The model is simple and elegant, and is easily implementable
• The model limits itself to the key characteristics of the system under consideration
• The model is focused primarily on control (i.e., leading to a guaranteed performance)
and much less on optimality.
53
While this is the first time the model has been documented as an explicit reference model, its
main insights have been taught to students at Eindhoven since several decades. Dozens of
students and alumni have implemented these insights in the companies that they conducted their
Master Thesis at, or were working. The fixed cycle times hierarchical approach has proven its
performance hence not only conceptually, but at a large scale in the process industry.
References
Bertrand, J.W.M., and W.G.M.M. Rutten, 1999. Evaluation of three production planning
procedures for the use of recipe flexibility, European Journal of Operational Research 115:1,
179-194.
Bertrand, J.W.M., J.C. Wortmann, en J. Wijngaard, 1990. Productiebeheersing en Material
Management. Houten: Educatieve Partners Nederland.
Doll, C.L., and D.C. Whybark, 1973. An iterative procedure for the single-machine multi-
product lot scheduling problem, Management Science 20(1), 50-55.
Elmaghraby, S.E., 1978. The Economic Lot Scheduling Problem (ELSP): Review and
Extensions, Management Science 24: 6, 587-98.
Fandel, G., 1987. Surplus or disposal quantities in optimal program planning in joint
production, Engineering Costs and Production Economics 12:1-4, 143-158.
Federgruen, A., and Z. Katalan, 1996. The Stochastic Economic Lot Scheduling Problem:
Cyclical Base-Stock Policies with Idle Times, Management Science 42:6, 783-796.
Flapper, S. P., Fransoo, J. C., Broekmeulen, R. A. C. M. and Inderfurth, K., 2002. Planning and
control of rework in the process industries: A review. Production Planning & Control 13:1, 26-
34.
Fransoo, J.C., 1993. Production Control and Demand Management in Capacitated Flow
Process Industries, PhD Thesis, Eindhoven University of Technology.
Fransoo, J.C., and W.G.M.M. Rutten, 1994. A typology of production control situations in
process industries, International Journal of Operations & Production Management 14:12, 47-
57.
Fransoo, J.C., V. Sridharan, and J.W.M. Bertrand, 1995. A Hierarchical Approach for Capacity
Coordination in Multiple Products Single-machine Production Systems with Stationary
Stochastic Demands, European Journal of Operational Research 86: 1, 57-72.
54
Gallego, G., 1990. Scheduling the Production of Several Items with Random Demands in a
Single Facility. Management Science 36:12, 1957-1592.
Hayes, R.H., and S.C. Wheelwright, 1979. Link manufacturing process and product life cycles.
Harvard Business Review, 57:2, 127–136.
Jordan, W.C., and S.C. Graves, 1995. Principles on the Benefits of Manufacturing Process
Flexibility, Management Science 41:4, 577-594.
Leachman, R.C., and A. Gascon, 1988. A Heuristic Scheduling Policy for Multi-Item, Single-
Machine Production Systems with Time-Varying, Stochastic Demands, Management Science
34:3, 377-390.
Roundy, R., 1989. Rounding off to Powers of Two in Continuous Relaxations of Capacitated
Lot Sizing Problems. Management Science 35:12, 1433-1442.
Silver, E.A., D.F. Pyke, and R. Peterson, 1998. Inventory Management and Production
Planning and Scheduling. 3rd Ed. New York: John Wiley & Sons.
Taylor, S.G., S.M. Sewart, and S.F. Bolander, 1981. Why the process industries are different,
Production and Inventory Management Journal 22:4, 9-24.
Wallace, T.F. (Ed.), 1984. APICS Dictionary, 5th ed., Falls Church, VA: American Production
and Inventory Control Society.
Winands, E.M.M., I.J.B.F. Adan, and G.J. van Houtum, 2011. The stochastic economic lot
scheduling problem: A survey, European Journal of Operational Research 210: 1, 1-9.
55
OPERATIONS MANAGEMENT IN
HEALTHCARE
Jan M.H. Vissers and Guus de Vries
Abstract The paper reflects on Will Bertrand’s contribution to operations management in healthcare. We
describe the involvement of the department with healthcare over the years. We provide more
information on healthcare as operations management context. Then we describe Will’s main
contributions to operations management in healthcare, i.e. the doctoral theses supervised and
the development of a hierarchical production control framework for hospitals together with the
authors. We also reflect on the relevance and impact of this work on research and education, and
make some concluding remarks on its inspiration for future work.
1. Introduction
Healthcare has been an area of applied research for the Department Industrial Engineering and
Innovation Sciences and its predecessors for many years. The adventure with healthcare started
in 1970 with the ‘Hospital Research Project’ at the Department of Industrial Engineering and
Management Science. The ‘Hospital Research Project’ referred to a group of 8-10 researchers
from four subdepartments (Organisatiekunde, Organisatiepsychologie, Operations Research,
Bedrijfseconomie) which collaborated in a multidisciplinary way in the investigation of planning
and management issues in healthcare (Merkx 1974). Topics of research were amongst others:
team nursing in hospital wards, nursing workload, outpatient appointment systems, radiology
planning. After more than a decennium of growth in activities and output the ‘Hospital Research
Project’ lost around 1990 its department-wide role for coordination of research in healthcare, and
the subdepartments continued in their own way and pace. The subdepartment of ‘Quantitative
aspects of production control’ (a follow up of the subdepartment of Operations Research)
gradually took over the role of focus point for healthcare research, which in time evolved into a
small research group (Wim Monhemius, Michel Kirkels) with a number of PhD students (Guus
de Vries 1984, Rob Kusters 1988, Petra Peters-Groot 1993, Jan Vissers 1994). When Wim
56
Monhemius handed over to Will Bertrand in 1987, Will ‘inherited’ healthcare. Rumours at that
time went that this was not out of free ‘will’, but that it can be credited to Wim Monhemius’
convincing power that healthcare is an important area for contributions by industrial engineers
and industrial engineering research. In this way Will became involved in healthcare.
Will’s view on healthcare at that time was - and still is - that it is a not well-organized sector in
which many costs are made that can be avoided by better planning and management.
But more importantly, he became gradually convinced that healthcare is the ultimate test for
many operations management principles and planning approaches. This paper provides an
account of Will’s contribution to operations management in healthcare through the eyes of two
colleagues working as professors of health operations management at the EUR in Rotterdam, but
having collaborated for many years with Will2
.
2. Healthcare as operations management context
Healthcare has become a major industry, with many people involved either as employees in
healthcare delivery organisations or as consumers of health care services. The UK’s National
Health Service is actually the largest employer in Europe. The rising cost of healthcare due to
new technologies and demographic trends (in particular, the ageing population), is a vitally
important issue for health care policy makers. At the same time there is a paradigm shift in the
service concept of healthcare. Patients are no longer prepared to accept poor quality service,
either in terms of long waiting times or inconvenient appointment systems, and expect that
services are well organized from a “customer” perspective. The service concept has shifted from
optimizing the use of resources to finding a balance between service for patients and efficiency
for providers. These developments have had an impact on the popularity of Operations
Management/Operations Research (OM/OR) in health care not only in terms of the number of
OM/OR applications but also of the scope of topics covered.
The characteristics of health OM/OR - which make it different from OM/OR in industry or in
commercial services - stem from the way healthcare organisations operate and from the type of
healthcare system in use in a particular country. A hospital organisation, for instance, consists of
units (outpatient departments, wards, operating theatres, etc.) which contribute to the processes
delivered by clinical specialties. Hospital management does not always have much control over
output, as core processes are often controlled by clinical specialists who in many European
2 Jan Vissers first as his PhD student and thereafter as postdoc at the Faculty of Technology Management, Guus de Vries as parttime professor in Healthcare Operations Management at the Faculty of Technology Management of Eindhoven University of Technology.
57
countries have a contract with the hospital but are not salaried. Therefore, the line of command
structure in hospitals is not always straightforward. Decision making is carried out in more of a
political arena in which the interests of different stakeholders need to be balanced.
Standardisation of product and process is difficult due to the high variation between hospitals and
also between clinicians within the same specialty. Moreover, medical professionals want to keep
their autonomy in the care delivery process. Although doctors manage the clinical process, and
nurses the nursing process, no one is in charge of the customer process as a whole. This makes
managing the service quality from the perspective of the customer a real challenge.
The health care system in use in a given country is another important influential factor in the
health care industry. Health care systems vary between countries in terms of their incentives for
reducing waiting lists or controlling the costs of care.
Countries with a health care system with more market competition tend to put more effort into
service improvement, whereas countries with a budgeting system put more emphasis on
improving efficiency. This paper focuses on European health care systems, which typically
provide coverage of most health care costs for all inhabitants and enable the costs of health care
expenditure to be controlled at a national level. This is also true of the health care systems in a
few countries outside Europe, such as Canada and Australia.
3. Development of the framework for hospital production control
As the OM healthcare research group had developed knowledge on operations management of
hospitals in the years 1990-2000 through master and doctoral theses and project work of Will for
the Elisabeth hospital in Tilburg (and later-on for the Maxima Medisch Centrum in Veldhoven),
we decided to use this knowledge for developing a framework that would help hospitals to plan
and control hospital processes. Inspired by the hierarchical planning frameworks developed for
industry by Will in collaboration with Hans Wortmann and Jacob Wijngaard (Bertrand,
Wortmann and Wijngaard, 1990), we asked ourselves the following questions:
- Why do hospitals need a dedicated framework?
- In what way do hospitals differ from manufacturing organizations and what are the
reasons why hospitals require a framework tailored to their needs?
- What are the requirements of such a framework for hospital production control?
If a dedicated hospital production control framework is required, what are the design
requirements to develop such a framework?
- How does the framework look like?
58
- What would be necessary to coordinate processes and resources at different levels of
planning in a hospital to be in control of production, service and efficiency?
- How can the framework be used?
- What role can the framework fulfill in production control of hospitals?
We worked for over two years to answer these questions, resulting in two papers published in
Production Planning and Control - a journal that according to Will was the number one journal
that published control frameworks and was therefore fitting our purpose. As this journal was not
known in healthcare management research, we also made a condensed paper version for Acta
Hospitalia, a scientific journal published in Dutch by Centrum voor Ziekenhuis en Verplegings-
wetenschap in Leuven, which is read by researchers and managers interested in applied research
in healthcare management.
We will summarize the main finding on questions 1 and 2 in this paragraph, present the
framework in paragraph 4 and reflect on its use for hospitals in paragraph 5.
Hospital management has limited possibilities to control hospital production, as hospital
production processes are driven by medical specialists who, however, do not manage that
process. We consider therefore the hospital as a virtual organisation, consisting of a number of
relatively independent businesses in a common framework. Each business unit functions as a
focused factory for a range of more or less homogeneous products. Production control principles
can be applied to each of these businesses, but not to the system as a whole. A number of
elements from classical production control theory can be applied also to health care, i.e. the use
of decoupling points, the bottleneck oriented approach, and the operational control between
production and market. However, important factors that need to be considered in health
production control are that often specifications on quality are not available at the start of the
process, and that there is strong interaction between the patient and the process. The conclusion
is that a dedicated framework for approaching hospital production control is necessary. The
specific characteristics of hospital care and its state of production control development are the
main arguments for this dedicated framework.
4. Framework for hospital production control
The framework is based on an analysis of the design requirements for hospital production control
systems (De Vries, Bertrand and Vissers (1999)) and builds on the production control design
concepts developed in Bertrand et al. (1990). The design requirements are translated into the
59
control functions at different levels of planning required for hospital production control. This
translation is built on notions of the hospital as a virtual organization with patient groups as
business units and a focused factory approach for the production control per business unit. In
short, we can distinguish a number of production control functions, which can be positioned at
different levels of planning in a framework (see Table 1).
Table 1: Production control functions distinguished in the planning framework for hospitals
decision focus
1
range of the services, markets and product groups, long-term resource
requirements, centrally coordinated scarce resources;
contracted annual patient volumes, target service and efficiency levels
2 amount of resources available at annual level to specialties and patient groups,
regulations regarding resource-use
3 time-phased allocation of shared resources, involving specialist-time
detailed number of patients per period
4 urgency and service requirements, planning guidelines per patient group
5 scheduling of individual patients, according to guidelines at patient group level
and resource-use regulations at resource level.
At the highest level decisions have to be made on the range of services provided, the markets one
wants to operate in and the product groups for each market.
Also decisions have to be made on the long term resource requirements of the hospital, which
scarce resources are centrally coordinated, what level of annual patient volumes one wants to
achieve and what level of service one wants to target for. These are all longer term strategic
decisions, which essentially do not belong to the domain of OM, but which have impact on the
management of operations at shorter terms.
The next level focuses on the amount of resources that is available annually to specialties and
patient groups, to ensure that the contracted annual patient volume can be realized. At this level
also the rules for using the resources need to be established to ensure that the target service and
efficiency levels are achieved. At the third level the focus is on the allocation of shared resources
60
in time, taking into account the availability of specialists and seasonal developments. This
requires more insight into the detailed numbers per patient group per period within the year. At
level four the urgency and service requirements per patient group need to be established, and the
planning guidelines per patient group. The fifth level regards the scheduling of individual
patients, according to the planning guidelines for the patient group and the resource-use
regulations for the resources involved.
Though the planning framework seems to be working only top-down, the need for each level and
the requirements for coordination are established bottom-up. At the lowest level individual
patients are coupled to resources in the day to day scheduling. This level in the framework is
called patient planning and control.
The way patients are operationally scheduled needs to be governed by rules established at patient
group level. Oncology patients, for instance, have different urgency and service requirements
from patients with varicose veins. Therefore, operational scheduling of patients needs to be
governed by what we called patient group planning and control. To allow for the planning of a
patient group resources need to be allocated, taking into account the availability of specialists and
personnel. This level is called resources planning and control, and includes also the time-
phased allocation of resources. The level of resources required results from the annual patient
volumes contracted, and the service and efficiency levels targeted for. This level is called patient
volume planning and control. Finally, the volume level is governed by the strategic planning
level, where, for instance, decisions are taken about which resources need to be shared or not.
This level is called strategic planning. At this level there is no control involved.
These levels of planning can be further elaborated (Vissers, Bertrand and de Vries (2001))
resulting in the planning framework as shown in Figure 1.
5. The use of the framework for hospitals
A framework for a hospital does not describe the optimal way to control hospital activities but
instead describes a logical way of coordinating hospital activities within the perspective of the
current hospital organisation. It is about what to do, and not about how to do it. Therefore,
frameworks are not meant to be implemented as such. They rather serve as a reference
background for the development of hospital production control systems, to show the weak spots
where improvement is necessary, and to position contributions from logistic theories to issues of
planning in the wider context of hospital planning.
61
Figure 1: Framework for production control of hospitals
STRATEGIC PLANNING
specialties &productrangepatient groups asbusiness units
patient flows
collaboration & outsourcingshared resources
resources
volume contracts# patients per patientgroupservice levels
expected # patients perpatient groupcapacity requirementsper patient group
rough cut capacity checktarget occupancy levels
allocation of leading shared resourcesbatching rules for shared resources
patient flows resources
patient flows resources
PATIENT VOLUME PLANNING & CONTROL
RESOURCES PLANNING & CONTROL
feedback onrealizedpatient flows
feed forwardon impacts of changesin population& technology
feedback ontargets forresource utilisation
feed forwardon servicelevel standards
feedback oncapacity useby specialty& patient groups
feed forward on availablecapacity perpatient group& specialty
projected number of patients per period
availability of specialistcapacity
patient flows resources
scheduling of patientsfor visits, admission &examinations
allocation of capacityto individual patients
patients resources
feedback oncapacity usereadjustmentservice levelstandards
feed forwardon batchcomposition& schedulingrules
PATIENT GROUP PLANNING & CONTROL
PATIENT PLANNING & CONTROL
LT demand-supply match
demand-supplymatch
demand-supplyspecialty
demand-supplyseasons
demand-supplypeak hours
days-weeks
weeks-3months
3months-1year
1-2 years
2-5 years
restrictions ontypes of patients
restrictions on types of resources
restrictions ontotal patient volumes
restrictions onamount ofresources
restrictions ondetailed patient volumes
restrictions onresource availability
restrictions on thetiming ofpatient flows
restrictions on the timing ofresources
62
We will illustrate the importance of the hospital production control framework for its use for
research and for education.
Guidance for OM research on hospital production control
The framework has served for developing applications of production control approaches to
different planning issues in hospitals, often in the form of projects involving postmaster students
from Mathematics in Industry at Eindhoven University of Technology. This resulted often in an
applied research paper in the form of a case study in a specific hospital setting. Examples of such
projects are:
- Master scheduling of resources for inpatient planning, involving regular beds (wards),
intensive care beds (intensive care unit), operating theatre rooms (operating theatre
department), specialist capacity (medical specialists). This proved to be a very productive
research line with a number of publications: firstly, a mixed integer planning approach for
defining the optimal mix of elective patients to realise target levels of utilisation of resources
involved (Adan and Vissers, 2002; Vissers, Adan and Bekkers, 2005); secondly, a paper on
comparison of hospital admission systems (Vissers, Adan and Dellaert, 2007); thirdly, a
paper illustrating the effects of stochastic distributions (Adan et al, 2009); fourthly, a paper
extending the approach to include also emergency admissions (Adan et al, submitted 2010).
- A patient group based business planning model for a surgical specialty, involving outpatient
and diagnostic departments, the operating theatre department and wards (Vissers, Adan, van
den Heuvel and Wiersema, 2005).
- Master scheduling of medical specialists, involving outpatient departments, the operating
theatre department, wards, and different sites (Winants, de Kreuk and Vissers, 2005).
- Cardiology patient flow planning, involving outpatient departments, cardio care unit and
wards (Vissers and Croonen, 2005).
- The framework has also inspired research at other universities, for instance a PhD study in
Maastricht on hospital system design (Molema, 2009) and a PhD study in Rotterdam on
modelling and management of variation in the operating theatre (Stepaniak, 2010).
Textbook for students in healthcare management
The research projects on Operations Management performed in the period 1998-2005 provided
together an interesting selection of applications that could illustrate OM contributions to
production control of hospitals.
Due to a growing interest in operations management in master programmes for healthcare
management at the institute of Health Policy and Management of Erasmus University Rotterdam,
63
there was a need for a textbook for this target group, that would contain the basics of operations
management terminology and principles and applications of OM in healthcare. We developed
therefore a textbook with 6 chapters on OM healthcare basics (operations, processes, resources,
unit/chain/network logistics, control framework) and 10 chapters on applications of OM in
healthcare. Some of these applications were from colleagues abroad but most of them could be
based on the projects performed at Eindhoven University of Technology. The book ‘Health
Operations Management, Patient Flow Logistics in hospitals’ was published in 2005 (Vissers and
Beech, 2005), with two contributions from Will (Bertrand and de Vries, 2005; Vissers, Bertrand
and de Vries, 2005).
6. To conclude
Healthcare is an important service industry that will increasingly benefit from operations
management research. There is no doubt that this topic should be on the agenda for development
of research and education of the Department Industrial Engineering and Innovation Sciences.
That is also the reason why a number of years ago a healthcare track was developed, consisting
of a bachelor program 'Industrial Engineering for Health Care' and a master track Operations
Management & Logistics for Health Care.
Will Bertrand has played an important role to continue the line of healthcare based research from
the pioneering era of the Ziekenhuis Research Project to the current position with renewed
discussions on the department’s role in healthcare. Using the Subdepartment OPAC and the
research school Beta as research basis, a series of theses was produced and a framework for
hospital production control was developed, that has put Eindhoven on the map in the Dutch and
international research landscape.
The fact that the hospital production control framework has still ‘selling power’ after more than
ten years since its inception, shows that the research performed has sustainable (Will) quality.
Nevertheless, nothing lasts forever. Healthcare is a dynamic sector and operations management
practice is developing fast, which will bring new challenges for operations management.
One of the challenges in the near future will be to update the framework, to reflect the increased
interest in process improvements for better customer service and higher efficiency. After all,
healthcare is a service that we appreciate very much as consumer, but whose costs we would like
to limit as healthy citizens.
64
References
I.J.B.F. Adan and J.M.H. Vissers. ‘Patient mix optimisation in hospital admission planning: a
case study. Special issue on ‘operations management in health care’ of the International Journal
of Operations and Production Management, Volume 22, Number 4, 2002, pp.445-461.
Ivo Adan, Jos Bekkers, Nico Dellaert, Jan Vissers, Xiaoting Yu. ‘Patient mix optimisation and
stochastic resource requirements: a case study in cardiothoracic surgery planning’, Health Care
Management Science, 2009 12: 129-141.
Ivo Adan, Jos Bekker, Nico Dellaert, Jully Jeunet, Jan Vissers. ‘Improving operational
effectiveness of tactical master plans for emergency and elective patients under stochastic
demand and capacitated resources’. Submitted for publication, 2010.
Bertrand J.W.M., J.C. Wortmann and J. Wijngaard. Production Control: A Structural and
Design Oriented Approach. Elsevier, 1990, Amsterdam.
Will Bertrand and Guus de Vries. ‘Lessons to be learned from Operations Management’. In:
Vissers and Beech (eds.). Health Operations Management. Patient flow logistics in health care.
Routledge, London and New York, 2005, p. 15-38.
P.M.A. Groot. Decision support for admission planning under multiple resource constraints.
Thesis TUE 1993.
R.J. Kusters. Opnameplanning in Ziekenhuizen, Proefschrift TUE 1988.
R. Mercx. Overzicht Ziekenhuis Research Project. TUE, 1974.
F.W.M. Molema. Hospital system design. Creating supply flexibility to match demand
variability. Thesis Maastricht University, 2009.
Pieter S. Stepaniak. Modeling and Management of Variation in the Operating Theatre. Thesis
Erasmus University Rotterdam, 2010.
J.M.H. Vissers. Patient flow based allocation of hospital resources. Thesis TUE 1994.
J.M.H. Vissers, G. de Vries en J.W.M. Bertrand. Een raamwerk voor productiebesturing van een
ziekenhuis, gebaseerd op logistieke patiëntengroepen. Acta Hospitalia 2001-2, 33-51.
Vissers J.M.H., J.W.M. Bertrand and G. de Vries. A framework for production control in
healthcare organisations. Production Planning and Control, Vol. 12, 2001, no.6, 591-604.
65
Vissers and Beech (eds.). Health Operations Management. Patient Flow Logistics in Health
Care. Routledge, London and New York, 2005.
Jan Vissers, Will Bertrand and Guus de Vries. Frameworks for health operations management.
In: Vissers and Beech (eds.). Health Operations Management. Patient flow logistics in health
care. Routledge, London and New York, 2005, p. 84-94
Jan Vissers, Ivo Adan, Miquel van den Heuvel and Karin Wiersema. A patient group based
business planning model for a surgical specialty. In: Vissers and Beech (eds.). Health
Operations Management. Patient flow logistics in health care. Routledge, London and New
York, 2005, p. 202-223.
Jan Vissers and Gijs Croonen. Cardio care units: modelling the interaction of resources. In:
Vissers and Beech (eds.). Health Operations Management. Patient flow logistics in health care.
Routledge, London and New York, 2005, p. 249-263.
J. Vissers en G. De Vries. Sleutelen aan zorgprocessen. Intreerede ErasmusMC/iBMG
Rotterdam, april 2005.
J.M.H. Vissers, I.J.B.F. Adan and J.A.Bekkers. Patient mix optimization in cardiothoracic
surgery planning: a case study. IMA Journal of Management Mathematics (2005) 16, 281-304.
Jan M.H. Vissers, Ivo J.B.F. Adan and Nico P. Dellaert. Developing a platform for comparison
of hospital admission systems: an illustration. EJOR Volume 180, Issue 3, 1 August 2007, pp.
1290-1301.
G. de Vries. Evenwicht in zorgvraag en zorgaanbod; besturing van de afstemming op
verpleegafdelingen. Proefschrift TUE, 1984.
Vries G. de, J.W.M. Bertrand and J.M.H. Vissers. Design requirements for health care
production control systems. Production Planning and Control, volume 10, 1999, no. 6, 559-569.
Erik Winants, Anne de Kreuk and Jan Vissers. Master scheduling of medical specialists. In:
Vissers and Beech (eds.). Health Operations Management. Patient flow logistics in health care.
Routledge, London and New York, 2005, p. 184-201.
R.E. Wulff. Het ontwerpen van ziekenhuisorganisaties. Een onderzoek naar de
organisatiestructuur van het algemene ziekenhuis. Proefschrift TUE 1996.
66
67
THE EUT MAINTENANCE RESEARCH
Geert-Jan van Houtum
1. Introduction
Prof. Will Bertrand was the successor of Prof. Will Geraerds as professor in Production
Control within the Department of Industrial Engineering around 1990. Geraerds was appointed
as full professor from 1972-1991 on the chair ‘Produktieplanning en -besturing’ (Production
Planning and Control); see Geraerds (1973) and Bertrand et al. (1991). Bertrand was
appointed in 1989 on the chair ‘Produktiebeheersing’ (Production Control); see Bertrand
(1989). As explained in Geraerds (1970, Section 4), there are many similarities between
production and maintenance systems. Geraerds worked on production control, but clearly
specialized in maintenance control. Bertrand clearly specialized in production control, but he
also worked in the maintenance area (he has been the first promotor of three PhD students in
this area). This fact, together with my own interest in the area of maintenance (see Van Houtum,
2010), motivated me to look back into the history of maintenance research at the Eindhoven
University of Technology and to devote my contribution for this LiberAmicorum to this topic.
The title of my contribution is “The EUT Maintenance Research”, which is a slight variant on
“The EUT Maintenance Model”. The latter is a framework for maintenance that has been
developed by Geraerds (1988, 1992). It is a useful framework to classify problems and research.
Although the framework got well-known within The Netherlands, unfortunately, it did not get
established in the international scientific literature. Below, in Section 2, I will briefly describe
this framework, and hopefully this motivates some of the readers to study the original work of
Geraerds in further detail. Here, I will use the framework to position the maintenance PhD
research projects that have been and are executed at the EUT; see Section 3. This will show that
really a whole variety of topics has been studied. Finally, in Section 4, I will recall a list of
trends in maintenance that was presented by Geraerds (1991) in his last lecture. What surprised
me when I saw this list, is that many of these trends are still actual! This means that either the
area of maintenance progressed only slowly in the past 20 years, or quite some of these trends
are simply long-term trends, or both. I will recall the trends, and I will reflect on what happened
in the meantime and what the current trends are.
68
2. The EUT maintenance model
The EUT maintenance model is a descriptive model that describes subfunctions (=
subprocesses) within maintenance and their interrelations, seen from the perspective of
industrial engineering and from the perspective of an organization that uses and maintains
technical systems. The intended use of the model is twofold:
• to classify scientific research on single subfuntions and of interrelated subfunctions;
• to systematically identify subfunctions in a practical case for which significant
improvements are possible.
The model is depicted in Figure 1 and contains the following subfunctions:
1) the technical systems to be maintained;
2) the internal resources (= capacities) to execute maintenance;
3) the external resources offered in the market by a diversity of contractors
4) the external resources offered specifically by the Original Equipment Manufacturers;
5) maintenance planning and control;
6) the inventory control of nonrepairable maintenance parts (consumables);
7) the maintenance planning and control of exchangeable/repairable spare parts
(repairables);
8) the evaluation of results;
9) the terotechnical feedback;
10) the methodology of design of a technical system;
11) the specification of requirements for a technical system;
12) the design of a technical system;
69
13) the manufacture of a technical system;
14) the design of the maintenance concept for a technical system.
The names of all subfunctions are more or less self-explaining. Further explanations are given in
Geraerds (1992). Some additional remarks:
• The word ‘terotechnical’ in function “9) the terotechnical feedback” comes from tero-
technology, a name invented by a British national maintenance center to get more
attention for the area of maintenance and in particular for the coherence with design
(see Geraerds, 1991).
• The maintenance model is generally applicable. It applies to production equipment,
transportation means, buildings, roads, and so on.
• As stated above, the model is formulated from the perspective of maintenance as a
function in an organization. Hence, decisions about acquiring new technical systems
and disposing of old systems are outside the scope of the model.
• In Figure 1, a dotted line divides all subfunctions in two classes. The subfunctions left
of the line are mainly taken care of by mechanical engineering, electrical engineering,
and so on. The subfunctions right of the dotted are mainly taken care of by industrial
engineering and management science.
Generally, an OEM produces technical systems for multiple users, in which the same
systems are used and maintained by multiple different organizations. The maintenance
model is formulated from the perspective of an organization that uses and maintains its own
technical systems. However, the same model can also be applied to other organizations that
maintain systems, such as maintenance contractors and Original Equipment Manufacturers
themselves.
3. Positioning of PhD projects
In the last thirty years, within Industrial Engineering at Eindhoven University of Technology,
eleven PhD projects with the core of their research in the area of maintenance were successfully
completed. The names of the PhD students, the topics of their projects, and the main
subfunctions of the EUT maintenance model to which they contributed, are as follows:
70
71
• Gits, 1984 (Promotors: Geraerds and Monhemius): Developed a framework for the
design of maintenance concepts. Contributed to subfunction 14.
• Geurts, 1986 (Promotors: Geraerds and Monhemius): Built a bridge between Gits’
framework and the kind of data on failure modes and failure times as encountered in the
records of maintenance organizations. Contributed to subfunction 14.
• Martin, 1994 (Promotors: Geraerds and Bemelmans, Co-promotor: Van Eijnatten):
Studied requirements for management information systems for the maintenance
function within an organization that uses and maintains technical systems. This project
covered all subfunctions on the right of the dotted line in Figure 1 (i.e., the subfunctions
1-9 and 14).
• De Haas, 1995 (Promotors: Bertrand and Christer, Co-promotor: Gits): Studied the
initial stock decision for repairables and the effect of working in overtime by a repair
shop. Contributed to the subfunctions 2 and 7 and their interaction.
• Verrijdt, 1997 (Promotors: De Kok and Theeuwes): Studied the effect of repair and
supply flexibilities in spare parts networks by which multiple users of technical systems
are served. Contributed to the subfunctions 2, 6 and 7.
• Rustenburg, 2000 (Promotors: Zijm and De Kok, Co-promotor: Van Houtum):
Developed multi-item spare parts models for the initial supply and resupply of spare
parts under budget constraints as observed in military organizations. Contributed to the
subfunctions 6 and 7.
• Keizers, 2000 (Promotors: Bertrand and Wessels): Developed a hierarchical framework
for the maintenance control function and decision support models for the external repair
resources. Contributed to the subfunctions 2, 3, 4, 5 and 7.
• Kranenburg, 2006 (Promotor: De Kok, Co-promotor: Van Houtum): Developed multi-
item inventory control models for spare parts networks serving multiple users of
technical systems. Contributed to the subfunctions 6 and 7.
• Vliegen, 2009 (Promotors: Van Houtum and De Kok): Studied the integrated planning
for spare parts and service tools in service networks. Contributed to the subfunctions 6
and 7.
• Öner, 2010 (Promotors: Van Houtum and De Kok): Studied the effect of component
reliability levels and building in redundancy in the design phase on total costs during
the whole life cycle of a technical system. Contributed to subfunctions 11 and 12.
• Büyükkaramikli, 2011, almost completed (Promotors: Bertrand and De Kok, Co-
promotor: Van Ooijen): Studied periodic capacity flexibility in maintenance
departments. Contributed to the subfunctions 2, 3, 5 and 7.
72
As we can observe from this overview, many of the subfunctions of the EUT maintenance
model have been studied and also interactions between them. The PhD projects in which Will
Bertrand has been involved, deal with maintenance capacity, planning concepts, and capacity
flexibility, which are the topics that have the largest overlap with production control.
Current PhD projects are focused on condition-based maintenance (Van Oosterom, Zhu;
subfunction 14), the whole loop of repairable parts (Arts; subfunctions 2, 3, and 7), control in
spare parts networks (Van Wijk, Tiemessen; subfunctions 6 and 7), sharing of spare parts by
multiple companies (Karsten; subfunctions 6 and 7), and the decision support systems for spare
parts planners (Driessen; subfunctions 6 and 7).
4. Trends
In this last section, I want to go back to the list of trends in maintenance that was presented by
Geraerds in his last lecture in 1991:
1. The reliability of components will increase considerably. Statistical analysis of failure
intervals will become less effective and will decrease.
2. Periodic, usage based maintenance will considerably decrease, while condition-based
maintenance will start to dominate.
3. One will more and more apply monitoring of the degradation state of components via
“process control”.
4. Together with the automation of condition-based maintenance, one will see a
centralization of inspections and analysis of the measurements (remote monitoring and
remote diagnosis).
5. The pressure on short downtimes after failures will lead to more modular systems,
where simple “go - no go” tests for functional components go hand in hand with quick
and simple replacements. Repair of failed components, if economically applicable, will
be carried out by specialized companies.
6. Both as the result of the decrease in failure rates and the requirement to react quickly
after a failure, periodic preventive maintenance will constitute a minor part of the
maintenance workload, and consequently maintenance planning and control will start to
look like the way we deal with fulfilling demand for taxi’s.
7. Because of the increasing focus on the core business by organizations, more and more
of the maintenance tasks will be outsourced, while at the same time the maintenance
services as an industry will develop co-makerships.
73
8. Especially in situations where calamities occur, e.g. because of large energy volumes,
chemical and nuclear processes, failure analysis for the sake of the prevention of
failures will start to use statistics of near-accidents.
We are now 20 years further in time, and we may conclude that Geraerds’ vision was excellent!
All above trends have been realized and/or are still going on. This is quite remarkable, and that
is why I wanted to recall this list.
Let me now further elaborate on the above list. I think that there is a certain ordering in the
above trends. We can distinguish ‘root causes’, and the other trends may be seen as
developments in reaction to the root causes. The following factors constitute the root causes:
• Technical systems have become more and more complex. Each new family of a given
type of system contains more functions and many more components than the previous
family. This is a natural process that will continue for many more years. The reliability
at machine level stays constant or is even decreasing, because this is required by the
users. Obviously, then the reliability per component has to increase considerably (trend
1).
• Due to the increased complexity of technical systems, more specialization is required
among the maintenance engineers. It is hard and not desired for individual users to
have their own specialists on the payroll. Further, being a specialist requires that you
practice your specialism with a sufficiently high frequency. This forms a driver for
outsourcing of specialized maintenance to parties that have a sufficient large scale for
this work (this relates to trend 7).
• Many companies have increased their focus on their core business and they made their
core processes leaner. As a result, these core processes have become more and more
dependent on the availability of the technical systems. These companies, being the
users of technical systems require high system availabilities and that implies that
corrective maintenance has to be organized such that failures are repaired within very
short times (trend 6). Especially the availability of spare parts and maintenance (or
service) engineers play an important role in this process.
• Together with a stronger focus on core business, companies also embraced modern
accounting methods such as Activity Based Costing and they have become better
aware of the costs of technical equipment. They know that purchasing price of
technical equipment is high, but they also know that you have high maintenance and
downtime costs during the long periods that the equipment is being used. This has led
to a focus on the Total Cost of Ownership (TCO). I.e., at the moment that new
74
equipment is being bought, a user is already interested in the total costs during the
whole life of the equipment.
The above factors are all drivers for the outsourcing of maintenance to the Original Equipment
Manufacturer (OEM) or a third maintenance party (trend 7). This holds especially for
specialized maintenance work and for the maintenance in the first years of new equipment.
There are multiple factors that make this outsourcing attractive for both the users of the systems
and the OEM’s and third parties:
- Pooling of maintenance resources: A party that does the maintenance for many
systems needs relatively less engineers and spare parts.
- Pooling of data: With an increasing reliability of components, one needs failure data of
more systems in order to apply statistical analysis. The OEM or a third party may have
enough data, while individual users often do not have enough data (see also the
problem raised under trend 1).
- Remote monitoring and diagnosis: When maintaining enough systems, it becomes
more attractive to invest in remote monitoring and diagnosis techniques (trends 3 and
4). These techniques have large fixed costs and it is only attractive to invest in it when
one has a sufficiently high number of systems to maintain. Technically, it is nowadays
very well possible to apply remote monitoring and diagnosis. Many technical systems
are connected to communication networks, and more and more users accept that it is
necessary that the maintainer can follow the status of systems.
In the case of outsourcing to the OEM, there are the following additional factors;
- Terotechnical feedback (holds for an OEM): By becoming responsible for the
maintenance for the systems produced by an OEM, the OEM can create feedback to
the designers. This ‘terotechnical feedback’ is often missing if all users do their own
maintenance. With the feedback, one gets a better design for new systems.
- System availability and TCO focus of designers: Traditionally, design departments
look at the cost price and the reliability when new systems are being designed. By
becoming responsible for the maintenance and the realization of high system
availabilities, they automatically will start to look at design alternatives that increase
availability and decrease TCO. This concerns e.g. incorporating more reliable parts,
building in redundancy, a stronger modular design (trend 5), and incorporating better
diagnosis tools.
75
In the discussion above, all trends come back except trends 2 and 8. Although I cannot support
it by numbers, I know from discussions with the industry that pure usage-based preventive
maintenance is applied as less as needed (cf. trend 2). It is generally seen as unattractive to loose
a part of the useful lifetime. Trend 8 concerns safety management, a topic that has become so
important that it forms a separate research field.
Let us now look ahead and predict what future big developments may be. Most likely, the root
causes that were listed above, will continue for many more years. The other trends which are
reactions to them, will also continue for many more years. Here, it is likely that the OEM gets
more maintenance service contracts for the first years of the lifetime of technical equipment and
third parties may take over after that. In industries where OEM’s close full maintenance service
contracts for all machines they produce, one will go even one step further and sell only a
function (with availability guarantees!) instead of a machine. This causes a delay into cash
inflow at the OEM and requires that financial parties step in to fill this cash gap. Another big
trend is that more and more advanced systems are co-designed with first-tier suppliers, and
maintenance has to be provided by a whole consortium of players (see e.g. the maintenance
concept for the Joint Strike Fighter). It may also happen that first-tier suppliers take a leading
role and try to pass OEM’s. Maintenance will be a challenging and exciting area for many more
years!
References
• Bertrand, J.W.M. (1989). Productiebeheersing; Groei naar volwassenheid. Intreerede,
Eindhoven University of Technology.
• Bertrand, J.W.M., Geurts, J.H.J., & Monhemius, W. (eds.) (1991). Onderhoud en
logistiek; Op weg naar integrale beheersing. Samsom/Nive, Alphen aan den Rijn.
• De Haas, H.F.M. (1995). The coordination of initial stock and flexible manpower in
repairable item systems. PhD thesis, Eindhoven University of Technology.
• Geraerds, W.J.M. (1970). Towards a theory of maintenance. Presentation at NATO-
Conference, Luxembourg. Also published in: Bertrand, J.W.M., Geurts, J.H.J., &
Monhemius, W. (eds.) (1991). Onderhoud en logistiek; Op weg naar integrale
beheersing. Samsom/Nive, Alphen aan den Rijn.
76
• Geraerds, W.J.M. (1973). Productiebeheersing en bedrijfskunde naar samenhang.
Intreerede, Eindhoven University of Technology.
• Geraerds, W.J.M. (1988). Onderhoud vanuit bedrijfskundig perspectief. Diesrede,
Eindhoven University of Technology.
• Geraerds, W.J.M. (1991). Onderhoud in vogelvlucht. Afscheidscollege, Eindhoven
University of Technology.
• Geraerds, W.J.M. (1992). The EUT maintenance model. European Journal of
Operational Research, 24, 209-216.
• Geurts, J.H.J. (1986). On the selection of elementary maintenance rules; With special
references on the estimation of the survival function from censored data. PhD thesis,
Eindhoven University of Technology.
• Gits, C.W. (1984). On the maintenance concept for a technical system; A framework for
design. PhD thesis, Eindhoven University of Technology.
• Keizers, J.M. (2000). Subcontracting as a capacity management tool in multi-project
repair shops. PhD thesis, Eindhoven University of Technology.
• Kranenburg, A.A. (2006). Spare parts inventory control under system availability
constraints. PhD thesis, Eindhoven University of Technology.
• Martin, H.H. (1994). On the determination of functional requirements in a maintenance
environment. PhD thesis, Eindhoven University of Technology.
• Öner, K.B. (2010). Optimal reliability and upgrading decisions for capital goods. PhD
thesis, Eindhoven University of Technology.
• Rustenburg, W.D. (2000). A system approach to budget-constrained spare parts
management. PhD thesis, Eindhoven University of Technology.
• Van Houtum, G.J. (2010). Maintenance of Capital Goods. Inaugural lecture, Eindhoven
University of Technology.
77
• Verrijdt, J.H.C.M. (1997). Design and control of service part distribution systems. PhD
thesis, Eindhoven University of Technology.
• Vliegen, I.M.H. (2009). Integrated planning for service tools and spare parts for capital
goods. PhD thesis, Eindhoven University of Technology.
Note: The lectures of Bertrand (1989), Geraerds (1973, 1988, 1991) and Van Houtum (2010)
are available at:
http://w3.tue.nl/nl/diensten/bib/digibib/publicaties_tue/redes_tue/
All PhD theses are available at:
http://w3.tue.nl/nl/diensten/bib/digibib/publicaties_tue/proefschriften_tue/
78
79
A TYPICAL AND HIERARCHICAL
WORKLOAD-ORIENTED APPROACH IN
EDUCATION
Remarks on the Dutch System for Vocational and Scientific Education
Corné W.G.M. Dirne
Abstract In this paper a comparison is made between a School for Vocational Education and Training, a
University of Applied Sciences, and a Research University, using some of the typical concepts
studied by Will Bertrand in his long research history. The paper doesn't claim to have a
scientific base, but it's the result of some reflections on personal experiences in the Dutch
system for tertiary education. The concepts used are the BWW-Typology for production
situations, the idea of hierarchical planning and control, and the use of workload control in
order to achieve required delivery performances. Education is a serious subject. Read this
paper with a wink of the eye.
1 Introduction
As we all know: doing research in the area of production control with Will Bertrand as coach
and writing a PhD-thesis with Will Bertrand as promoter requires positioning of the research
context in the well-known typology for production situations of "BWW", the application of the
concepts of hierarchical production planning and control, and consideration of the effects of
workload control on delivery performance and capacity planning (e.g. (Bertrand et al., 1990)
and (Bertrand et al., 1998)). This was true in the previous century (e.g. (Dirne, 1990), (Fransoo,
1993), (Ooijen, 1996)), but still is true in this century (e.g. (Nyen, 2005), (Mincsovics, 2009)).
In this article I will contemplate on my experience in the Dutch system of vocational and
scientific education. This contemplation clearly will be based on Will's indoctrination on the
subjects mentioned above, but will also violate all the basic rules Will has taught me on doing
scientific research and on writing reports and articles. To start with the latter: I will not write in
80
the first-person plural ("we"). And on top of that: my considerations are personal, based on my
experience, validated on only three case studies (n=3), and clearly not proven using
mathematical models or simulation studies. Whether or not this paper could be the start of a
reflective cycle as opposed to merely a case-study (Aken, 2004), I'm not sure. What I do know
is that the paper is a result of my reflective cycle. And hopefully, and probably, this cycle is not
round yet.
In this paper I will discuss observations made during my career in education in the Dutch
context. Because of the "logic" of the steps in this career, I will start with discussing the
complexity, the planning hierarchy and the workload control of a University of Applied
Sciences (also known as a school for Higher Vocational Education, or "HBO"). Then I will
proceed with discussing similar observations in the area of a school for Vocational Education
and Training ("MBO"). And finally I will end with a short discussion on the educational context
of a Research University (... because that's where it all started!).
2 The application of Production Planning and Control concepts on
Educational Systems.
The hypothesis in this paper to be checked is that both the BWW-Typology to describe the
complexity of a situation and the paradigm of hierarchical planning and control and workload
control, help to understand the similarities and differences of the different educational systems I
have been working in so far. Clearly, Will's insights have survived the turn of the century. But
will they survive this paper as well? Let's see. In the first subsection I will discuss the
application of the BWW-Typology. Than in the second subsection I will study in some more
detail the concepts of hierarchical planning and control, and workload control, in order identify
the main issues to be discussed in the next sections.
2.1 BWW-Typology The BWW-Typology, as described in (Bertrand et al., 1998) and taught everywhere I teach, is
based on two dimensions of complexity, i.e. material complexity and capacity complexity (see
Figure 2.1).
As indicated above, I will discuss the complexity of the educational situation of the Dutch
school systems following the basic characteristics of this typology. Obviously, applying the
concept of material complexity on an educational system requires the description of the students
considered. The complexity of students is not so much their physical structure (although some
81
Capacity
Complexity
High Low-volume Component
Manufacturing Shop ("job shop") Project Shop
Low Process Industry Shop Mass Assembly Shop
Low High
Material Complexity
Figure 2.1: BWW-Typology of Production Units (Bertrand and Wortmann, 1992).
The capacity complexity in an educational system is determined by the way teaching tasks can
be distributed over the teaching staff. This is largely determined by the flexibility of the teaching
staff, both in the timing of their work and in their level of multi-tasking. Also, the specificity of
the routing a student has to follow to complete his or her programme and the number of times
each lecturer will appear somewhere along that path, determines the complexity of the situation.
2.2 Hierarchy and Workload Control Typical concepts studied in most research in which Will Bertrand was involved, are the
aggregation and decomposition of the planning and control problem in different production
settings, and the workload oriented release of orders to lower levels of control in order to create
"solvable" problems on each level (e.g. (Bertrand, 1983), (Bertrand et al, 1990), (Fransoo,
1993), (Ooijen, 1996)). In other words, in hierarchical systems planning is based on
decomposition of the problem into "solvable" problems, using aggregation at different levels,
and hierarchical co-ordination. As Stadtler noted "... like decomposition, aggregation serves to
reduce problem complexity. It also can diminish uncertainty ..." (p.25, (Stadtler and Kilger
(2000)). One important rule may be obtained from most of these studies: if in a particular shop:
- flexibility is low, both in multi-tasking and in the timing of the availability of capacity,
- required operation times are uncertain,
- routings are complex,
- due dates for some jobs are tight,
then the release of work to this shop should be based on a careful control of the WIP-level,
especially of the bottleneck of the shop. Utilization rates caused by the "tight jobs" should be
kept low.
Increasing the flexibility and reducing the complexity and uncertainty will make it possible to
increase the WIP-level and control the WIP in a more aggregate manner. In other words, small
82
teams with a diversity of tasks require flexibility and enough decision freedom to cope with the
capacity (and "material") problems that may appear.
3 University of Applied Sciences
Working as a lecturer at a University of Applied Science (a "HBO-University" if you like) has
some similarities with working at a Research University, but also is different in many ways. In
order to explain this, I will use the setting of the team I'm working in nowadays as a case study.
This team is responsible for the programme of Industrial Engineering and Management at
Avans University of Applied Sciences in Tilburg. The hypothesis to be checked is that both the
BWW-Typology to describe the complexity of a situation and the paradigm of hierarchical
planning and control and workload control help to understand the similarities and differences of
the different educational systems I have been working in so far. Let's see.
3.1 Applying the BWW-Typology to a HBO-University In this section I will discuss the complexity of the educational situation of the team in the case
study. In order to do so, I will apply the BWW-Typology in a "slightly" adapted way: I will use
the dimensions of material complexity and capacity complexity as in the original typology, but I
will use different characteristics to describe these complexities (as discussed in subsection 2.1).
3.1.1 Material Complexity
The diversity of the "raw materials" of an HBO-University is quite large. Students entering the
university may have a background in Senior General Secondary Education ("HAVO", see
Figure 3.1). For these students this type of higher education is the most logical next step
towards a high professional degree. The usual age for these students entering the HBO-
programme is seventeen. Some of these students have an IQ high enough for a full Research
University, but lack the discipline for self-guided studying or abstract thinking without clear
direct practical implications. Others may have a slightly lower IQ and usually find abstract
conceptual reasoning difficult. In many cases these students are still quite young in their
behaviour.
Nowadays, a large part of the students entering the HBO-programmes do not have a HAVO-
background, but an MBO-background ("Senior Vocational Education and Training"). A
percentage of almost 40% is not uncommon (see Figure 3.2). Usually these students have a 4-
year programme as background ("MBO-4"). Students with an MBO-background tend to be
more focussed and motivated than the students with a HAVO-background, probably due to age,
83
experience (in practical and professional settings) and financial drive. However, confronted with
complex problems to solve, the tendency of these students is to skip the analysis-phase and
implement immediately a "solution". Languages (Dutch and English) and mathematical skills
may be a problem for these students, although that largely depends on the MBO-school they're
coming from. To give a bit of an extreme but real example: for a few students dividing a
fraction by another fraction, may be too difficult if a calculator is not allowed. An interesting
phenomenon, new for me as a lecturer, is the fact that some of these students tend to give up any
effort for understanding a new theory or concept if they don't succeed in doing so the first time.
And they do accept formulas and tools without a thorough explanation of the background, as
long as they understand how to apply them and professionals indicate that these tools indeed are
used in practical situations..
Finally, a small amount of the first year students have a VWO-background ("University
Preparatory Education", see Figures 3.1 and 3.2). These students prefer an HBO-programme
over a Research University because of the practical professional focus, the lack of discipline for
self-guided studying, or the expected effort required at a Research University (or any
combination). Students with a VWO-background usually do not accept tools and formulas
without any discussion on background or derivation. And they easily tend to get bored if the
pace is slowed down.
3.1.2 Capacity Complexity
Usually the team of lecturers operating in an HBO-programme is not very large, especially if the
programme is in some area of technology. Often teams of eight to twelve lecturers have to cover
the entire programme. As a result, the routing of students during their four years of studying
contains many loops, i.e. they get the same lecturer over and over again. Also, it is important for
the lecturers to get to know the students on a personal basis. Attendance of workshops and
classes often is obligatory and often all lecturers are also tutor of several students. Therefore, it
is almost impossible for a student to "hide" in general or even to avoid frequent contact with a
particular lecturer.
Lecturers are required to increase their skills and broaden their knowledge in order to be able to
perform all of their tasks.
84
Figure 3.1: Diagram of the Dutch Education System
(http://www.uu.nl/university/international-students/EN/whyutrechtuniversity/education/dutcheducationalsystem/Pages/default.
aspx).
Figure 3.2: Division first years HBO-students 2007 over background (Takkenberg and Kapel,
2008).
85
As in production, different tools are used in teaching in the process of producing the output
required. The idea that one type of tool is sufficient for the entire population of students, has
long been superseded. This is especially true for the divers input of the HBO-process. Books,
lecture notes and exercise books (such as (Bertrand et al, 1998) and (Durlinger and Dirne,
1992)) may be adequate for some students, but it always fascinates me to see how concentrated
and quiet other students suddenly get when I start a video. Even if it's just a boring simple video
of a lecturer before a white board! And I'm even more amazed by the change in the students
attitude towards learning when they have to do something in a real life situation. Sometimes it's
like looking at Dr. Jekyll and Mr. Hyde. Therefore, confronting students with real life situations
from the beginning of their study, is important for their motivation and the effectiveness of the
programme. Due to the number of practical settings required, the fact that the level of
complexity of these settings shouldn't be too high for students in the first two years, and due to
the fact that -compared to graduates of a master of science programme- a large percentage of
graduates of a bachelor of engineering programme tend to start their career in SMO's ("Small
and Medium sized Organisations"), most companies providing these practical settings are SMO-
companies.
Preparing these kind of tools and learning environments requires time, creativity, networking
and organisational talent.
3.1.3 Conclusion
The material complexity at an HBO-University is average to reasonably high especially due to
the differences in student backgrounds. The capacity complexity could be characterized as
average to reasonably high as well, not only because of the loops in the routing of a student
passing lecturers, but especially because of the frequent involvement of real life situations
where each situation is likely to have at least some particular characteristics.
3.2 Hierarchy and Workload Control at an HBO-University Clearly, the kind of tasks to be performed by lecturers in a HBO-setting is comparable to the
teaching tasks at a Research University. The main differences from a planning and control point
of view are:
- The part of the utilization rate due to teaching tasks, is much higher. As a result usually the
part of working hours that is blocked due to fixed time schedules, is much higher.
- The amount of time required to answer questions of individual students or groups of students
at unscheduled moments, usually is higher as well. Making reservations in the time tables for
these moments of contact is a good way of increasing the predictability of the workload, but
86
doesn't reduce the number of hours required and clearly will increase the blocked time slots
even further.
- It is a known fact that on days where no classes are planned, students are likely not to study
at all, but, for instance, tend to earn extra money on part time jobs. Therefore, in time tables
usually classes are spread throughout the week instead of being concentrated on particular
parts of the week. Probably the only exception to this observation is the Friday, but that
exception is not caused by planning and control considerations, but by the grey-coloured hair
of many lecturers.
In order for a lecturer to organise and schedule all the other tasks he or she has to perform, some
degree of flexibility is required. Some possibilities to achieve this flexibility are:
- for each lecturer, at least one day in the week is left unscheduled;
- the time horizon for fixing time tables shouldn't be any larger than strictly necessary, thus
creating the possibility of rolling horizon planning and leaving more flexibility for solving
future planning problems;
- creating multi-tasking possibilities by educating and improving the skills of the lecturer
instead of increasing the tendency towards further specialization;
- individual lecturers are given some freedom in their classes and exams in order to be able to
adapt to the student population at hand and the actual situation, as long as they still achieve
the final attainment level of their course (this requires for these final attainment levels not be
too detailed).
Fortunately, a team does have influence on the accents and layout of the programme. The final
achievement levels that are determined on a national level, are aggregate enough to allow for
local accents and design. In such a setting, creativity is stimulated and teams are encouraged to
adopt to local and changing circumstances.
From a workload planning and control point of view, at the department-level the aggregate
workload and the workload of the bottleneck resources can be planned and controlled. On the
team-level the workload of each individual lecturer could be controlled. Clearly, the exact time
schedule for each day is for the individual lecturer to be controlled by use of a more or less
advanced personal diary. Whether indeed these possibilities for planning and control are used, is
very dependent on the organisational setting of the team. To be more precise: on the choices
made by management.
87
4 School for Vocational Education and Training
Before I started as a lecturer at Avans University of Applied Sciences, I was managing director
at a School for Vocational Education and Training ("MBO"). To be more precise: of the unit
"Business and Organisation" of this school. It was my first position outside a university. Note
that about 46% of all students that continue their education after secondary school, follow a
study programme at an MBO-school (as opposed to 34% HBO and 20% University (CBS,
2010)). The observations made in this section are based on one case study, combined with some
overall statistics.
4.1 Applying the BWW-Typology to a MBO-School Again, using the BWW-Typology in an adapted manner, I will discuss the complexity of the
educational situation of a School of Vocational Education and Training (short: MBO-School).
First by discussing the material complexity of the situation (the students). Then by discussing
the capacity complexity (the teachers and organisational structure).
4.1.1 Material Complexity
Before I will get into more details on the material complexity, let me first describe two
incidents.
One of the first phone calls I received as a managing director was from a parent telling me that
her son would be away for two months. He was "out on a visit for two months", she told me. I
didn't get it at first until she said to me a bit annoyed: "Well, you know, he's got an address the
next two months where he can't leave ...". Then I got it.
In September the teams in my unit organised an evening for the parents of the new first year
students. As a managing director I also gave a presentation that night, explaining the
programmes and educational organisation of our unit. I gave some examples of the subjects that
their sons and daughters would be getting, like on administration and commerce. Afterwards the
parents were sent away with the mentor of the class of their son or daughter. One couple stayed
behind. They told me they didn't know which class their son was in. "No problem", I told them.
It's a new school and all of that, and clearly we would be able to find the right class by looking
into the school administration. Unfortunately, we couldn't find his name on any of our lists. The
parents were puzzled, and so were we. Then the mother took her cell phone and called her son.
She was told by him that he didn't study on a MBO-school yet, but still was studying on the
VMBO-school (see Figure 3.1) further on in the street. And that he didn't study business or
organisation or something, but electrical engineering.
88
Need I say more? Maybe I do.
In order to understand the material complexity of a MBO-school more thoroughly, I will
subdivide the student population into four categories (see Figure 4.1).
Level 1-2 BOL 1/2 BBL 1/2
3-4 BOL 3/4 BBL 3/4
BOL BBL
Type of programme
Figure 4.1: Four categories of MBO-students.
The levels indicate the difficulty of the programme (and also indicate the number of years
required to complete the programme). So a level 4 programme requires 4 year and is the highest
level a student may achieve on a MBO-school. This level may give access to a HBO-
programme, especially if the HBO-programme is in the same area as the MBO-programme. A
level 1 programme requires only 1 year and is the lowest level a student may achieve.
Unfortunately, this level is too low to be regarded as a official professional degree. Usually,
most programmes are offered in two types. "BOL" is the regular type of programme where
students go to school five days a week (apart from their periods of practical experience, of
course). In a "BBL"-type of programme a similar diploma may be achieved by a totally different
route, i.e. the students works for -in most cases- four days a week and comes to school only one
day a week. In this programme the student has to perform all kind of prescribed tasks and
assignments "in-company".
In many cases, level 1 students never successfully finished any school at all. They may have no
VMBO-diploma. Often their social circumstances are quite difficult. Positive feedback is
unknown to many of them. Success is not a part of their vocabulary. Setting up a Christmas-tree
with two or three students together on the agreed day could be quite an achievement. Coming to
school on 75% of the days required often is an achievement as well. Anger and disappointment
often results into aggression, partly because their language skills are low or because they don't
know how else to express themselves.
In the same school, although not necessarily the same building, students study to successfully
complete their level 4 programme. They study to become a reliable bookkeeper, a motivated
teaching assistant, an expert welder, a highly skilled mechanic.
89
One aspect that should be mentioned to understand the complexity even further, is the number
of immigrant students (see Table 4.1). Many of these students are highly integrated and have a
stable social background. But for others, that's not the case. Their parents don't speak Dutch.
The social environment and norms and values at home are totally different from those at school.
And both of these social settings are totally different from the street settings where they spend a
lot of their time. Obviously, the kind of attention, the pedagogical approach, and the level of
education on, for instance, the Dutch language, should be totally different for these students
than for students with a more stable background.
4.1.2 Capacity complexity
Understanding the capacity complexity of an MBO-school requires understanding of the
national structure of this part of the Dutch educational system. The final achievement levels for
each study programme are not determined on a local level, but on a national level. For each
category of
Table 4.1: Percentage immigrations for each MBO-level (source: CBS-website; 10 June 2010)
level total immigrants % immigrants
1 23392 10281 44%
2 130259 40303 31%
3 141970 34730 24%
4 226057 52484 23%
total: 521678 137798 26%
professions one of the seventeen national Centres of Expertise (Dutch: "Kenniscentra") is
responsible for setting up, monitoring, and adapting the final achievement levels, described in
the so-called qualification files (Colo, 2010).
The procedure for designing or changing these levels are quite complex and many parties are
involved (including usually the larger employers). Before the introduction of the competences-
based qualification files in this century, the final achievement levels were described in much
more details. For some study programmes lists of hundreds of items to be achieved were not
uncommon. The monitoring and control of all the study programmes is quite intense. At a
90
particular moment in time, around 2004-2005, both the Dutch Inspectorate of Education and a
special Centre of Expertise on Examinations would check the programmes, requiring many
documents and data to be collected on a regular base (every year or every other year).
Because of the structure being so complex and of the control being so tight, the freedom for
teachers to create new lessons and teaching material was (and sometimes still is) limited. The
best way to "keep out of trouble" was to make use of ready-for-use material provided by
publishers of educational books. And to use the ready-for-use exams that could be bought from
the Centres of Expertise and were guaranteed approved by the Centre of Expertise on
Examinations. As a result creativity among the teaching staff was reduced to practically zero
(except for some highly motivated Don Quixotes who believed in fighting windmills).
Then the new competences-based structure was introduced, requiring real-life situations to be
included in the programmes as much as possible and stimulating the setting up and organisation
of complex proficiency tests on a local base. The introduction of this new system was co-
ordinated nationally, in a project-oriented manner, and every school and team was to introduce
the new system in a couple of years. Although the logic behind the system was and still is not
bad at all, especially for vocational education and training, both from the point of view of
change management and from the wish for diversity in study programmes to accommodate the
diversity in students, a worse scenario could not have been followed. Suddenly teachers were
required to perform tasks they didn't agree with nor were prepared for, and they risked of being
confronted with all kinds of problems afterwards when inspected by the Dutch Inspectorate of
Education or the Centre of Expertise on Examinations.
One of the consequences of all these directives and changes was that teachers more and more
were checking, sometimes even on a daily base, whether their workload still corresponded with
the official collective agreement ("CAO"). And the number of rules that could be violated when
creating a time schedule for the next period, was quite large. Making a new schedule was like
solving a mathematical optimisation problem with many constraints, with no guarantee of
finding a viable solution.
4.1.3 Conclusion
Clearly, the material complexity at an MBO-School can be considered high to very high. Due to
the special attention each student may require, especially in the lower levels, sometimes it may
seem as if no standard programme could be applied. The diversity in learning competences and
skills is high, while the moment and amount of time required from a teacher to help a student is
rather unpredictable. On top of that, due to a lack of flexibility, caused by the reasons mentioned
91
above, the capacity complexity is high as well as is the specificity of the programme to be
followed ideally.
4.2 Hierarchy and Workload Control at an MBO-School As compared to the time schedules of lecturers in a University of Applied Sciences, the time
schedules of teachers in an MBO-school are even much more fixed. On average, the number of
hours a full-time teacher will spend giving classes would be about 22 (with a possible maximum
of 25). The number of hours available for preparation is less than for a lecturer at a University
of Applied Sciences, at least by 50%.
As you can imagine, the flexibility of teachers in an MBO-school was and probably still is not
very high. Taking away their books, standard exams and classrooms doesn't make them more
flexible. It makes them more insecure. Forcing level 4 teachers to teach to level 2 students
doesn't improve the quality of a study programme. Generally speaking, level 2 students do not
need teachers focussing on state of the art knowledge and equipment. They need socially en
empathically engaged teachers, speaking their language, able to motivate them to reach a bit
higher than before. Level 2 teachers having to teach more advanced knowledge and skills to
level 4 students do not come home at night with a satisfied feeling of having done a good job.
Demanding more certified teachers doesn't improve this flexibility.
Planning and controlling the workload of the teachers in the teams was a tough job. It required a
detailed spreadsheet-based program where workload was calculated in four digits after the
decimal point. For instance: one teacher in my unit was employed for 0.7969 fte. Not for
0.7970, not for 0.7968, but for 0.7969 fte. And she was not the only one ... The more teachers
were likely to refer to the official collective agreement, the less possibilities were left to plan
workload in a hierarchical way, such as a more aggregate workload control on team level and
the more detailed planning and scheduling in the teams themselves. Asking the teachers to come
school one day before the opening of the school year could be violating the agreements on the
collective holidays. Not to mention the difficulties encountered when teachers got ill and had to
be replaced. In 2008 the percentage absence through illness for teachers in the MBO-sector was
almost 5,1% (Stichting Onderwijsarbeids-marktfonds, 2008).
5 Research University
Things have changed at universities since I've left the TU/e. My ego is not that big that I assume
that these changes have anything to do with my departure. I've had the privilege of being part of
92
that change for some time. My last academic year I worked at the university was the year 2002-
2003. If I would follow the outline of this paper in this section as well, I would have to discuss
the present situation at the TU/e. However, it would be too presumptuous of me to claim to
know enough details of this situation to be able to do so. As I understand, the last couple of
years Will Bertrand even entered the Departmental Board. Things have changed too much!
The only observations I would like to share, since they probably are still true nowadays, are:
- The planning and control situation at a HBO-University resembles more the situation at a
Research University than the situation at a MBO-School.
- Usually the flexibility of timing is larger at a Research University than at a HBO-University,
while the flexibility in the sense of multi-tasking abilities tends to be less.
- The creativity and readiness for changes is much higher at the Research University and
HBO-University than it is at a MBO-School. The need for changes and creativity probably is
higher at HBO-Universities than it is at Research Universities.
- The educational and pedagogical skills and attitudes required from a teacher, increase rapidly
as the "level" of the educational setting decreases. I know that even today at Research
Universities there are a lot of discussions on the quality of classes, study materials and
lecturers. And rightly so. But the necessity of this quality is far less as compared to, for
instance, the MBO-Schools. I know now from experience that students at a Research
University indeed are far more self-guiding than students elsewhere.
6 Reflection
You may wonder whether there is an educational situation in the Netherlands with low material
and low capacity complexity. Well, there was once: the old Master-Mate system where the
master only had one student and could fully determine by himself what programme his mate had
to follow and how much time was required to teach the mate all he had to know. Learning a
craft could be considered quite a process. Change-over time to a new mate usually was very
large. Such a system of patience, dedication and slow growth seems to be difficult to re-create
and sustain in our modern, fast society.
And the "job shop"? Would that be the student shopping at many different schools and school
systems in order to get a degree? Or maybe the part-time student who has completed several
advanced courses and has many years of experience, and now tries to get an official degree?
93
In any case, the old fashioned class-oriented system resembles more the assembly shop. One
teacher for one class during one year, with clear and identical end terms for each student.
Complexity in this system was and is the diversity of the students in the class .
I know. This paper is not really a highbrowed scientific paper. It's not the result of a thorough
research project. Actually, it's not a result of any research at all (apart from some literature
research I've done to write this paper). It's just a reflection on my experiences in three different
type of educational settings in the Netherlands.
The step from a position as a director of education at a university to a managing director
position at a MBO-school was interesting and by all means a challenge. I'm glad I made this
step. I'm glad Will supported me all the way. And motivated me. Our real first encounter was an
oral exam. He asked me a question and, while I was thinking what the question was about and
why I hadn't read that part of the lecture notes he was referring to, all he did was looking out the
window and being silent. It was a quiet and painful exam. It got me motivated.
As Will always told me: in research and education it's not about capacity planning and
monitoring hours or workload; it's all about motivation!
References
Aken, J.E. van (2004). Management Research Based on the Paradigm of the Design Sciences:
The Quest for Field-Tested and Grounded Technological Rules. Journal of Management
Studies, 41:2, March 2004, 219-246.
Bertrand, J.W.M. (1983). The effect of workload dependent due-dates on job shop performance.
Management Science, 29, 799–816.
Bertrand, J.W.M., Wortmann, J.C., Wijngaard, J. (1990). Production Control - a Structural and
Design Oriented Approach. Amsterdam. Elsevier Science Publishers.
Bertrand, J.W.M., Wortmann, J.C. (1992). Information systems for production planning and
control: developments in perspective. Production Planning & Control, Vol.3, No.3, 280-289.
Bertrand, J.W.M., Wortmann, J.C., Wijngaard, J (1998). Productiebeheersing en material
management. Houten, Nederland. Educatieve Partners Nederland BV.
94
CBS (2010). Statistisch Jaarboek 2010. Den Haag, CBS.
Colo (2010). Qualifications Structure. http://www.colo.nl/qualifications-structure.html.
Accessed 29 November 2010.
Dirne, C.W.G.M. (1990). Production Control for Flexible Automated Manufacturing Stations in
Low Volume Component Manufacturing. PhD-Thesis, TU Eindhoven.
Durlinger, P.P.J., Dirne, C.W.G.M. (1992). Produktiebeheersing en material management -
werkboek. Leiden/Antwerpen. Stenfert Kroese Uitgevers.
Fransoo, J.C. (1993). Production Control and Demand Management in Capacitated Flow
Process Industry. PhD-Thesis, TU Eindhoven.
Mincsovics. G.Z. (2009). Studies on Tactical Capacity Planning with Contingent Capacities.
PhD-Thesis. TU Eindhoven.
Nyen, P.L.M. Van (2005). The integrated control of production-inventory systems. PhD-Thesis,
TU Eindhoven.
Ooijen, H.P.G. van (1996). Load-based work-order release and its effectiveness on delivery
performance improvement. PhD-Thesis, TU Eindhoven.
Stadtler, H. (ed.), Kilger, C. (ed.) (2000). Supply Chain Management and Advanced Planning:
Concepts, Models, Software and Case Studies. Berlin. Springer-Verlag.
Stichting Onderwijsarbeidsmarktfonds (2008). Verzuimanalyse Mbo-sector - 3e kartaal 2007
t/m 2e kartaal 2008. Groningen, Arboservicepunt.
Takkenberg, D, Kapel, R. (2008). Van mbo en havo naar hbo. Centraal Bureau voor de
Statistiek, Sociaaleconomische trends, 2e kwartaal 2008, 26-31.
95
THE IMPACT OF NON-COMPLIANCE TO
PRODUCTION CONTROL PRINCIPLES
ON PERFORMANCE OF REAL-LIFE
PRODUCTION SYSTEMS
J.M. Keizers
Abstract Models on production control cannot be built without assumptions. These (mostly) small-scale
models help us to represent real-life situations and to understand the relationships between the
input and output variables in these situation. However, we also know upfront that these small
scale models contain errors, compared to the real-life situations. When implementing the
knowledge obtained from these models in real-life situations, it might seldom be the case that a
performance is being achieved that is being forecasted based on the model. Rather than
assuming that this gap is due to the theoretical assumptions upon which the model is based, this
paper presents – based on three case studies - an approach to make this gap visible. While
models are based and developed on steady state behavior and aggregate performance statistics,
this approach is based on analyzing the individual (sales/production/purchase-) orders. In this
way, additional information on the behavior of the planners is being created, which can be fed
back to the planners, in order to structurally improve performance. This approach can be of
help for companies which keep on investing in new supply chain processes and software, but
continue to suffer from performance which keep on underperforming.
Introduction
The research schools on operations management continue to develop and improve knowledge
on inventory and production planning and scheduling. This knowledge is slowly being absorbed
by companies and implemented in real-life production situations, via various channels: e.g. via
new software applications of ERP software providers, via graduate projects, via former students
and via research projects between research schools and private companies. Despite this growing
96
knowledge and improved algorithms in software packages, companies still struggle to control
and structurally improve the performance of their processes. It is widely known, that part of this
struggle is caused by the difference between the underlying assumptions used for building the
production control models and the real-life situation. However, also human behavior can
increase of decrease the gap between the model and the real-life situation.
Wiers (1996), for instance, studied real-life planning and scheduling systems and concluded that
these are much richer than theoretical models, both in terms of the complexity of the problems
to be solved and the intelligence and control means available for the planners solving the
problems.
It is now widely agreed that just applying the results of theoretical research does not solve the
real-life planning problem. However, human planners and schedulers who are integral parts of
the planning and scheduling functions in a manufacturing system can use the theoretical results
as a guideline for their practice, but they still are fully aware of the complexity of the real-life
problems and the various means available for tackling the problems (McKay et al. 1995a,
1995b). Research by Crawford et al. (1999) has shown that planners in manufacturing systems
have many different roles, ranging from mediator and problem solver, to information channel
and problem escalator. Planners therefore seem to do more than just taking decisions; they seem
to influence the process in various ways, and seem to take into account all kinds of information
that is not represented in the theoretical models that are being used for studying the planning
and control processes
This paper builds further on the work done by Bertrand in the past decades. In his work,
Bertrand tries to decomplexify the production situations under research in hierarchical
dependent models or solutions, where the output of each model is used as an input condition for
the next lower level. Furthermore, Bertrand explicitly pays attention to the robustness of the
solutions, as he assumes that – by definition - real-life situations differ from the assumptions of
the theoretical models. This paper is based on the research of Bertand and Keizers in the late
90’s. In Keizers, Bertrand & Wessel (2003), the gap between actual planning procedures and
their analytical planning model in a maintenance organization is being described by the factors
which are not included in a proper way in the planning procedures. The authors conclude that
planners do not adequately include the information on (uncertainty of) processing times, lead
times, the number of shops involved in the various projects and the mixture between planned
and unplanned jobs in their planning processes. As a result, the performance is below threshold.
This information can be fed back to the planners. However, this feedback information still lacks
the structure and knowledge for the planners on how to adequately use the identified lacking
information in their planning processes. In this paper, we go one step further and describe for
three different cases the actual gap between the analytic models and the real-life situation by
97
analyzing the individual planning decisions, rather than analyzing aggregate statistics on the
data about the planning decisions. Based on this, we show the power of these types of analyzes
and even describe the content of the gaps. These analyses are becoming more and more
possible, because of the growth of the ERP software during the last decades. These ERP-
systems (and their add-ons) are increasingly containing state of the art planning and scheduling
algorithms and storing all kind of data on single purchase, sales and production orders, which
are relevant for the analyses shown in this paper.
In the next sections of this paper, three different cases will be presented. Each case presents a
(part of a) production system which suffers from underperformance, even though production
control models are implemented, which contain at least some analytical knowledge. However,
unless these models, the production systems show a performance which is below the expected
or target values. In each of the three case studies, the performance is not being analyzed by
studying averages or steady-state behavior, but by studying the individual performance of the
relevant objects (e.g. production orders, sales orders, purchase orders) and comparing their
individual performances to the performance as suggested or calculated by the model.
Case 1: Low inventory turns because of non-compliance on Make-to-Order products
Company I produces a commodity product with a large variety in dimension of the end product
(length, width, height, weight). Some of these products are made to stock (MTS), whilst other
products are only made to order (MTO). Although procedures are in place for distinguishing
between MTO and MTS, an unacceptable part of the actual inventory consists of non- and slow-
moving inventory. After studying the procedure for classifying articles in either MTO or MTS,
the adherence of the planners to the MTO procedures is being analyzed, using Figure 1. This
graph shows that - although planners distinguish MTS and MTO production - for 23% of the
production orders the output on MTO production is more than planned. And for 49% of the
production orders, the output on MTO production is less than planned, which requires a new
production order with again a substantial probability of producing too much. In this way,
overproduction on MTO articles leads to unwanted stock with high potential on becoming
obsolete. The gap exists in the fact that although clearly defined procedures exist for assigning
articles either as MTO or MTS, still non-moving stock is being created buy production
supervisors who take the freedom to increase efficiency by increasing the batch sizes. High
level research on aggregate characteristics of the
98
10005000-500-1000
99
95
90
80
7060504030
20
10
5
1
Quantity under/overproduced on production orders
Pe
rce
nt
0
77
49
(Each dot represents a production order)Production output versus production planning
Figure 1. Impact of production-schedule compliance on inventory turns
inventory mix does not give the required knowledge which can be fed back to the production
supervisors, who have a natural preference to run efficient batch sizes. However, this analysis
on individual production order level does tell you what to feed back to your production
supervisors in order to avoid non-moving inventory as much as possible: stick to the plan.
Case 2: Impact of master data quality on delivery performance
Company II develops technical systems and uses a subcontractor for purchasing/ manufacturing
the sub-assemblies and for assembling the final technical system. The delivery performance of
company II to its customers is under pressure and falls below their target of 90% in time. As
Figure 2 shows, this is caused by a delivery performance of their subcontractor which is also
below targets. At first, pressure is put on their subcontractor to make sure they improve their
performance. Performance of the subcontractor slightly increases, but still does not reach the
target performance. Once studying the performance of the individual purchase orders to the
subcontractor and comparing the attributes of the individual purchase orders to their standards –
as described in the material master data of the associated articles – a pattern is being found.
99
50403020100-10-20-30-40
99,9
99
95
90
80706050403020
10
5
1
0,1
Actual delivery date minus requested delivery date
Perc
ent
0
65
(each dot is a purchase order line in 2010)
Figure 2. Delivery performance of suppliers
The company defines 7 different standards for their material master data in their ERP system
which need to be met in order to classify the quality of the master data as acceptable. These
standards range from “valid price in ERP system”, to “purchase lead time in ERP aligned with
contractual agreed lead time” and to “re-order points filled in ERP for Make-to-Stock articles”.
For each article number, the number of errors (i.e. non-satisfying standards) can be counted and
assigned to this article number, which logically can be range from 0 (i.e. all standards are met)
to 7 (i.e. none of the seven standards are met). When studying the delivery performance of the
company’s supplier, whilst distinguishing material numbers based on their number of non-
satisfying standards, the following can be concluded.
Whereas each dot in Figure 3 represents the actual delivery date minus the requested delivery
date, it can be seen in the plot that the delivery performance is a decreasing function of the
number of non-satisfying standards: the more errors in the master data, the worse and the more
unstable the performance of the supplier. Or stated the other way around, in case the quality of
the material data is up to standards, the delivery performance of the suppliers is closest to the
requested performance. Clearly, the purchasing department sends purchase orders to their
suppliers which contains data – or
100
50250-25-50
99,999
90
50
10
10,1
50250-25-50
99,999
90
50
10
10,1
0
Actual delivery date of suppliers minus requested delivery date of suppliers
Pe
rce
nt
01
2 3
0
0123
master dataerrors in
Number of
Panel variable: Number of errors in master data
Each dot is a purchase order line in 2010
77% 66%
58% 61%
Figure 3. Impact of master data quality on supplier’s delivery performance
is triggered based on data - which does not reflect the real-life process and the real-life
capability of the supplier. This case shows that at first glance the production system is non-
capable of reaching the target performance, whilst management is confident that they are using
the right principles for production planning. In fact, they are using the right principles, but the
non-compliance of the master data makes that performance is below standards.
Case 3: Impact of customer’s sales order lot size behavior on delivery performance
Consider Company III. This company delivers customer-specific articles. Some of the articles
are delivery from stock, some of them are made-to-order. The company was being faced with a
poor delivery performance on their make-to-stock items. Like Figure 4 shows, in month 8, the
delivery performance of the company turned out to be only 20%. Further research showed that
the inventory policy was not aligned with the service level agreements that the company has
agreed with their customers. For example, for some items it has been contractually agreed that
these should be delivered from stock, but in real-life it turned out that for most of them the
replenishment levels
101
100806040200
99
95
90
80
7060504030
20
10
5
1
Actual delivery date minus confirmed delivery date
Perc
ent
89
10
Month
(each dot is a delivery line)
Figure 4. Delivery performance of Make-to-Stock items
1000050000-5000
99,9
99
9590
80706050403020
10
5
1
0,1
Ordered quantity minus contracted reorder quantity
Perc
ent
0
Customer ACustomer BCustomer CCustomer D
Customer
(each dot is a sales order)
Figure 5. Adherence of customers to frame contracts
102
where not aligned with the contractually agreed sales order lot size of the customer. After
solving this, the delivery performance started to improve during month 9 and 10. At the start of
month 10, all replenishment levels where aligned with the contractually agreed sales order lot
sizes, and management of the company predicted a delivery performance over 90%.
Although Figure 4 shows a strong improvement of the delivery performance - please note that
the lateness of some sales orders may be caused by non-aligned inventory policy in prior
months – still part of the sales orders showed an unacceptable delivery performance. Rather
than concluding that the production control principles are not sufficient for this real-life
situation, further research of the individual sales order lot sizes showed the solution. Each make-
to-order item has inventory parameters (reorder level and reorder quantity), which are based on
the contractually agreed sales order size (e.g. “Company III ensures to keep 5000 pieces of part
AB1234 on stock”). However, Figure 5 shows that there can be a huge difference between the
contractually agreed sales order size and the actual sales order size.
Customer B and Customer C comply more or less to the agreed lot-size. However, especially
customer D orders in about 65% of the sales orders, a quantity which exceeds the quantity on
which the company has been anticipating. As a result, an additional production run is required
and therefore, the order cannot be delivered from stock.
Conclusion
Models on production control cannot be built without assumptions. These small-scale models
help us to represent real-life situations and to understand the relationships between the input and
output variables in the real-life situation. However, we also know upfront that these small scale
models contain errors, compared to the real-life situations. When implementing the knowledge
obtained from these models in real-life situations, it might seldom be the case that a
performance is being achieved that being forecasted based on the model. Rather than assuming
that this gap is due to the theoretical assumptions upon which the model is being based, we have
presented in this paper – based on three cases - an approach to make this gap visible. This
approach is based – instead of investigating the steady state behavior and aggregate
performance statistics, as are used for building the small scale models – on analyzing the
individual (sales/production/purchase-) orders. In this way, additional information on the
behavior of the planners is being created, which can be fed back to the planners, in order to
structurally improver performance. This approach can be of help to make for companies which
103
keep on investing in new supply chain processes and software, but continue to suffer from
performance which keep on underperforming.
References
Bertrand, J.W.M. and D.R. Muntslag (1993). Production control in engineer-to-order firms.
International Journal of Production Economics, 30: 3-22.
Bertrand, J.W.M. and J.C. Wortmann (1981). Production control and information management
for component manufacturing shops. Elsevier, Amsterdam.
Bertrand, J.W.M., J.C. Wortmann and J. Wijngaard (1990). Production control: a structural and
design oriented approach. Elsevier, Amsterdam.
Crawford, S., B.L. MacCarthy, J.R. Wilson and C. Vernon (1999). Investigating the Work of
Industrial Schedulers through Field Study. Cognition, Technology & Work, 1: 63-77.
Keizers, J.M. (2000). Subcontracting as a Capacity Management Tool in Multi-Project Repair
Shops, Ph.D. Thesis, Technische Universiteit Eindhoven.
Keizers, J.M., J.W.M. Bertrand and J. Wessels (2003). Diagnosing Order Planning Performance
at a Navy Maintenance and Repair Organization using Logistic Regression. Production and
Operations Management, 12(4): 445-463.
McKay, K.N., F.R. Safayeni and J.A. Buzacott (1995a). Schedulers and planners: what and how
can we learn from them. In: D.E. Brown and W.T. Scherer (ed), Intelligent scheduling systems.
Kluwer, Dordrecht, 41-62.
McKay, K.N., F.R. Safayeni and J.A. Buzacott (1995b). 'Common sense' realities of planning
and scheduling in printed circuit board production. International Journal of Production
Research, 33: 1587-1603.
Wiers, V.C.S. (1996). A quantitative field study of the decision behaviour of four shop floor
schedulers. Production Planning and Control, 7: 383-392.
104
105
WHAT AND HOW OF PLANNING AND
CONTROL OF OPERATIONAL PROCESSES
Ton de Kok
1. Introduction
I had heard about Will Bertrand before I joined his group only indirectly. It was clear that he
was a good researcher and knew what the practice of planning and control was about. Only
when we started our discussions on the fundamentals of planning and control, i.e. the nature of
primary (operational) processes and its consequences for effective planning and control, I truly
learned about the depth and breadth of his knowledge. Looking back, I think I only implicitly
made a wise decision: instead of reading all the things he read (in particular the classics from
Holt et al (1960), Forrester (1961), Anthony (1965), Conway et al. (1967), and Galbraith (1973),
to name a few), I just listened to him, absorbed and tried to translate his thoughts into mine,
often trying to develop basic quantitative models that might help further understanding.
Listening is extremely efficient, provided you listen to knowledgeable people. I think Will
shares my vision that listening can also be extremely inefficient when listening to self-indulgent
guru-type people. Both Will and I have had our share of that.
Though Will and I continued our discussions on the fundamentals of planning and control until
now and will continue them as long as we live, we only published one paper together in the
Dutch journal Bedrijfskunde (cf. Bertrand and De Kok (1999)). The topics discussed below may
provide some explanation for this fact. In section 2 what and how of planning and control is
discussed. In the subsequent sections recurring topics during our discussions on the
fundamentals are discussed: commonality in section 3, assumptions about the future in section
4, and modeling time in section 5. The epilogue, section 6, intends to draw some conclusions
and indicate food for thought for others. By no means I claim that out of our discussions we
have distilled the do’s and don’ts of planning and control. But I do claim that we have discussed
matters that most scientists and practitioners never consider explicitly, while these matters
matter for effectiveness of implementation of planning and control frameworks.
106
2. What and how
Arriving at LBS in 1992 and reading the famous blue book (Bertrand et al (1990)) and purple
book (Bertrand et al (1990)), and re-reading them both, I wrote in the upper right-hand corner of
my whiteboard: What! How? For me the “what” in planning and control is the structuring of
decisions functions aligned with the decision authorities in organizations. The “how” in
planning and control is the implementation of these decision functions in information systems
and working procedures. Both the “what” and “how” can be treated at various levels of details,
depending on the objective one has with developing scientific or professional knowledge.
Operations Research, my background, is mostly concerned with the how question. Starting from
a model describing a decision function ( i.e. objective function, constraints, decision variables
with their domains) methods are developed to determine optimal decisions. If that is not
possible due to NP-hardness or the curses of dimensionality, heuristics are proposed showing
adequate decision quality. Late 1970’s Ackoff (1979) concluded that Operations Research as a
science had failed to deliver its 1950’s promises by focusing only on mathematical methods and
mathematically tractable models. He argued that OR should be “re-conceptualized”. When
reading Ackoff’s paper in the late 1980’s while working as an OR expert consultant, I disagreed
with Ackoff: I worked in a very successful group of OR and Statistics specialists proving him
wrong day-by-day. OR worked! And still does, by the way. Yet I did not get the point of
Ackoff’s paper. Though Ackoff was disappointed about OR’s impact on practice, his argument
was a scientific one. Using the wording above, Ackoff made a plea to first discuss the “what” of
decision making, before we discuss the “how” of decision making. I can only support this and
Will Bertrand made me clear why.
Already in the 1970’s Will Bertrand, together with Hans Wortmann and Jacob Wijngaard,
studied planning and control from a conceptual point of view. Keywords were hierarchical
planning, feedback and feed forward, uncertainty.
Influenced by Galbraith (1973) and Hax and Meal (1975) they developed a generic framework
for production planning and control. The decomposition into aggregate planning, material
coordination, workload control and work order release is fundamental. Probably over 1000 MSc
students have used the framework in various industrial sectors, ranging from process industries
to capital goods industries. The Eindhoven framework for production planning and control is the
generically applicable "what". But how about the "how"?
107
Figure. The Eindhoven framework for production planning and control.
The "how" of workload control found its form in various combinations of an upper bound on the
total amount of work (in time units) and machine dispatching rules. Together with Ton van de
Wakker (1993) and Henny van Ooijen (1996) Will Bertrand showed that their simple rules
enable low variability lead times at reasonably high utilization rates for job shops with simple
and complex job structures. Key concepts included differentiation in throughput times and
exchange of slack time between different work orders.
In his work Will Bertrand focused on workload control, work order release and Production Unit
control. For these functions he developed clear guidelines for the "how" of planning and control.
He paid less attention to material coordination, with exception of his work with Zuiderwijk and
Hegge (Bertrand et al. (2000)), which we discuss in some detail below. However, the fact that
even in chaotic shops (PU's) throughput times can be controlled is fundamental to the validity of
so-called planned lead times. These planned lead times are the means to decompose material
coordination, workload control and PU control. As material coordination deals with work order
Workloadcontrol
PU - Control
Aggregateproductionplanning
Materialcoordination
Work orderrelease
aggregate deliveryplan
PU-state(detailed)
(detailed)
(aggr.) releasepattern
work orderpriorities
sales information
(capacities, inv.)goods flow state
information (aggr.)sales + marketing
PU-state(aggregated)
capacity +capacity use
108
releases to multiple PU's in parallel under uncertainty in end-item demand, the reduction of
complexity by assuming constant lead times is enormous. Eventually it made it possible to
develop an alternative for MRP I under the acronym SBS, which stands for Synchronized Base
Stock policies (cf. De Kok and Fransoo (2003)), that seems to provide a valid description of real
world value networks. To understand this statement one should be aware of the fact that the
workload control concept assumes 100% material availability. In order to ensure validity of
workload control, the material coordination function should propose work orders to be released,
of which material availability has been checked. If so, the workload control function provides
the necessary information to the work order release function needed for timing the actual
releases to the shop floor. Below we argue that material coordination is periodic by nature. The
work order release function translates discrete time decisions into continuous time decisions.
The impact of guaranteed material availability should not be underestimated. If material
availability is checked after work order release, then continuous work order cannibalization is
the consequence, or regular idleness of resources. Both lead to a delay of output to the market or
successor PU's. To manage the consequences of delays, excessive manual labor is required,
together with additional slack in the form of safety stock, safety lead time and safety capacity.
In my view, this explains the success of the implementation of the SBS material coordination as
described in De Kok et al (2005).
Thus, workload control, ensuring reliable (predictable) lead times, and material coordination
based on the BOM and these constant lead times, ensuring 100% material availability, create a
consistent "how" of planning and control.
3. Exploiting commonality or not
Above we argued that material coordination should ensure a 100% availability check before
communicating work orders to the work order release function. Though we explicitly mentioned
SBS, based on the echelon stock concept and an explicit allocation mechanism, there are
multiple alternatives for material coordination decision support. The material coordination
problem can be formulated as an LP, or one may apply some finite loading heuristic, where
material requirements are satisfied one after the other according to some priority sequence. In
De Kok and Fransoo (2003) it is shown that LP's extremal solutions subsequently generated in a
rolling schedule context, are not robust against demand uncertainty. The same holds for priority
sequences. Having said this, still there are many alternatives for material coordination under a
concept like SBS (i.e. echelon-order-up-to-policies).
109
To clarify this, we consider an assemble-to-order environment, where customer orders are filled
from component stock. Bertrand et al (2000) propose a so-called hierarchical pseudo item Bill
Of Material as the means for material coordination. This BOM structure effectively separates
the material coordination of the product family under consideration into material coordination of
a common set of items and sets of items defined by options that can be chosen by customers.
Each set of items, i.e. pseudo item, is controlled effectively as a single item and independently
of the control of other pseudo items.
It is easy to see that the quantitative analysis in Bertrand et al (2000) is of a heuristic nature, as
the independent control of pseudo items yields an overall performance that depends on
interactions between demands for pseudo items, which is ignored. Furthermore, differences in
lead times of components are ignored. Still, the idea of separating control of different pseudo
items generated a discussion on material coordination between Will and me from a conceptual
point of view.
Consider an organization, where multiple account managers are responsible for different market
or product segments. The items in the product or market segments have common and specific
components. For the specific components the account manager is responsible for material
coordination. For the common components a separate organizational entity is responsible for
material coordination. Though all of this seems obvious, there is a fundamental issue: the
account managers are responsible for meeting sales targets, while meeting these sales targets
depends on the availability of the common components. There is a separation of responsibilities,
yet authority is not separated as account managers cannot decide on common material
availability themselves. Here we identify a trade-off between the benefit of clear responsibility
and authority separation and the benefit of pooling the demand for common items. Will
Bertrand emphasized the first aspect, whereas I emphasized the second aspect. How to settle this
matter?
In my view only empirical research can provide insight into the trade-off to be made, as the
benefit of clear and aligned responsibilities and authorities are hard to quantify. Still,
quantitative analysis may shed light on the benefits of exploiting the commonality embedded in
the BOM. Unfortunately we are faced with a fundamental issue: for arbitrary BOM's there is no
hope to find optimal material coordination policies, not even under the classical (simplifying)
assumptions of stationary demand for end-items and linear holding and penalty costs. Hence the
benefit of exploiting commonality depends on the particular material coordination policy
chosen: LP, SBS, finite loading, MRP.
As commonality exploitation is only relevant under demand uncertainty, we use SBS policies as
coordination mechanism; as under these policies we can accurately compute long-run average
110
holding costs under service level constraints. Based on our experiments we found the following
insights.
1. Commonality does not tell the whole story. The lead time structure is of equal
importance. Commonality can only be exploited if specific items have relatively short
lead times.
2. Commonality has the greatest impact in situations where fast movers have more or less
equal demand rates and demand for each of the fast movers has moderate to high
variability.
3. Exploiting commonality is counter-productive if low cost volatile demand end-items
share components with high cost moderately variable demand end-items.
The third situation is due to the fact that the material coordination mechanisms known to date
are non-optimal and cannot manage this type of situation well. We might think that the third
situation is unlikely to occur, as we expect high cost items to have lower demand and higher
variations. However, fast moving items may show considerable variations due to the bullwhip
effect. Another relevant situation is where components and subassemblies are used as spare
parts to support the maintenance of an installed base. Demand for spare parts may be highly
variable, and the spare part is less expensive than the final product it is part of. Thus, it may be
wise to keep separate safety stocks for spare parts.
The consequence of the above is that in many situations we cannot exploit commonality to its
full extent and it may be even unwise to exploit it. Taking into account the intangible benefits of
alignment of responsibility and authority, Will Bertrand may be right in not exploiting
commonality in many practical situations.
4. Do we know the mean?
With a suitcase full of models and methods I arrived at Will's group in 1992. Will emptied my
suitcase in one stroke as he noted in a discussion that stationary models are not valid; we do not
even know the mean of random variables that realize themselves in the future! I thought it wise
to ignore this statement, as researchers must be productive: publish or perish.
Having said this, the problem of dynamics in demand processes has kept me busy ever since, be
it only now and then. Over the years I developed results for stationary models, from these
111
derived policies that can be used in non-stationary situations, and implemented them with the
help of many others. With a series of MSc students we developed value network models
describing as-is situations and validated them with data from ERP systems. To give some idea
of our approach:
1. Scope the value network to be analyzed.
2. Select "critical" items.
3. Create the resulting BOM and lead time structure.
4. Compute average stocks and average lot sizes for all items over a relevant period in
time using data from ERP.
5. Compute actual operational service levels (e.g. fill rate) over the same period of time
using data from ERP.
6. Compute means and standard deviations of all end-items over the same period in time
using data from ERP.
7. Use BOM, lead times, average stock, mean and standard deviations of end-items in a
quantitative (value network) model and compute the model operational service level.
8. Compare model and actual operational service levels.
The overall findings are quite positive. Given the fact that the data about demand for an
individual end-item are usually limited, we cannot expect very accurate results. Think of
throwing a dice 12 times. You will not find 2 times 1, 2 times 2, etc. However, when averaging
over all operational service levels, one finds that model outcomes and actual outcomes are quite
close. The inevitable conclusion is two-fold:
1. Real life demand data can be assumed to be generated by a stationary demand model.
2. The average inventories across a value network determine the operational performance
of this value network.
I consider these conclusions inevitable as the model outcomes could not be similar to the actual
outcomes, otherwise. This is obvious for the first conclusion, but the second conclusion
warrants some explanation.
We should be aware that the performance of a value network depends both on the operational
control concept and its parameters (safety stocks, safety times, lead times and lot sizes). In De
Kok and Fransoo (2003) it is shown that meeting a given fill rate requires different safety stocks
when using LP in rolling scheduling then when using SBS policies. It is reasonable to argue that
safety stocks needed to meet fill rate targets in practical situations where MRP is used in
conjunction with human interventions are again different. The behavior of a value network
under SBS policies is clearly different from its behavior in practice. Still, model outcomes
112
resemble actual outcomes. These two observations can be consistent when observing that
different parameters for different material coordination concepts may yield the same average
stocks. (Note that in our eight-step approach we do not use control parameters from ERP).
Apparently, the actual system with its average stocks performs more or less the same as the
system under SBS policies with the same average stocks.
From the above we may conclude that modeling individual end-item demand as stationary i.i.d.
demand is warranted. Unfortunately, when deciding on safety stocks, lot sizes and (planned)
lead times, or even the design of the manufacturing and distribution network, and the design of
the product family under consideration, we must guesstimate the mean and variance of end-item
demand. This can be seen as the Achilles heel of any design problem. Our findings are only in-
sample findings. We used historical data and showed that our process model of the value
network is correct. So probably when applying the same approach next year we will find the
same. Yet the decisions must be taken now, before knowing mean and variance of end-item
demand. Are all our efforts in vain?
In my view our efforts enable a quantum leap in performance. Whereas before the development
of quantitative models for general value networks only ad-hoc rule of thumb methods have been
used for setting parameters, we now know that quantitative models can be used effectively for
this purpose, subject to our ability to forecast demand. This is a huge reduction in problem
complexity. On top of that one finds that in many cases exponential smoothing is the most
effective forecasting method for individual end-item demand (Makridakis and Hibon (2000)).
This implies that extrapolation of demand is effective in many situations. This leads to the
following approach to set tactical parameters for a given value network.
1. Scope the value network to be analyzed.
2. Select "critical" items.
3. Decide on BOM, lead time structure and lot sizes.
4. Decide on operational end-item service level (e.g. fill rate) requirements over the future
period under consideration.
5. Decide on means and standard deviations (forecast errors) of all end-items over the
same period in time.
6. Use BOM, lead times, service level requirements, mean and standard deviations of end-
items in a quantitative (value network) model and determine optimal safety stocks.
7. In case results are not satisfactory w.r.t. costs and investments, go back to step 3.
8. Translate safety stocks to control parameters in ERP.
113
Clearly, the above approach is iterative. It may seem undoable to take all decisions in steps 3-5.
Yet we start from an existing situation and we can use improvement plans w.r.t. lead times, lot
sizes and forecasting to generate scenarios efficiently. Thus we create a methodology where
knowledgeable people identify improvement directions and demand trends, and decision
support tools handle the intrinsic complexity of optimization of safety stock parameters for
value networks under demand uncertainty.
The feasibility of the above approach has been field tested in MSc thesis projects at Océ, Bayer,
Grass Valley and P&G, amongst others. The essential and most interesting part of the approach
is the scenario development. The data gathering activities needed for this often reveal short-term
gains in master data cleaning, as well as inconsistencies and errors in control parameters. This
data gathering also reveals that, despite implementation of tools such as SAP APO, by and large
work order release decisions are based on human judgment: in our cases lot sizing parameters in
planning systems do not have a relationship with average lot sizes derived from historical data
from ERP systems. This even more supports our claim that a more rigorous approach towards
parameter setting can yield substantial improvements.
In conclusion, we do not know the mean demand. Yet extrapolation of historical data may be
effective for a large part of the product portfolio. For the remainder of the product portfolio
demand scenarios must be developed. The demand information together with the other value
network information can be used for the analysis of alternative value network scenarios. From
the results of this analysis decisions can be taken concerning control parameters and even the
design of the value network and associated product portfolio.
5. Discrete or continuous time
Around 2005 Will and I discussed whether time should be modeled as a continuous or discrete
variable. Clearly, in reality time is continuous. Having worked primarily on discrete time multi-
item multi-echelon models, I consider time a discrete variable in the context of material
coordination. As Will worked primarily on workload control, employing continuous time
queuing models, his position was that time should be modeled as a continuous variable: another
opportunity for a debate leading to mutual learning.
I think we came to terms along the following lines. Both aggregate planning and material
coordination encompass by nature multiple production units (PU’s) and multiple business
functions. Both aggregate planning and material coordination are driven by forecasts of demand.
Forecasts of demand are the result of a periodic decision making process involving sales and
marketing, as well as manufacturing and supply chain management. Decision making in
114
companies follows a recurrent calendar: daily, weekly, monthly, quarterly and yearly decision
are made. A yearly decision should set a frame of reference, if not constraints, for quarterly and
monthly decisions, which in turn set a frame of reference (constraints) for weekly and daily
decisions. Most decisions frequencies come naturally: short-term decisions are taken more
frequently than mid-term decisions. Now and then frequencies of decisions are changed
(increased) because of changes in the environment, such as ash clouds and financial crises.
From the above it follows that decision support models for aggregate planning and material
coordination should use time as a discrete variable. As time is continuous in reality this
modeling error has to be taken into account within the decision support models for aggregate
planning and material coordination. Any modeling error results into slack in time or quantity.
One might argue that the ultimate goal must be to increase the frequency of decision making, so
that this type of slack can be eliminated. However, this line of thought ignores the fact that
human decision making takes time, so that the decision interval should at least be longer than
the decision time. Furthermore, the quality of information used in decision making may not
improve by taking decisions more frequently. The converse may be true, as estimation of
exogenous variables, such as demand, may introduce more noise into the data used as the
frequency of estimation increases. It is easy to show that in that case the quality of decision
making deteriorates. Hence there must be an optimal decision frequency for each decision to be
taken, dependent on the quality of the input data for that decision under different frequencies of
decision making. It is well possible that the frequencies used in practice are close to optimal, as
these frequencies are based on experience over many years.
One of the key inputs for decision making is the so-called state of the system under
consideration. In the context of the decision framework of Bertrand et al. (1990) the state of the
system consists of the net stock and outstanding orders of each item, customer orders to be
satisfied, as well as available resources over the planning horizon. As decision making takes
time, the decisions taken are implemented during a future time interval that starts at some point
in time in the future. This implies that for each recurrent decision making process we do not
merely record the state of the system; we estimate the state of the system. Consider as an
example a weekly planning process. The decision making process is executed on Thursday
afternoon, based on data extracted on Wednesday morning and the decisions concern work
order releases for Monday next week. At the moment of decision making we estimate the state
on Monday morning from the real-time state on Wednesday morning, by making assumptions
about order completions, inventory replenishments and work order releases between Wednesday
morning and Monday morning. The more accurate these estimations the less slack in time and
quantity is needed to protect against uncertainty. In quantitative models this phenomenon is
hardly ever taken into account. In my view this phenomenon supports the concepts of planned
115
lead times and the need for reliable lead times as pursued by workload control. It also explains
why planners spend so much time in gathering information from the shop floor. One way or
another they need to estimate the short term future, with the sole purpose of estimating the state
of the system in the planning system for the start of the interval in which decisions are
implemented.
In summary, real time is continuous, but decision support systems for material coordination and
aggregate planning should assume discrete time. Whether workload control and work order
release are discrete time or continuous time processes, depends on the PU resource
configuration and the work orders to be released. Here the debate between Will and me is still
open.
Epilogue
In the above I discussed a number of topics that have been subject for discussions between Will
Bertrand and me. In my view these topics are essential for the implementation of the framework
for production planning and control developed by Bertrand et al. (1990). Some progress has
been made on the how of planning and control, but much more needs to be done. The dominant
paradigm in planning and control is still based on rolling scheduling using mathematical
programming models. The fundamental difference between processes under complete certainty
about the relevant future and process under uncertainty about the relevant future is ignored
completely. At the same time basic inventory and queueing models from stochastic Operations
Research are used naively to set planning parameters. Will Bertrand’s research focuses on
principles for planning PU’s with realistic configurations under implementable control rules.
My research focuses on principles for material coordination for realistic value networks under
implementable control rules. We have shown effectiveness of these principles by implementing
them with companies. Not on a large scale, but in pilot projects we have shown the effectiveness
of these principles. In my view we need to make further steps in translating the principles of the
“what” and “how” of planning and control into implemented systems and procedures. We need
feedback from empirical studies to test our hypotheses and tools. We are on the brink of making
this happen. If Will Bertrand’s plan for the PDEng program of the department of IE&IS of
Eindhoven University of Technology is implemented, within 5 years we can implement our
concepts, methods, and tools at more than 50 different companies. We can check this in 2017.
The learning continues.
116
References
Ackoff, R.L., 1979, The Future of Operational Research is Past. The Journal of the
Operational Research Society, Vol. 30, 93-104.
Anthony, R.N. 1965. Planning and control systems: A framework for analysis. Boston: Harvard
Business School Press.
Bertrand, J.W.M., Zuijderwijk, M. and Hegge, H.M.H., 2000, Using hierarchical pseudo bills of
material for customer order acceptance and optimal material replenishment in assemble
to order manufacturing of non-modular products, Int. J. Production Economics 66, 171-
184.
Bertrand, J.W.M., and Wortmann, J.C. 1981. Production control and information systems for
component-manufacturing shops. Amsterdam: Elsevier.
Bertrand, J.W.M., J.C. Wortmann and J. Wijngaard (1990). Productiebeheersing en material
management (in Dutch), Leiden: Stenferd Kroese.
Bertrand, J.W.M., Kok, A.G. de (1999). Macht en operationele ketenbeheersing in de
industriële productie (in Dutch). Bedrijfskunde: tijdschrift voor modern management,
71(2), 15-23.
Bertrand, J.W.M., Wortmann, J.C., and Wijngaard, J. 1990. Production control: A structural
and design oriented approach. Amsterdam: Elsevier.
Conway, R.W., Maxwell, W.L., and Miller, L.W. 1967. Theory of scheduling. London:
Addison-Wesley.
De Kok, A.G., and Fransoo, J.C. 2003. Planning supply chain operations: Definition and
comparison of planning concepts. Pages 597–675 of: De Kok, A.G., and Graves, S.C.
(eds), Handbook in Operations Research and Management Science, Volume 11:
Design and Analysis of Supply Chains. Amsterdam: Elsevier.
Forrester, J.W. 1961. Industrial dynamics. Massachusetts: The M.I.T Press.
Galbraith, J. R. (1973). Designing Complex Organizations. Addison Wesley.
Hax, A.C., and Meal, H.C. 1975. Hierarchical integration of production planning and
scheduling, 53–69 of: Geisler, M.A. (ed), Logistics. Amsterdam: Elsevier.
117
Holt, C.C., F. Modigliani, J.F. Muth and H.A. Simon (1960). Planning, Production,
Inventories and Workforce. Englewood Cliffs: Prentice Hall.
Hopp, W.J., and Spearman, M.L. 2000. Factory physics: Foundations of manufacturing
management. 2nd edn. New York: McGraw-Hill.
Kok, A.G. de, F. Janssen, J. van Doremalen, E. van Wachem, M. Clerkx and W. Peeters
(2005). Philips Electronics Synchronizes Its Supply Chain, Interfaces 35, 37–48.
Makridakis, S. & Hibon, M. (2000). The M3-Competition: results, conclusion and
implications, International Journal of Forecasting, 16(4), pp. 451-476.
Schneeweiss, C. 1999. Hierarchies in distributed decision making. Berlin: Springer-Verlag.
Van de Wakker, A.M., 1993, Throughput Time Control and Due Date Reliability in Tool &
Die Shops, PhD Thesis Eindhoven University of Technology.
Van Ooijen, H.P.G., 1996, Load-Based Work-Order Release and its Effectiveness on Delivery
Performance Improvement, PhD Thesis Eindhoven University of Technology.
118
119
IN THE GALLERY OF CELEBRITIES
Marc Lambrecht
Abstract This paper is written in honor of Prof. J.W.M. Bertrand. The paper deals with a problem where
both of us are interested in, namely, the issue of production smoothing. The research topic of
volatility in demand and production has been the subject of numerous studies. Theoretical work
goes back to the early fifties and there is an extensive literature in macroeconomics. The issue
of volatility was reactivated through the renewed interest in the bullwhip effect. My attention in
this paper will go to a single authored paper written by Will: Balancing production level
variations and inventory variations in complex production systems. This paper was published in
1986 in the International Journal of Production Research. I will focus on the importance of that
paper and I will position the paper in the extensive literature dealing with the problem.
1. Introduction
I am honored to contribute to the Friendship Book dedicated to Prof. J.W.M. Bertrand (allow
me to address you as Will). We both belong to the same generation of researchers. We never
published together but our research lines crossed each other on several occasions. In this paper I
would like to talk about our personal crossroads of research. Will wrote papers on research
issues that became popular (or became research waves) decades later (and not the other way
around). It’s like an avant-garde painter, being ahead of time. That’s my criterion to belong to
the gallery of celebrities. In this paper I will illustrate this by focusing mainly on one single
authored paper written by Will: “Balancing production level variations and inventory variations
in complex production systems” (Bertrand 1986). The objective is to position this paper in the
literature on production smoothing and the bullwhip phenomenon. Production smoothing has
also been on my research agenda. In section 2 we will summarize the early literature on
production smoothing. In section 3 we position the work of Will Bertrand in that research area.
In section 4 we summarize the major findings concerning production smoothing in the empirical
operations management and macroeconomics literature. In section 5 we conclude.
120
2. The bullwhip problem and production smoothing :
overview of theoretical results.
The bullwhip effect is a short-hand term for a dynamical phenomenon in supply chains. It refers
to the tendency of the variability of order rates to increase as they pass through the echelons of a
supply chain towards producers and raw material suppliers. Hau Lee (1997) wrote two
important articles on the subject and put it again on the research agenda. Since then, numerous
papers were published (for a summary (see Disney and Lambrecht, 2007)) on the subject and
the Beer Distribution Game was omnipresent in classrooms all over the world. Jay Forrester
(1961) was among the first researchers to describe this phenomenon of the amplification of
oscillations of the material flow as one moves farther away from the customer. The bullwhip
problem is basically a problem of managing volatility in demand and how companies react to
that volatility.
There is ample anecdotic evidence that many companies experience significant extra costs due
to these supply chain problems. Bloated stocks sat alongside empty racks and display shelves. It
is a formidable job for logistics managers to design order management systems that optimally
match pipelines to the marketplace. It is not the objective of this paper to evaluate all possible
solutions to that problem, instead we will focus on one specific research line, namely order
smoothing replenishment rules and production level smoothing models (in short, smoothing
models). Production/ordering smoothing models have been around for many years. Pioneering
work was done by Holt, Modigliani, Muth and Simon (1960). This famous HMMS paper was
compulsory reading in almost all operations management classes. In a private conversation with
Will I tried to find out whether he still uses, today, the HMMS model in class. It really looks so
old fashioned. But remember that in de gallery of celebrities nothing and nobody is old
fashioned. Just replace HMMS by Sales & Operations Planning and we transfer the problem to a
contemporary business issue. The HMMS model was at the start of a whole series of papers in
the area of aggregate production planning. Magee (1956-1958) also produced pioneering work
on the development of smoothing rules. The Bertrand 1986 paper to be discussed in the next
section also refers to the work of Magee. In the 1950s and 1960s a number of production
smoothing rules were developed (Simon (1952), Vassian (1955), Deziel and Eilon (1967).
More recent work on smoothing replenishment rules can be found in Dejonckheere et al. (2003),
Balakrihnan et al. (2004) and Disney and Lambrecht (2007)). Caplin (1985), Baganha and
Cohen (1998) and Kelle and Milne (1999) study the relationship between (the variance of) (s,S)
replenishment ordering policies and the variance of demand.
121
Smoothing is a well-known method to reduce variability. No wonder that production smoothing
models also received a lot of attention in the macro-economic literature (see section 4). Early
theoretical investigations of optimal inventory control and production behavior established that
if production costs are convex, then it is optimal for a firm to only partially adjust output in
response to a change in its inventory position. This production smoothing hypothesis
investigates whether production is smoothed relative to sales from a macro-economic
perspective.
It must be clear that many methodological approaches can be used to solve the above problem.
Control theory, Laplace transform (continuous time), z-transform (discrete time) and various
control engineering techniques can be used for this purpose. At this point it is good to mention
that one of the first papers by Will Bertrand deals with cybernetics and control (Kickert,
Bertrand et al. 1978).
3. Is smooth smart : Balancing production level variations and
inventory variations.
Since large variations in order/production rates are undesirable because of the high cost
implications, taming or dampening order variability, seems to be a dominating operations
strategy. But we have to be careful not to focus only on one side of the production smoothing
“coin”. In developing production or replenishment rules we have to consider the impact on the
inventory variance as well. Dampening may have a negative impact on customer service due to
inventory variance increases. Today, this observation is widely accepted, and almost all research
papers deal with both the order rate variance and the inventory variance.
The 1986 paper by Will Bertrand: “Balancing production level variations and inventory
variations in complex production systems” is dealing with the problem mentioned in the
previous paragraph. Let’s discuss the paper and point out how important the paper is in the
literature. The paper is situated in a material requirements planning (MRP) context and deals
with the variability of the master production schedule (MPS). I consider this very typical for the
work of Will, that means, at least the problem must exist and there must be a managerial issue.
Moreover, there must be a methodological contribution.
Allow me to quote some paragraphs from the paper. I ask the reader to go through the 2010-
2011 literature on the same topic and realize that it is still a major research problem. In the
introduction we read “We will derive a new production control decision rule that incorporates
122
production smoothing. However, the use of such a rule may lead to increased inventory
variations. We will show how the resulting variances in required production and in stock depend
on the characteristics of the production system (lead times, smoothing factors, capacity
requirements)”. This is what I call a clear problem statement. The introductory paragraph of the
paper finishes as follows: “We will show how to build and use a cost model for balancing the
safety stock costs and the production level variation costs for the type of situation under
investigation”. More than twenty years later, exactly that problem was on my personal research
agenda. Our research lines crossed each other. The visionary character of the 1986 paper is
further illustrated by the fact that the following questions are answered in the paper: What is the
optimal smoothing parameter? How to quantify the trade-off between reduction of spread of
production versus an increase in spread of physical stock? At what stage of a production system
is production smoothing more attractive? Will was not able to solve the problem analytically,
still today, the optimization model is very hard to solve.
4. Production smoothing : empirical evidence
Production smoothing has often been considered an appropriate operations management
strategy. One of the important features of the Toyota Production System is Heijunka. Heijunka
stands for production leveling and the purpose is to facilitate stability in the production process.
Production smoothing is also desirable with the combination of seasonality and stochastic
shocks (Cachon et al., 2007). An often heard argument in favor of smoothing is when
production costs are convex. Production flexibility is the opposite strategy. In a production
flexibility strategy companies opt to “chase demand” instead of producing at a constant rate
throughout the year. Such a strategy minimizes inventory build ups. All this illustrates that the
issue of production smoothing is very controversial in practice. Consequently, there is a need to
empirically validate our research outcomes. In recent years we experience a rise in the area of
empirical operations management. The simple question is whether production is more variable
than sales (in retailing, wholesaling or manufacturing). Allow me to summarize a number of the
major findings. Blinder (1986) argues that the production smoothing idea is all wrong. Some
authors will find overwhelming evidence against the production smoothing model, whereas
other authors find empirical evidence that confirms the smoothing hypothesis.
Cachon et al. (2007) gives several reasons why the empirical evidence is not aligned with the
production smoothing hypothesis. There is a conflict between theory and observation on the
smoothing issue. In a recent study Cachon et al. (2007) wrote a paper with a very suggestive
title: In search of the bullwhip effect. These authors conclude, based on a large data set, that the
123
bullwhip effect is often not observed in industry-level data. More research is needed to undo this
knot.
5. Conclusion
In this paper I focused on a publication written by Will Bertrand in 1986. The paper deals with
production level variations and inventory variations. Both operations management researchers
and economists are interested in the problem. The research question raised in that paper has a
long history and empirical evidence and shows that the issue is controversial. The debate is far
from finished; we have to deepen our understanding of the problem. The 1986 paper plays a
central role in this fascinating production smoothing debate.
References
Balakrishnan, A., Geunis, J. and Pangburn, M.S., (2004). Coordinating supply chains by
controlling upstream variability propagation. Management & Service Operations Management 6
(2), 163-183.
Baganha M. and Cohen, M.S., (1998). The stabilizing effect of inventory in supply chains.
Operations Research 46 (3), S72-S83.
Bertrand, J.W.M., (1986). Balancing production level variations and inventory variations in
complex production systems. International Journal of Production Research, 24(5), 1059-1074.
Blinder A., Lovell, M.C. and Summers L.H., (1981). Retail inventory behavior and business
fluctuations. Brookings on Economic Activity, 1981(2), 443-520.
Blinder A., (1986). Can the production smoothing model of inventory behavior be saved ?
Quart. J. Econom., 101, 431-453.
Cachon, G., Randall T. and Schmidt G.M., (2007). In search of the bullwhip effect.
Manufacturing & Service Operations Management, 9 (4), 457-479.
Caplin, A.S., (1985). The variability of aggregate demand with (s,S) inventory policies.
Econometrica, 53, 1395-1410.
124
Dejonckheere, J., Disney, S.M., Lambrecht, M.R. and Towill, D.R., (2003). Measuring and
avoiding the bullwhip effect : a control theoretical approach. European Journal of Operational
Research, 147 (3), 567-590.
Deziel, S.M. and Eilon, S., (1967). A linear production-inventory control rule. Production
Engineer, 43, 93-104.
Disney S.M., Lambrecht, M.R., (2007). On replenishment rules, forecasting, and the bullwhip
effect in supply chains, Foundations and Trends in Technology, Information and Operations
Management, 2(1), 1-83.
Dooley, K.J., Yan, T., Mohan, S. and Gopalakrishnan M. (2010), Inventory management and
the bullwhip effect during the 2007-2009 recession : evidence from the manufacturing sector.
Journal of Supply Chain Management, 46 (1), 12-18.
Forrester, J. (1961). Industrial dynamics. Cambridge, MA :MIT press.
Holt, C.C., Modigliani, F., Muth, J.F. and Simon, H. (1960). Planning production, inventories
and work force. Englewood Cliffs : Prentice-Hall.
Kahn, J. (1987). Inventories and the volatility of production. American Economic Review, 78,
667-679.
Kelle, P. and Milne, A. (1999). The effect of (s,S) ordering policy on the supply chain.
International Journal of Production economics, 59, 113-122.
Kickert, W.M., Bertrand, J.W.M., and Praagman J., (1978). Some comments on cybernetics and
control. IEEE Transactions on Systems, Man and Cybernetics, 8(11), 805-809.
Lee, H.L., Padmanabhan, V. and Whang, S.,(1997a). The bullwhip effect in supply chains.
Sloan Management Review, 38 (3), 93-102.
Lee, H.L., Padmanabhan, V. and Whang, S., (1997b). Information distortion in a supply chain:
the bullwhip effect. Management Science, 43 (4), 546-558.
Magee, J.F., (1956). Guides to inventory control, part II : problems of uncertainty. Harvard
Business Review, 34, 103-116.
Magee, J.F., (1958). Production planning and inventory control. McGraw-Hill, New York, CA.
Milne, A., (1994), The production smoothing model of inventories revisited, The Economic
Journal, Royal Economic Society, 104, 399-407.
125
Simon, H.A., (1952). On the application of servomechanism theory to the study of production
control. Econometrica, 20, 247-268.
Vassian, H.J., (1955). Application of discrete variable servo theory to inventory control. Journal
of the Operations Research Society of America, 3, 81-96.
126
127
EEN EXCURSIE DOOR EEN ORDER
ACCEPTATIE MODEL EN HET GEBRUIK
DAARVAN TEN BEHOEVE VAN HET
ONTWERPEN VAN DE
BEHEERSING VAN EEN ÉÉN-FASE
PRODUCTIE-VOORRAAD SYSTEEM
Jacob Wijngaard
Samenvatting De inrichting van een supply chain en de wijze van beheersing ervan is een ontwerp. Het is
belangrijk te weten hoe zo’n supply chain werkt. Het moet een beheerst systeem zijn. Dat lukt
alleen maar als de supply chain bestaat uit eenvoudige elementen die eenvoudig worden
samengesteld. Een relevante deelverzameling van systemen bestaat uit de productie-voorraad
systemen die gedomineerd worden door één productiefase. Er zijn meer producten en meer
ordersoorten, met verschillende prioriteiten en levertijden. En meer resources die deels
uitwisselbaar zijn. In deze bijdrage wordt een eenvoudig model geëxploreerd dat wellicht van
dienst kan zijn bij het ontwerpen van een adequate beheersing voor dit type van systemen.
1. Inleiding
In ons gezamenlijke werk was het begrip “Ontwerp” belangrijk (zie Betrand et al. 1990). De
inrichting van een supply chain en de wijze van beheersing ervan is een ontwerp. Het is
belangrijk te weten hoe zo’n supply chain werkt. Dat lukt alleen maar als de supply chain
bestaat uit eenvoudige elementen die eenvoudig worden samengesteld. Je moet er bovendien
voor zorgen dat de supply chain werkt in een beheerste omgeving. Ik hoef dat hier verder niet
uit te leggen. Het gaat in die ontwerpbenadering vooral om “eerste orde effecten”. Mijn rol in
de samenwerking was om met behulp van eenvoudige modellen inzicht te krijgen in zulke
“eerste orde effecten”. Het lijkt me gepast om me in deze bijdrage aan die rol te houden. Hier
128
daarom de behandeling van een eenvoudig supply chain model. Het betreft een situatie met
twee vraagklassen, beperkte productiecapaciteit en verschillende klantenorder levertijden en
prioriteiten. Naar mijn mening is er nog veel onbekend over het effect van voorraadvorming in
zulke situaties. Het betrekken van orderacceptatie is daarbij wezenlijk.
De nadruk in dit verhaal ligt op de presentatie en bespreking van een numeriek voorbeeld. Een
soort van excursie dus. Op basis daarvan hoop ik toch wat conclusies te kunnen trekken.
Gezien de exemplarische aard van de excursie, zijn dat natuurlijk alleen maar indrukken en
vermoedens. In de laatste sectie besteed ik kort aandacht aan de kwestie of dit model ook
bruikbaar is bij het ontwerp van de productiebeheersing voor een klasse van systemen uit de
praktijk. Namelijk de klasse van productie-voorraadsystemen die gedomineerd worden door één
productiefase. Er zijn meerdere producten en meerdere ordersoorten, met verschillende
prioriteiten en levertijden. En er zijn meerdere resources die deels uitwisselbaar zijn.
1.1. Model We onderscheiden twee vraagklassen, en . Het aankomstproces is Poisson ( ). De
vraag per keer volgt een verdeling met distributiefunctie . De gevraagde klantenorder
levertijd is . De netto opbrengst per eenheid is . Orders kunnen wel of niet
geaccepteerd worden. Voor de productie is er één machine met snelheid 1. Die kan aangezet
worden voor een bepaalde runlengte ( tijdseenheden). Na afloop van die run kan er verlengd
worden voor weer een runlengte of kan er gestopt worden. Het aanzetten van de productie
brengt geen extra kosten of tijdverlies met zich mee. Er kan op voorraad geproduceerd worden.
De voorraadkosten zijn per eenheid product per tijdseenheid en de tekortkosten zijn per
eenheid per tijdseenheid. De indices en zijn niet toevallig gekozen. Ik wil spelen met de
situatie waarbij klasse een hogere prioriteit heeft dan klasse . Dat kan zijn omdat of
omdat er alleen beslissingsregels worden toegelaten waarbij alle orders voor moeten worden
geaccepteerd. Een grotere klantenorder levertijd kan het gemakkelijker maken de klasse
prioriteit te geven.
1.2. Excursie Alleen klasse
We bekijken eerst het geval waarbij er alleen orderklasse is, en . Alle orders van dat
type worden geaccepteerd. De enige beslissing is de productiebeslissing. De machine wordt
uitgezet als voorraadniveau bereikt wordt. We gaan uit van het geval
en de vraag per order uniform [0,2] verdeeld. Figuur 1 geeft simulatieresultaten voor
en variabel. De runlengte is 10000 tijdseenheden. De opbrengst heeft alleen
129
invloed op de hoogte van de totale opbrengst, niet op de plaats van het maximum, omdat alle
orders geaccepteerd worden.
Figuur 1: Opbrengst als functie van productielimiet bij
en uniform [0,2] verdeelde ordergrootte.
We zien een heel normaal verband waarin het balanceren van voorraadkosten en tekort kosten
geïllustreerd wordt. Bij (ongeveer) ligt het optimum.
Twee orderklassen
Nu laten we ook toe dat er van orderklasse mag worden geaccepteerd. Ook voor deze orders
is de klantenorder levertijd gelijk aan 0. We nemen aan dat bij aankomst van de order bekend is
hoe groot die is en dat de order ook gedeeltelijk mag worden geaccepteerd. De aankomst-
intensiteit en de ordergrootte verdeling worden ook aangenomen bekend te zijn. Voor de hand
ligt dat in dit geval de optimale acceptatieregel gekarakteriseerd kan worden door een
acceptatielimiet, een voorraadniveau zodanig dat er bij een voorraad geen orders van
klasse worden geaccepteerd en dat bij een hogere voorraad, , nooit meer dan wordt
geaccepteerd. Dat dit een redelijke verwachting is kan als volgt worden ingezien. Formuleer het
probleem als een stochastisch dynamisch programmeringprobleem. Laat de waarde zijn
bij voorraad , na acceptatie en de waarde bij voorraad , en een order ter grootte
waarvan de order acceptatie beslissing nog moet plaatsvinden. Dan geldt:
2,1
2,2
2,3
2,4
2,5
2,6
2,7
2,8
6,00 5,00 4,00 3,00 2,00 1,00
Reward(S)
130
Gezien de aard van de voorraad- en tekortkosten, mag je verwachten dat de waardefunctie
concaaf is. Daaruit volgt het bestaan van zo’n acceptatielimiet . Bij de in Figuur 1 gekozen
parameter setting blijken (ongeveer) optimaal te zijn.
Ter illustratie wordt in Figuur 2 de opbrengst als functie van gegeven. Het probleem kan ook
wel analytisch numeriek worden aangepakt. Dat blijft hier buiten beschouwing.
Tot zover niets nieuws. En zeker niets onverwachts.
Figuur 2: Opbrengst als functie van de acceptatielimiet ( ), bij
en uniform [0,2] verdeelde
ordergrootte, voor beide ordersoorten..
Positieve levertijd, equivalentie van geval en geval
De eerstvolgende stap is kijken naar het effect van een positieve levertijd. Er kan aangesloten
worden bij het artikel van Wijngaard en Karaesmen (2007). Daar wordt in de toestandsdefinitie
gebruik gemaakt van de virtuele voorraad, , gedefinieerd door:
,
met en de uit te leveren orders voor het interval Orders
“zie” je in zo’n aanpak gedeeltelijk als punten in het verleden, op interval , en
gedeeltelijk als punten in de toekomst, op interval . De ordergrootte betrek je
ook bij de toestand, evenals het type van de order.
2,52,55
2,62,65
2,72,75
2,82,85
2,92,95
33,05
4,7 3,7 2,7 1,7 0,7
Reward(s)
131
Op die manier zie je gemakkelijk dat elke beslissingsregel voor het geval met positieve één op
één correspondeert met een beslissingsregel voor het geval met , en omgekeerd. Ook de
order acceptatie onder de corresponderende regels correspondeert.
Het verschil in performance volgt uit de volgende relatie:
,
met de leegloop gedurende . Daaruit volgt dat het verschil in performance
tussen de situatie met en de situatie met niet groter kan zijn dan . En ook
dat als geldt dat de optimale beslissingsregel voor het geval met productie voorschrijft
zolang , die regel ook optimaal is voor het geval , zolang . Alleen wordt er
dan niet gereageerd op , maar op . Het verschil in performance is gelijk aan .
Dat komt omdat dan geldt dat . Voor niet te grote geldt dus dat het
geval met equivalent is aan het geval . Ik beperk me daarom hierna tot het geval
.
Positieve levertijd bij één orderklasse
Als de bezettingsgraad gegeven is, zoals in het geval met maar één orderklasse, waarvan alle
orders geaccepteerd moeten worden, kun je het hier boven genoemde resultaat direct toepassen.
In Figuur 3 (getrokken lijn) wordt de invloed van de levertijd geïllustreerd. Er wordt gestuurd
op basis van en de wordt constant gehouden. Zolang neemt de opbrengst lineair
toe, met snelheid . Daarna is eerst die toename kleiner en gaat de opbrengst al snel
dalen. Dat komt omdat dan de virtuele voorraad groot kan zijn en daardoor de productie stopt,
terwijl er toch stock-outs ontstaan. kan positief zijn, maar tussen en treedt zoveel
leegloop op omdat groter dan wordt, dat toch negatief wordt. Als je de
productieregel aanpast met deze productiecheck blijft de opbrengst veel langer (bijna) lineair
toenemen. Zie de onderbroken lijn in Figuur 3. Volledig lineaire toename is een bovengrens van
wat je kunt winnen met een positieve levertijd. Toepassing van een productielimiet ,
onafhankelijk van de levertijd, in combinatie met een buiten-voorraad check bij grotere
levertijden is dus over een brede range van levertijden bijna optimaal.
132
Figuur 3: Opbrengst als functie van de levertijd , bij
en uniform [0,2] verdeelde ordergrootte.
Positieve levertijd klasse orders bij ook klasse orders
Nu het geval van twee orderklassen, levertijd voor klasse en levertijd 0 voor klasse . Ook
dan is het zinnig te reageren op de virtuele voorraad:
,
met de orders voor H gedurende . Nu geldt:
,
met de orders voor L die gedurende worden geaccepteerd en
de leegloop gedurende . Ook nu geldt weer zolang .
Het positieve effect van een levertijd vergroting van orders voor klasse wordt daarmee, voor
kleine gelijk aan , met de bezettingsgraad van klasse orders. Zie
Figuur 4 (getrokken lijn) voor een illustratie. De parameterwaarden sluiten aan bij die in de
eerdere figuren. Voorbij kan de prestatie verbeterd worden door toevoeging van een
check op de actuele voorraad ( ) en de al eerder genoemde productiecheck. Die
resultaten worden
0
0,5
1
1,5
2
2,5
3
3,5
0 1 2 3 4 5 6 7 8 9 10
Reward(l)
Met check
Bovengrens
133
Figuur 4: Reward als functie van de levertijd , bij
en uniform [0,2] verdeelde
ordergrootte voor beide ordersoorten.
ook gegeven. Ook bij toepassing van beide checks blijken de resultaten bij grotere behoorlijk
achter te blijven bij de bovengrens. Nader onderzoek zou moeten aantonen of dat ligt aan de
bovengrens of aan de regel.
Ook weigeren van klasse orders?
Dit is zo ongeveer al het eind van de exploratieve wandeling. Het geval waarbij ook orders uit
klasse H geweigerd kunnen worden is niet erg interessant, zeker niet bij deze opbrengsten. Bij
deze opbrengsten moet je (bijna) alle orders uit klasse H toch accepteren. Zie Figuur 5 voor een
illustratie. Daarin worden totale opbrengst en bezettingsgraad als functie van de acceptatielimiet
gegeven. Er mag ook gedeeltelijk geaccepteerd worden. Vergelijkbaar met de eerder besproken
acceptatie van orders uit klasse L. De maximale opbrengst wordt pas bereikt bij een
acceptatielimiet van (ongeveer) -2. Bij die acceptatielimiet is de bezettingsgraad nog maar 2%
lager dan maximaal. De maximale opbrengst is kennelijk heel gevoelig voor de bezettingsgraad
(de totale opbrengst verandert veel sneller dan de bezettingsgraad). Anders zou het effect van
een positieve levertijd ook nog weer anders geschat moeten worden. Stel is de maximale
reward, gegeven dat de bezettingsgraad gelijk aan is. Dan geldt dat de maximale opbrengst bij
levertijd (voldoende klein), gegeven een
2
2,2
2,4
2,6
2,8
3
3,2
3,4
3,6
3,8
0 1 2 3 4 5 6 7 8 9 10
Reward(l)
Voorraadcheck
Productiecheck
Bovengrens
134
Figuur 5: Opbrengst als functie van de acceptatielimiet ( ), bij
en uniform [0,2]
verdeelde ordergrootte.
bezettingsgraad , gelijk is aan . Stel het optimum ligt voor bij . Dan is
een dalende functie van .
Bij een wat lagere bezettingsgraad kun je het zicht beter benutten. Maar gezien de gevoeligheid
van de totale opbrengst als functie van de bezettingsgraad leidt dit effect niet tot een heel ander
optimum. Het is een grappig effect, maar het lijkt voor de praktijk niet erg belangrijk.
2. Conclusies
Het lijkt inderdaad een niet of nauwelijks onderzocht probleemgebied, terwijl het wel relevant
is. Natuurlijk zit men in de praktijk met verschillen in urgentie en waarde tussen de
verschillende ordersoorten. En die gaan vaak gepaard aan een verschil in levertijd. Of moet je
dergelijke verschillen maar liever negeren? Ik kom daar nog op terug in de laatste paragraaf.
Theoretisch onderzoek naar de structuur van de optimale beslissingsregel voor het geval dat
beide levertijden gelijk aan 0 zijn, is toch nog wel interessant. Hierboven is er van uitgegaan dat
er ook gedeeltelijk mag worden geaccepteerd en dat dat niet onevenredig duur is. Dan lijkt voor
de optimale beslissingsregel te moeten gelden dat er tot een bepaald voorraadniveau
geaccepteerd moet worden. Dat moet beter worden gecheckt. En ook zou bekeken moeten
worden hoe de optimale beslissingsregel eruit ziet bij alleen maar volledige acceptatie?
0
0,5
1
1,5
2
2,5
3
2 1 0 -1 -2 -3 -4 -5 -6 -7
Reward(s)
Bezettingsgraad(s)
135
Het effect van een kleine levertijdverhoging is voor het geval alle orders geaccepteerd moeten
worden gelijk aan , met de totale bezettingsgraad. Als maar een deel van de orders
geaccepteerd hoeft te worden is dat effect gelijk aan , met de bezettingsgraad ten
gevolge van de hoge prioriteit orders.
Die lineariteit blijft ook (ongeveer) voor langere levertijden gelden, als er maar voor de hand
liggende checks aan de beslissingsregel worden toegevoegd. Wanneer alle orders geaccepteerd
moeten worden, is dat een productiecheck om ervoor te zorgen dat de productie (op basis van
de virtuele voorraad) niet te snel wordt uitgezet. Bij beslissingen over acceptatie van orders
met een lagere prioriteit moet er ook een check op de fysieke voorraad worden toegevoegd. Het
lineaire effect is dus (redelijk) robuust en kan wellicht daarom in het ontwerp van planning en
beheersing gebruikt worden. Het is de moeite waard te onderzoeken of dit “lineaire gebied” nog
verder kan worden uitgebreid via andere eenvoudige toevoegingen aan de beslissingsregel.
Dat het effect van een grotere levertijd evenredig is met laat zien dat een lagere
bezettingsgraad het grotere “zicht” beter bruikbaar maakt. Om die reden mag men bij langere
levertijden de neiging tot minder orders accepteren verwachten. Dat is een wat paradoxaal
effect dat nader onderzoek vraagt. Meer leuk dan nuttig overigens, lijkt me.
De excursie betrof een situatie met maar één product. Het gebruik van de resultaten in een
situatie met meer producten is niet eens zo moeilijk. In plaats van twee orderklassen voor
hetzelfde product kun je bijvoorbeeld denken aan twee verschillende producten, een product
waarvan alle orders geaccepteerd moeten worden en een product waarvan men ook orders
mag weigeren. Van staat men voorraadvorming toe.
Als er een order van binnenkomt beslist men of er voldoende voorraad van is om even om
te schakelen naar en die order mee te nemen. Aangenomen wordt dat de levertijd voor net
voldoende groot is om dat mogelijk te maken. Het resulterende model komt vrijwel overeen met
wat hierboven is behandeld. Er zijn natuurlijk allerhande verschillende situaties met meer
producten, maar als het allemaal via één resource gaat is aggregatie in de regel wel mogelijk. In
het geaggregeerde model moet je dan nog wel dat onderscheid naar verschillende prioriteiten en
verschillende levertijden maken.
136
3. Ontwerpen
De uiteindelijke vraag is of probeersels met dit model kunnen bijdragen aan het ontwerpen van
een systeem voor de beheersing van de in de inleiding geschetste klasse van productie-voorraad
systemen: één productiefase, meerdere producten en meerdere ordersoorten, met verschillende
prioriteiten en levertijden en meerdere resources die deels uitwisselbaar zijn. Zie het
proefschrift van Bertrand en Wortmann (1981) voor een uitgebreide behandeling van de
ontwerpbenadering in het algemeen. Een belangrijke ontwerpvraag is hoe je het design process
model moet kiezen of liever de design process modellen. Dat hangt samen met de formal
decision procedure die je beoogt en de vrijheid die je daarbij nog voorziet voor de beslissers.
De beslissingen die hier aan de orde zijn, zijn: Ga je deze resource inzetten? Voor welk
product? Moet deze order geaccepteerd worden? Voor de resource beslissingen is belangrijk
wat de inzetbaarheid van de resources is. Dat punt komt ook al aan de orde in het proefschrift
van Bertrand en Wortmann (1981). Later hebben Graves en Tomlin (2001) hier aanzienlijk aan
bijgedragen en het in de context van de supply chain theorie getrokken. Het model hierboven
draagt hier niets aan bij. Ik ga er al vanuit dat het op de één of andere manier mogelijk is in het
ontwerp te decomponeren naar een aantal los van elkaar staande of eventueel geneste resources.
Het decomponeren naar enkelvoudige resources wordt overigens niet fundamenteel anders als
er ook rekening gehouden moet worden met verschillende prioriteiten en levertijden.
De beslissing over resource gebruik en de samenhang ervan met order acceptatie komt wel aan
de orde in het model. Aggregeren over verschillende producten met verschillende
productiekosten, verschillende voorraadkosten en verschillende tekortkosten is niet
problematisch, zolang je de productmix maar redelijk kunt schatten en zolang de levertijden
maar gelijk zijn. In de model excursie hierboven wordt die aggregatie al voorondersteld. De
nadruk ligt hier op de vraag hoe je het aanhouden van slack in balans kunt brengen met het
accepteren van extra orders. Daarbij kan kennelijk geaggregeerd worden over de al
geaccepteerde orders. Nog beschikbare levertijd telt even zwaar mee in de slack als fysieke
voorraad. Daarbij is wel belangrijk dat de orders in het orderboek allemaal dezelfde levertijd
hebben. Hoe je slack moet tellen bij orders met verschillende levertijden moet nader onderzocht
worden. Het is niet te verwachten dat daar eenvoudige regels voor zijn. Op basis daarvan is er
iets voor te zeggen in het design process model er maar van uit te gaan dat alle orders die niet
meteen geaccepteerd worden allemaal dezelfde levertijd hebben.
Met behulp daarvan kan dan een formal decision procedure afgeleid worden voor de beheersing
van de slack. Die laat nog ruimte bij het accepteren van niet verplichte orders: yield
137
management. Daar zou je nog een aparte formal decision procedure voor kunnen afleiden op
basis van een ander design process model.
Misschien dus toch wel een bruikbaar model voor een redelijk goed herkenbaar praktisch
probleem. Maar ik realiseer me dat er nog veel moet gebeuren. Het door Bertrand en Wortmann
(1981) ontwikkelde ontwerp voor de componenten fabriek is wat mij betreft nog steeds een
lichtend voorbeeld.
Referenties
Bertrand, J.W.M. en J.C. Wortmann (1981) Production Control and Information Systems for
Component Manufacturing Shops, Elsevier
Bertrand, J.W.M., Wortmann, J.C. en J. Wijngaard (1990) Production Control – A Structural
and Design Oriented Approach, Elsevier
Graves, S.C. en B.T. Tomlin (2003) Process flexibility in supply chains, Management Science
49, 907-919
Wijngaard, J. en F. Karaesmen (2007) On the optimality of order base-stock policies with
advance demand information and restricted production capacity, OR Spectrum, 29, 643 – 660
138
139
WILL BERTRAND AND DOUBLE
MATCHING QUEUES
Onno J. Boxma
Abstract The organ transplantation process is in this paper modeled as a double matching queue. The
customers are patients in need of an organ. The servers are organs, which arrive sequentially
and randomly. There is a waiting line of customers and a waiting line of organs; at most one of
the lines is non-empty. In each of the lines there is ‘impatience’: customers may leave the queue
because their health deteriorates, while organs cannot be preserved inde finitely.
We present an exact analysis of this queuing system under exponentiality assumptions, deriving
the queue length distributions, the waiting time distribution of patients, and the rates of
unsatisfied demands and of organ outdatings.
Introduction
It is an honour and a pleasure to contribute to the Liber Amicorum for Will Bertrand. Will has
played a very important role in strengthening the relations between the Department of Industrial
Engineering & Innovation Sciences and the Department of Mathematics and Computer Science
– and the group of Stochastic Operations Research (SOR) in the latter department has strongly
benefitted from this. When I joined the SOR group as successor of Jaap Wessels in 1998, I was
asked to join the management team of Beta. Will chaired those meetings, as scientific director of
Beta. Looking back, those meetings were more enjoyable, constructive and productive than
almost any other business meetings I have attended in various boards and committees. The
presence of persons like Ton de Kok and Henk Zijm inevitably led to very lively and creative
meetings, and Will was the person who always managed to keep us on the right track. His
persistent emphasis on quality turned Beta into a strong scientific research school, and has
raised the standards of Operations Management in Eindhoven.
Probably the most important item on Will’s agenda in those Beta meetings was to enhance real
interaction between persons and groups with (sometimes very) different backgrounds. All of us
were convinced that such interaction would bring considerable added value to Beta, but
140
realizing real collaboration between groups with quite different languages and interests turned
out to be extremely challenging. In the case of SOR, real collaboration with other Beta groups
has indeed been established; e.g., in the past few years there have been joint PhD projects with
Ton de Kok and Geert-Jan van Houtum (in particular, the PhD projects of Erik Winands, Josine
Bruin and Sandra van Wijk). Will personally played an important role in the thesis research of
Rene Bekker, a PhD student in the SOR group. In the huge queuing literature it is almost
exclusively assumed that the server works at constant speed. However, speaking from a long
experience, Will emphasized the fact that a human server in a factory does not work at constant
speed, and that the server speed depends on the workload. If the workload is low, the server
might slow down; if the workload is very high, the server might ‘panic’ and also slow down.
A medium workload would then lead to the highest server speed. Triggered by Will’s
statements and questions, Rene Bekker extensively studied queuing models in which the server
speed is a function of the workload [1, 2].
I am most grateful to Will for the above-sketched important role he has played in strengthening
(the interactions within) Beta. In addition, I would like to express my gratitude to Will for the
fact that our contacts invariably were pleasant, friendly and collegiate.
In this Liber Amicorum I would like to discuss – briefly and quite superficially, I’m afraid – the
stochastic process of organ transplantations. The phenomenon of waiting for organs is modeled
using a queuing system. It leads to an operations management problem that contains many
aspects which, hopefully, will appeal to Will: (1) it gives rise to an unconventional queuing
model, (2) it also contains aspects of an inventory model, (3) it is related to health logistics, and
(4) it requires a lot of modeling that could benefit from the interaction between researchers from
very different disciplines. Regarding the latter, I am benefitt ing from the expertise of Professor
Israel David on transplantations.
Much of the remainder of this introduction is based on, or even taken from, a joint paper with
David, Perry and Stadje [3]. It focuses on the modeling aspects of the organ transplantation
process.
Organ transplantations and double matching queues
The problem of managing a list of patients waiting for transplantation has attracted the attention
of operations researchers from the mid eighties [4, 5, 6, 7, 8, 9, 11, 12, 15]. Zenios et al. [13]
contains an excellent introduction to the modeling of live-organ transplantations by means of a
waiting list. The problem is a hybrid of queuing and inventory aspects, the inventory being a
collection of organs ‘on the shelf’. Considering patients who suffer from organ failure (kidneys,
livers, hearts, etc.) and who register on a transplantation list as ‘customers’ waiting for service,
one is tempted to take recourse to queuing models, but several non-standard features need to be
141
taken into account. First, the ‘servers’ are organs usually donated from the dead; they arrive se-
quentially and randomly, and the transplantation itself only takes a negligible time. One thus
faces a double matching queue whose two lines (one of organs and one of patients) can be both
empty but, normally, are not both non-empty simultaneously. Second, both queues are affected
by influences which may cause reductions in their lengths without matching’s taking place. One
major issue is the health deterioration of waiting patients. In the language of queuing theory,
one speaks of impatience, a random positive time-lag, which is assigned to any arriving
customer (patient). If a customer’s ‘patience’ runs out before being served, he reneges. A
second, dual, major issue is that live organs such as kidneys or livers cannot be preserved for
more than a certain period of time (often about two days), since freezing them is not possible.
This establishes a link with stochastic inventory theory where one speaks of a perishable
inventory system (PIS) whose output process is split into satisfied demands and outdatings. Ob -
viously, the two types of untimely departures – impatience and outdating – are not symmetric.
In [3] a model is proposed that captures in full the double-queue nature of the problem and takes
health deterioration (customer impatience) as well as organ outdating (server removal) into
account.
In the basic model presented there, a FIFO regime of patients is assumed and it is shown how
tools available from queuing theory can be used to derive explicit results regarding the
processes of unsatisfied demands, of outdated organs and of waiting patients , and other
important information about the efficiency of the issuing process.
The starting-point in [3] is the age process of the model, viz.: (i) if the shelf is not empty, the
age process represents the evolution of the age of the oldest organ on the shelf; (ii) as long as
the shelf is empty, the age process represents minus the waiting time of a (virtual) arriving
patient whose patience would be long enough to eventually receive the first available organ.
Notice that age here can be negative, hence age should be interpreted in a broad sense. A key
idea is to flip its graphical representation over by 180 degrees, yielding the virtual outdating
process (VOP). The VOP provides the time until the next outdating of an organ if the demand
process is stopped right now. The VOP leads to the key performance criteria of the model. A
second key idea in [3] is to note the analogy of the VOP with the workload process in a certain
M/G/1-type queue with impatience. Analysis of the latter queue leads in [3] to explicit results
for important performance criteria of the organ-transplant system, for the special case of
deterministic organ outdating: organs last a fixed amount of time.
In the next section, we provide an exact solution for the purely Markovian case in which organ
142
interarrival times, customer interarrival times, outdating times and customer patience times are
all independent and exponentially distributed.
The model in [3], as well as the model in the present paper, may be viewed as prototype models
that provide a good first approximation to reality. Much further research is needed, though.
Below we question (1) the fact that there may be a non-empty queue of organs on the shelf, and
(2) the assumption that organs are issued to customers in order of customer arrival.
Ad (1). In most Western countries, the line of organs on the shelf is empty most of the time; a
longer line of waiting patients gives rise to a larger rate of abandonments (deaths) and hardly to
a larger rate of transplantations. However, this may change, e.g., due to new legislation. In
several Asian and South-American countries, permission from the families of potential donors is
not required, leading to a much larger arrival rate of organs and a more ‘balanced’ double
matching queue.
Ad (2). In most organ transplant situations, the issuing policy of organs is not FIFO. For
example, the conditions of the patients may play a role, and the level of matching; e.g., there
may be a mismatch between blood types of donor and potential recipient. However, for a large
subset of the population of patients, a FIFO policy still provides a good representation.
1. Analysis We consider the following model. Perishable items (donated organs) arrive at the organ bank
according to a Poisson process with rateλ . The arrival times of the demands (patients needing
transplantation) are independent of those of the items (in the case of kidneys this may not be
quite true) and form a Poisson process, of rate µ .
A demand that upon its arrival finds the shelf of items not empty is satisfied immediately by the
oldest organ present. Demands that arrive at an empty organ bank join the line of waiting
demands (if any); newly arriving organs are assigned on the spot to waiting demands on a first -
come-first-served basis.
Each demand possesses its own exponential patience time, which can be interpreted as the time
until the patient’s physical condition no longer allows carrying out a transplantation. These
patience times are assumed to be independent, identically distributed, random variables which
are independent of the arrival times of organs and demands. We assume them to be
143
exponentially distributed with rate 0η > . If the waiting time of a patient exceeds his patience he
abandons the waiting line without receiving treatment. The ‘shelf lifetimes’ of the stored
organs, i.e., their maximum usage times, are independent, identically distributed random
variables, again independent of all other processes. We assume them to be exponentially
distributed with rate 0γ > . Each organ is stored until it either satisfies some demand or, after
some time on the shelf, is outdated (and then scrapped).
According to the above description two connected queuing systems of FCFS type are generated.
The first queue consists of the stored organs on the shelf while the second one is given by the
line of patients waiting for these organs. The two queues cannot be simultaneously non-empty,
but it is possible that both of them are empty.
Let { }, 0tX t ≥ be the number of waiting customers, if there are no organs, and let it be minus the
number of organs on the shelf, if there are no customers; 0tX = corresponds to the case that
there are neither organs nor customers, tX j= corresponds to the case of j waiting customers,
1,2, ,j =
tX k= − corresponds to the case of k organs on the shelf, 1,2,k = .
Due to the exponentiality and independence assumptions, { }, 0tX t ≥ is a Markov process; it
even is a birth-and-death process. The presence of outdating and impatience implies that the
process { }, 0tX t ≥ is ergodic for all parameter values. We shall determine the steady-state
distribution
( )( )
( )0 0
: lim , 1,2, ,
: lim , 1,2, ,
: lim 0 .
j t t
k t t
t t
p X j j
q X k k
p q X
→∞
→∞
→∞
= = =
= = − =
= = =
PP
P
The balance equations for the process are:
( )( )
1
1
, 1,2, , (1)
, 1,2, , (2)j j
k k
p j p j
q k q k
µ λ η
µ µ γ−
−
= + =
= + =
It follows that
144
01
01
, 1,2, , (3)
, 1,2, , (4)
j
jm
k
kn
p p jm
q q kn
µλ ηλ
µ γ
=
=
= =+
= =+
∏
∏
and with the normalization condition 0 1 11j kj k
p p q∞ ∞
= =+ + =∑ ∑ it follows that
[ ] 1
0 01 1 1
1 . (5)j k
j m n
p qm nµ λ
λ η µ γ
∞−
= = =
= = + ++ +∑∏ ∏
Having determined the steady-state distribution of the ‘queue length’ process, we now turn to
various performance measures which are key characteristics of the efficiency of the organ bank:
1. the waiting time of customers;
2. the rate of the times at which demands leave unsatisfied;
3. the rate of the times of organ outdatings.
1. The waiting time of customers.
Let W denote the steady-state waiting time of a tagged customer with the special feature that this
one customer has infinite patience. First we observe that, by PASTA, ( ) 10 kk
W q∞
== =∑P .
Secondly, again by PASTA, an arriving customer sees j waiting customers with probability
, 1,2, ,jp j = and sees neither a customer nor an organ with probability 0p . Let jW denote the
waiting time of such a tagged customer who sees j customers, 0,1,j = . Then jW is the sum
of 1j + independent random variables: 0 1j jW Z Z Z= + + + , where iZ denotes the time it
takes until the number of customers ahead of the tagged customer reduces from i to
1, 1,2,i i− = , and where 0Z denotes the time till the arrival of an organ in an empty system. It
is readily seen that iZ is exponentially distributed with rate iλ η+ , because it is the minimum
of an exp ( )λ organ interarrival time and i patience times of customers, all of which are exp ( )η .
Hence the LST (Laplace-Stieltjes transform) of jW is given by
0
0
. (6)j jj
sW sZ Z
i
ie ei s
λ ηλ η
− − + +
=
+ = = + +∏E E
145
One can easily invert this expression, obtaining a weighted sum of exp ( )iλ η+ terms; cf. also
Section 5.2.4 of [10].
So far we assumed that the tagged customer has infinite patience. Let us now drop this
assumption, and calculate the probability that the tagged customer becomes impatient. The
probability that the tagged customer becomes impatient before being served when meeting j
customers upon arrival, 1,2,j = (and for 0j = the probability of becoming impatient before
being served when finding zero organs and zero customers) equals jWe η− E . Indeed, with H
denoting the (exponentially distributed) patience time of the tagged customer, one has
( ) ( )0
.jWntj jt
W H e d W t e η∞ −−
= < = < = ∫P P E
From (6),
( ) ( )0
. (7)1 1
jj
W
i
iei j
η λ η λλ η λ η
−
=
+ = = + + + +∏E
The interpretation of ( )0W H λλ η
< =+
P is trivial: 0W H< if an organ arrives before the
patience of the tagged customer runs out. The interpretation of ( ) 2jW H λ λ ηλ η λ η
+< =
+ +P is
obtained by observing that the patience of the tagged customer should first exceed the minimum
of an organ interarrival time and the patience time of the customer ahead of him (which has
probability 2
λ ηλ η++
, and subsequently the remaining (still exponential) patience of the tagged
customer should exceed an organ interarrival time. One can now similarly interpret
( ) ( )0.
1j
j i
iW Hi
λ ηλ η=
+< =
+ +∏P
It hardly requires an extra effort to derive an expression for the LST of jW given that jW H< .
First observe, with ( )A denoting the indicator function of event A:
( ) ( ) ( )0
. (8)j jsW s Wnt stj jt
e W H e e d W t e η∞− − +− −
= < = < = ∫E P E
Hence
( ) ( )0
, (9)1
jj
sWj
i
ie W Hi sλ η
λ η−
=
+ < = + + +∏E
which in combination with (7) yields:
146
( )( )0
1. (10)
1j
jsW
ji
ie W H
i sλ η
λ η−
=
+ + < = + + +∏E
2.. The rate of the times at which demands leave unsatis fied
Let us denote by µ the rate of unsatisfied demands. Clearly,
( ) ( )( )
( )( )0
0 0 0 1
1 1* . (11)
1 1
j
j j jj j j m
j jp W H p p
j m jη ηµµη λ λ η η λ
∞ ∞ ∞
= = = =
+ + = > = = + + + + + ∑ ∑ ∑ ∏P
3. The rate of the times of organ outdatings
Let us denote by *λ the rate of organ outdatings. That outdating rate immediately follows from
the following conservation law:
* *. (12)λ λ µ µ− = −
To see (12) note that since λ is the arrival rate of organs into the system, it is also their
departure rate out of the system in steady state. Thus, the left-hand side of (12) represents the
long-run average rate of organs which are not outdated, i.e., are used to satisfy demands, while
the right-hand side of (12) is just the rate of satisfied demands. As a result, once we have
computed *µ , the rate *λ is also known.
References
[1] R. Bekker (2005) Queues with State-dependent Rates. PhD Thesis, Eindhoven University
of Technology.
[2] R. Bekker, S.C. Borst, O.J. Boxma and O. Kella (2004) Queues with workload-dependent
arrival and service rates. Queuing Systems 46, 537-556.
[3] O.J. Boxma, I. David, D. Perry and W. Stadje (2010) A new look at organ transplantation
models and double matching queues. EURANDOM Report 2010-013; to appear in
Probability in the Engineering and Informational Sciences, 2011.
[4] David, I. and U. Yechiali (1985) A time dependent stopping problem with application to
live organ transplants. Operations Research 33, 491-504.
[5] I. David and U. Yechiali (1990) Sequential assignment match processes with arrival of
147
candidates and offers. Probability in the Engineering and Informational Sciences 4, 413-
430.
[6] J.C. Hornberger and J. H. Ahn (1997) Deciding eligibility for transplantation when a
donor kidney becomes available. Medical Decision Making 17, 160-170.
[7] C.H. Lee, G.M. Chertow and S.A. Zenios (2008) Optimal initiation and management of
dialysis therapy. Operations Research 56, 1428-1449.
[8] R. Righter (1987) The stochastic sequential assignment problem with random deadlines.
Probability in the Engineering and Informational Sciences 1, 189-202.
[9] R. Righter (1989) A resource allocation problem in a random environment. Operations
Research 37, 329-338.
[10] S.M. Ross (2007) Introduction to Probability Models. 9th ed. Academic Press, London.
[11] X. Su and S.A. Zenios (2006) Recipient choice can address the efficiency-equity trade-off
in kidney transplantation: a mechanism design model. Management Science 52, 1647-
1660.
[12] S. Ur, M. Trick and D. Sleator (1992) Analysis of online algorithms for organ allocation.
In: J. van Leeuwen (ed.), Algorithms, Software, Architectures. Information Proceedings
92, Vol. I, Elsevier (North-Holland), 458-464.
[13] S.A. Zenios, G.M. Chertow and L.M. Wein (2000) Dynamic allocation of kidneys to
candidates on the transplant waiting list. Operations Research 48, 549-569.
[14] S.A. Zenios (1999) Modeling the transplant waiting list: a queueing model with reneging.
Queueing Systems 31, 239-251.
[15] S.A. Zenios (2002) Optimal control of a paired-kidney exchange program. Management
Science 48, 328-342.
148
149
ON THE ONSET OF WEAK MONOTONICITY
RESULTS ON LATTICE FRAGMENTS
Andreea B. DRAGUT
Abstract Monotonicity and submodularity properties are insights into the structure of the optimal
decision policies. They reduce the exponential growth of the dynamic programming (DP)
formulation when looking for the optimal policies of a Markov Decision Process (MDP). In this
paper we describe an R&D control problem that led to the extension of the previously known
monotonicity results for the case of state and action spaces being fragments of lattices.
Keywords:
R&D tasks, discrete Markov decision process, monotonicity, supermodularity
1. Introduction
This article is a tribute to professor dr.ir. J. Will M. Bertrand, or simply Will Bertrand for the
people who worked with him.
Will Bertrand’s research is situated in the realm of the control of operational processes. In the
early years, operational and operations initial research was mostly based on quantitative
modeling, with the main purpose of rather solving real life problems than developing theoretical
scientific knowledge.
Nevertheless, one of the big impacts of the industrial use of simple mathematical models,
together with the increasing presence of information systems, was to reshape the organizations
and eliminate traditional human tasks. At the same time, the idealized research approach
neglected the creations of new functions and responsibility areas for humans in operational
processes planning and control.
As a consequence, very early in the development, idealized technical, logistical and
mathematical problems became the main focus of research in this field. Throughout all his
150
working years, Will Bertrand however always kept emphasizing the modeling aspect:
operational processes need to have explanatory and predictive models carefully and adequately
developed. Through his empirical research, he was constantly showing that the existing
idealized models failed to satisfactorily help implementing real-life solutions, rendering the
whole process painfully complex and often unsuccessful. Will Bertrand found that one of the
main shortcomings of those existing idealized models was the failure to properly take into
account the human factor implications at all planning and control levels.
In fact, all planning, scheduling and control processes have always been depending on both
technological and human assets of an organization. Of course, operational process models were
not entirely neglecting the human component, yet their approaches were greatly limited. They
constrained humans to always behave deterministically, predictably, in a ’stationary’ and
emotionless fashion, and independent of each other.
What exacerbated the abovementioned shortcomings, rendering them acutely critical, was that
in the last two decades customer and supplier market lost resource stability, acquired increased
dynamics, while at the same time competition and process complexity also greatly increased.
Therefore ignoring so many relevant factors was hampering the quest for more responsiveness,
leading to increasingly critical competitive disadvantages.
All these circumstances led to a sea change in the operational processes research field, of which
Will Bertrand was an integral part: he was constantly widening his operational control horizons,
and contributing to the knowledge development from the perspective of the human engineer,
scheduler and developer.
Let us take a closer look at these aspects. In general, in this field as of course in many others,
research methodologies span techniques from empirical research and simulation-based research
to formal analysis of production control models. Moreover, the empirical research is usually
doubled by the development of mathematical tools whenever the need arises. Will Bertrand’s
wide research encompassed both empirical and mathematical models, as a bridge between these
two streams. PhD theses supervised by him which were focusing on empirical research
identifying factors influencing operational processes, were then followed by more mathematical
ones attempting to create solvable models based on these factors.
Interestingly enough, this pattern of pairing up showed e.g. with his last PhD students: Wenny
Raaijmakers and then Cristina Ivanescu, and also Kim van Oorschot and Andreea Dragut. On a
general note, we must point out that it was never easy to go along this research path: there were
constant critics from both sides. Nevertheless, the complexity of processes ended up being
151
brought to light especially thanks to both types of research studies. They, on the one hand, made
clear which characteristics were being neglected during mathematical modeling, and on the
other hand required mathematicians to thoroughly justify the reasons behind all such neglects.
However, in a world of publish or perish, Will Bertrand made an honor point of never signing a
pure mathematical paper, while he provided the insight of real-life problems for an entire series
of papers that his PhD students signed with mathematical researchers, such as Jaap Wessels, Jan
van der Wal, and Ivo Adan. I would also like to mention, on a very personal note that for my
own PhD thesis Will Bertrand encouraged me to slightly extend the mathematical theory of
lattice programming, in order to include applications for R&D control. It turned out, as
expected, not to be an easy path. Yet, a number of years later, these weak monotonicity results
are now considered a part of the common body of mathematical techniques of lattice
programming for the Markov decision processes, being reviewed in the new Wiley
Encyclopedia of Operations Research and Management Science (EORMS) (see Dragut 2011).
The goal of this paper is to describe the R&D control problem that led to the extension of the
previously known monotonicity results for finite discrete-time Markov decision processes. It is
Will Bertrand who suggested the investigation of the optimal control policies for this problem.
In general, monotonicity and submodularity properties are insights into the structure of the
optimal decision policies. They reduce the exponential growth of the dynamic programming
(DP) formulation when looking for the optimal policies of a Markov Decision Process (MDP).
If the action spaces are real sublattices, monotonicity often allows for feasible numerical
solutions (Topkis 1998, Section 7.9 in Puterman 1994, p. 210) for discrete-time
multidimensional MDPs.
Nevertheless, control problems often exhibit only general partially ordered sets as state/action
spaces. If we strengthen the assumptions and we require the action space to be a lattice, the
classical monotonicity results can analytically characterize the optimal solutions of such
problems. However, they lead to an undesirable large increase of the state/action space, possibly
canceling the computational gains brought by monotonicity. (see Dragut 2006).
By setting up a new mathematical framework in which monotonicity results hold for state and
action spaces which are fragments of lattices, we can analytically characterize and also
numerically study the structure of optimal control policies for our R&D control problem, as well
as for new classes of problems.
152
This paper is organized as follows. In Section 2 we discuss the operational characteristics that
led to a Markov decision process (MDP) to model R&D projects with concurrent R&D tasks. In
Section 3 we prove the existence of robust weakly monotonic nondecreasing optimal policies
and we describe the optimal value variation function of the degree of specification of the
characteristics of the new product in the beginning. In Section 4 we end with some final
remarks.
2. Operational Characteristics of R&D projects
The aim of this section is to bridge the gap between R&D project operational characteristics and
the formulation of a quantitative model for the dynamic performance control of such a project.
For a quantitative model, we are forced to make some simplifications of the real-life project. A
better mathematical understanding of the structure of the optimal strategies for our simplified
model emphasizes relationships between model variables. It might also be helpful in more
complex situations where good heuristics are needed.
2.1. Creative human activities The work on a single creative activity can actually be seen as a series of steps. In the so-called
insight problem solving model (Wallas 1926, Smith and Blankenship 1999, and Sio and
Rudowicz 2007), the individuals start working on a problem, which results in some progress
(the preparation stage).
Rarely the problem might be solved at once. In most cases the individuals either reach a
cognitive state of fixation and the problem will remain unsolved, or after some incubation time,
with a sudden insight, they might solve the problem. There is thus a memoryless property in any
creative activity since the amount of time one spends on a task does not bring closer the
solution. It seems thus natural to model a creative activity as an exponentially distributed
random variable. In the following model we will see an R&D task as a sequence of creative
activities to be solved.
153
2.2. Operational Characteristics of R&D Projects
As in most production environments, both time and capacity are important. What differentiates
the R&D projects from other production projects are the creative type of work to be done, the
use of engineers' time as the main resource, and the fact that it is impossible to execute all (or
even the majority of all) R&D tasks of an R&D project within reasonable time, let alone more
projects.
Work definition and execution
Every project induces one or more concurrent R&D tasks to be solved before a given deadline.
In R&D, information is lacking with regard to what exactly needs to be the end product, and
how it should be produced (Britton and Glynn 1989). Thus, the exact amount and type of work
contained in an R&D task is typically unknown beforehand. To define an R&D task, we use a
definition similar to (Beeftink et al. 2008, Amabile 2005): a series of creative work activities in
which novel and potentially useful ideas are required to solve ill-defined problems. Ill-defined
problems are problems for which a clear path to solution is unknown (Sternberg and Davidson
1999) and for which only a creative problem solving approach might lead to a solution (Cropley
1999).
Task uncertainty makes planning ahead difficult. Thus, the planning starts with partially
described tasks: for each task n a number of planed activities is foreseen. Working on an R&D
task helps an engineer to gain more knowledge about the task and finally to fully describe all the
activities needed to accomplish different performance levels (van Oorschot et al. 2005).
Usually, the amount of work-in-process varies heavily over time, and the moments at which
work arrives are often so random that the arrival process of work can be described as a Poisson
process. More precisely, we consider that new activities arrive at each of the tasks in progress
according to a Poisson process. All activities planned or new are creative activities and they are
modeled as an exponentially distributed random variable.
Quality of Service
Any limitation of the capacity assigned to a task is negatively correlated with the quality of
service. This is a performance measure that indicates the quality of the actual outcome of a case
compared to the most favorable outcome achievable for that task.
We consider that each task n has different levels of performance, giving the quality of its
execution. For each task n, a minimal performance level lmin(n) has to be achieved, in order to
have a rewardable fully functional new product. For some tasks this minimal level can be zero.
154
Each level l consists of a list of planned activities with solving times being (i.i.d. exponentially
distributed) random variables. To attain a level, we assume that the engineer has to sequentially
execute the task at all previous levels. To enable an easy computation of the remaining solving
time Sn of the task n, we assume that for any given task n, its levels will have an identical
number of activities na(n) per level.
Decisions concern how to employ the available capacity, i.e., how much capacity to assign to
the work-in-process and how much for the new work. They are taken in terms of which tasks to
solve and to what extent. As long as the R&D project ends up with a fully functional product,
the capacity decisions are depending mainly on the work on the one hand, and the benefit of
carrying on with work-in-process on the other.
Even in standard production environments the advantages of increasing the pace of work—by
working under schedule pressure—can be offset by losses in productivity and quality. As
pointed out in the empirical study of (Nepal et al. 2006), the negative effects of schedule
pressure arise mainly by working out of sequence, generating work defects, cutting corners, and
losing the motivation to work. The effects on human beings are also supported by medical
research (Melchior et al. 2009).
We consider that the actual processing capacity is mainly given by the cumulated time of the
team of engineers. Our time capacity control ideas build on the idea of workload regulation as
described in Activation Theory (Gardner, 1986; Gardner and Cummings, 1988). This theory
suggests
that there is a certain optimal level of workload (neither too high or too low) at which
individuals perform best, and
that individuals try to influence their workload towards the level at which they perform
best. (Gardner, 1986, Gardner et al. 1988)
We can estimate the required processing capacity if all R&D tasks involved with all arriving
case-inducing work were to be executed, using the decomposition of R&D tasks. We can limit
the work under schedule pressure by requiring at each decision point that the remaining
workload should not exceed the engineers’ time capacity.
Under the hypothesis of a deficient time capacity we do not need to ensure that the team of
engineers has enough work to do as long as the economic benefit of working exists. The benefit
of working can be modeled straightforwardly by using reward functions depending on the
performance levels of each task.
155
Deficient resource capacity control
Despite a range of possible actions, such as the introduction of overtime, temporary staff
employment and multi-skilling of staff, the resource capacity will commonly be insufficient to
execute all (or even the majority of all) tasks of an R&D project, let alone more projects .
For the following problem description:
One team of engineers is confronted with a set of concurrent R&D tasks. These tasks are
available simultaneously or can arrive over a specific planning horizon. The performance level
up to which the team will execute each task must be decided such that some performance
criterion is optimized.
three essential on-line decisions must be taken in scant-capacity operating environments:
1. admission of new work, and
2. continuation or termination of work-in-process.
3. cancellation of work
Based on the previous decomposition of R&D tasks, they can be formulated in terms of
performance level control. Continuation of an R&D task can be seen as setting as a goal a
higher performance level than the currently achieved one. Termination of an R&D task appears
when the achieved performance level is judged to be sufficient. Cancellation occurs when the
termination decision is taken for an R&D task having the achieved performance level the level
zero. Thus, any new work can be admitted by default and subsequently, if needed canceled.
Next we will set up a mathematical framework, in which we can formally take the decision on
which performance level has to be achieved for each R&D task. This leads to a new class of
decision models, which fits naturally within the field of MDP control. Further, we study the
structure and characteristics of the optimal control policies obtained.
In the next section we propose a fairly straightforward mathematical control system formulation
of a R&D project in terms of a Markov decision process.
3. Mathematical Control System
In our control system an R&D project has a number T (from 0 to T-1) of review periods and N
concurrent design tasks allocated to a team of M engineers from the first review period.
156
Each task n has different levels of performance, giving the quality of its execution. For each
R&D task n, a minimal performance level lmin(n) has to be achieved, in order to have a
rewardable fully functional new product.
Each level l consists of a list of planned activities with solving times that are (i.i.d.
exponentially distributed Exp(µ)) random variables. To attain a level, we assume that the
engineer has to sequentially execute the task at all previous levels.
New activities arrive at each of the tasks in progress during the review period t according to a
Poisson process of rate λ(t). All activities planned or new are solved with a rate μ .
For simplicity, for any given task n, its levels will have an identical number of activities na(n)
per level. Thus, the remaining solving time Sn of one level of the task n, assuming the previous
levels solved, is a r.v. Erlang(na(n), μ).
Other input parameters (global variables) for each n from 1 to N:
Lmax(n) : the task’s maximal number of performance levels;
amin(n) : minimal work (in term of levels) to be done in a review period;
(at the beginning of review period t ):
β(t): the required current safety margin; β(t) is a real positive less than 1;
l(n,t) :the achieved performance level for the task n
The decision time points: are the project’s review points t
The immediate rewards ρt t=0,T-1 are zero. The final reward is zero if the minimal
required levels are not acquired in the final state. Otherwise, ρT(xT) is a nondecreasing function,
which is typical for cumulative market payoff.
The state space and the action space: The state set X(t) and the action set At(xt) in the
state xt give the achieved, respectively target performance level of each design task n such that
the remaining workload of the team of engineers should not exceed their maximal solving
capacity with a probability greater than safety margin β(t) .
For t=0 the only state is zero, otherwise the state xt =(l(1,t),…,l(N,t)) describes what was
solved for each design task n. If t<T we ask a feasible state to achieve a functional (minimal
levels) product with a safety probability
( )
( )( )
−≤∑∑
+==tTMSn
nl
tnli
N
n
min
1,1Pr .
157
For any t<T, the action at in the state xt decides how many other levels above xt we want to
perform. The level up to which the task n may be solved after the feasible action at was taken is
l(n,t)+at(n) such that
( ) ( )ttTMSn
na
i
N
n
t
β≥
−≤∑∑
==
)(
11Pr .
The transition probabilities:
The transition probabilities describe the behavior of the R&D project between two control
points. In this control system the probability that after taking the action at in the state xt, a
transition from state xt to state xt+1 occurs is given by the probability of solving in one unit of
time (which is actually the length of a review period [t,t+1)) the activities belonging to the
performance levels newly achieved in state xt+1. More precisely, the transition probability is
given by probability that the remaining solving time of the tasks in state xt up to the
performance levels from xt+1 is contained in an interval of length ε around the length of a
review period [t,t+1).
We need a low computational model preserving a maximum of features of the underlying
working R&D process.
One classical approach is to allocate the tasks to engineers and afterwards to derive a complete
schedule for each engineer. In the scheduling theory, very often the engineers (resources) are
viewed as individual servers working on R&D tasks. Some human factors might be considered
as: learning effects (Biskup, 2008, Gordon et al. 2008), deteriorating jobs (see for an extensive
survey Bachman 2002 and Cheng et al. 2004, 2005, Alidaee and Womer 1999), a rate-
modifying activity playing the role of the insight flash (see Lodree and Geiger 2010, Min Ji and
Cheng 2010) and even polynomial optimal allocation can be derived for certain classes of R&D
tasks completion times.
The main advantage of such an approach is its relatively low computational complexity, but this
view on the R&D project suffers severely from modeling issues. Usually an engineer has a
number of R&D tasks to solve and is allowed to switch back and forth between tasks at
moments chosen at her/his own discretion. Also, very seldom an engineer can work alone
without any exterior influence on his R&D tasks. Engineers share information (as for example
technological breakthroughs) and they suffer interruptions in order to pursue as a team the
progress of the entire NPD project.
158
Moreover, due to both technological and financial restrictions, it is rarely the case that the total
engineers’ time capacity is sizeable enough to be capable of executing all work ordered to the
system. Thus, even for the sake of taking decisions on what work has to be done, a complete
workshop schedule might be superfluous and might require an unnecessary effort.
We argue that more features of the working process can be captured by letting the engineers
(resources) act as a pool. This means that work is not allocated to a specific engineer (resource)
in advance, but rather shared by the team of engineers (resources) on a dynamic basis. If their
joint capacity is insufficient to do all work ordered to the system, then at review time decisions
must be made as to when to stop/cancel R&D tasks, and which tasks to stop/cancel. These
decisions will depend on the number of R&D tasks already in the system and the progress the
team of engineers done in solving them.
To make such a pool model computationally tractable we need to make simplifying assumptions
about the patterns of switching from one R&D task to another, as well as about what
phenomenon might occur while working on R&D tasks.
The empirical studies done in task pacing literature present a series of different patterns in
which work efforts can be distributed over time over the course of a task or project. Their
common feature (Sonnentag 2003, Beeftink et al. 2008) is that while working on a creative task
human beings tend to switch from moments of being highly engaged in their work to moments of
being engaged. While working they can experience interruptions (the need to switch the current
task at an externally determined moment in order to perform either new activities or activities
belonging to a different task) and breaks (the possibility of switching between already existing
tasks at their own discretion).
Thus, this stream of research suggests another empirical argument against a detailed workshop
schedule since imposing one leads to unnecessary interruptions. While breaks are helping
human beings to solve more insight activities and reaching fewer impasses, interruptions affect
them negatively (Zijlstra et al. 1999, Eyrolle and Cellier 2000, Bailey and Konstan 2006).
However, some interruptions, as the ones related to tasks’ interdependence cannot be avoided.
The simplifying assumptions made in Dragut and Bertrand 2008 concerned both a task pacing
behavior as well as the switching between tasks. The task pacing behavior considered was that
at any time instant the team either worked with the probability p on the planned activities of the
R&D tasks, or, with a probability 1-p, suffered interruptions.
To account for the engineers’ behavior of switching at their own discretion between the existing
R&D tasks all the tasks are assigned to the team and no detailed schedule is provided.
159
Analytically such a model can be also low computational if we consider:
• the team of engineers working as a whole on the performance levels of the R&D tasks
indicated by the action at
• a common granularity for the decomposition of any task, and thus a uniform measure of
the magnitude of an R&D task (namely the number of activities involved).
We recall that each task n has different levels of performance, giving the quality of its
execution. Each level l consists of a list of planned activities with solving times random
variables i.i.d. exponentially distributed Exp(µ). To attain a level, we assume that the engineer
has to sequentially execute the task at all previous levels. New activities arrive at each of the
tasks in progress during the review period t according to a Poisson process of rate λ(t). All
activities planned or new are solved with a rate μ.
Under these conditions the completion time of the work allocated to the team (i.e. the
makespan) of a queuing system with M engineers (resources) working together is the same no
matter the order in which the engineers might decide to do their activities.
In Dragut and Bertrand 2008 this queuing estimation model was tested successfully on real life
data from engineers working on R&D tasks.
Proposition 1 Given the queuing estimation model from Dragut and Bertrand 2008, the sum of
the transition probabilities p(xt,at,xt+1) taken over all the states in X(t+1) is less than one, and
thus they can be considered as transition probabilities in a Markov process.
Proof. See appendix.
This proposition derives a way of computing transition probabilities which satisfy the Markov
condition, i. e. the sum of pt(xt , at , xt+1) taken after all xt , xt+1 in X(t+1) is less or equal to one.
Thus, for an R&D project we obtain a control system which is Markov Decision Process having
partially ordered state space, and action space. Its action and state space are upper bounded
inf-sublattices of the lattice NN . The action space description may lead to the existence of
unordered points in the argmax set, as well as in its top set (see Topkis 1998 for lattice basics).
If we enlarge the action space to a lattice instead of a partially bounded subset, for all these
unordered actions there will exist a greater action that will be optimal. Thus, classical
monotonic optimal solutions in the sense of Topkis 1998 can be derived.
However, in this case the sought after computational gains do not occur (see Dragut 2003,
Dragut 2006) due to the large increase of the action space.
160
An alternative solution which was strongly supported by Will Bertrand was to extend
monotonicity results for our type of action and state spaces, namely, subsets without holes of
vector lattices (see Dragut 2003, 2011 for a formal definition). This was done in Dragut 2004,
Dragut 2011. Using this slight extension of the mathematical theory one can obtain robust
weakly monotonic optimal policies under the conditions of the subsequent Theorem from
(Dragut 2004, Dragut 2011).
4. Results
Theorem If the state/action spaces are bounded subsets without holes and the family
{ })(|)( tXxxA ttt ∈ is expanding and ascending for the MDP model, there exist weakly
nondecreasing optimal decision rules if the rewards are nondecreasing and superadditive as
well as the sum of trasition probabilities taken over any increasing subset of X(t). (Dragut 2003,
Dragut 2011)
Proof. See appendix.
In microeconomics and in theories of production and consumer choice, supermodularity of a
utility function is equivalent to products being substitutes. Thus, having a monotonic type of
optimal policy analytically proves that in R&D projects after we ensured the achievement of a
functional final product, having done more levels of performance for one R&D task
compensates for doing less performance levels of a different one.
The extension of the mathematical theory served its declared purpose of making computational
tractable the derivation of optimal policies of the control system. Thus, in Dragut 2003 we could
compute the weakly monotonic optimal policies and we could conduct a series of simulation
studies.
We have considered that the characteristics of the new product are underspecified/over specified
if we have low/high values for the number of planned activities, and vice versa for the rate of
arrival of unplanned activities during solving process. We synthetically generated large number
data sets with tasks of variable sizes for each µ. The simulation suggested that if the product is
less specified at the beginning, the optimal value increases. This effect diminishes with the
increase in the solving rate activities, i.e. the case of a better or experienced team. This supports
the idea that because of the time pressure, R&D decisions must often be made quickly with only
partial information available. Early involvement in the development is better, especially for
161
inexperienced teams of engineers. Uncertainty and under specification of product features are
inherent.
Also in general, the weakly monotone structure of an optimal policy can serve as a basis for
deriving heuristics. In the case of our control system a trivial way of doing it is to cut down on
the action space. The state and action spaces are large. However, one can show that many
actions are surely sub-optimal and can therefore be omitted from the model. In other words, if it
is optimal for a team to continue working on a task for which its minimal level is already
achieved, then this decision is optimal as well for a number of tasks of equivalent value. From a
numerical point of view, for any given state, the size of the action space can essentially be
reduced with an exponential factor.
5. Conclusion
Professor Will Bertrand suggested the investigation of optimal control policies for R&D
problems, during a joint research work on their appropriate modeling framework.
In this paper, we have shown how this investigation actually led to the extension of the
previously known monotonicity results for finite discrete-time Markov Decision Processes.
Usually control problems often exhibit only general partially ordered sets as state/action spaces.
By strengthening the assumptions to obtain full lattices for these sets, we can use classical
monotonicity results to analytically characterize the optimal solutions, but at the expense of
large increases in state/action space sizes, possibly offsetting computational gains.
We have shown how to set up a mathematical framework in which monotonicity results hold for
state and action spaces which are only fragments of lattices. We have thus shown how we can
analytically characterize and also numerically study the structure of optimal control policies for
our R&D control problem, as well as for new classes of problems.
References
1. Alidaee, B., Womer, N.K. (1999). Scheduling with time dependent processing times:
review and extensions. Journal of the Operational Research Society, 50, 711–720.
2. Amabile, T. M., Barsade, S. G., Mueller, J. S., and Staw, B. M. (2005). Affect and
creativity at work. Administrative Science Quarterly, 50(3), 367-403.
162
3. Bachman, A., Janiak, A., Kovalyov, M.Y. (2002). Minimizing the total weighted
completion time of deteriorating jobs. Information Processing Letters, 81, 81–4
4. Bailey, B. P., Konstan, J. A. (2006). On the need for attention-aware systems:
Measuring effects of interruption on task performance, error rate, and affective state.
Computers in Human Behavior, 22(4), 685-708
5. Beeftink, F., Van Eerde, W., and Rutte, C. G. (2008). The effect of interruptions and
breaks on insight and impasses: Do you need a break right now? Creativity Research
Journal, 20, 358-364.
6. Berglunda, M. and Karltun, J. (2007). Human, technological and organizational
aspects influencing the production scheduling process, International Journal of
Production Economics 110, 160–174.
7. Biskup, D. (2008). A state-of-the-art review on scheduling with learning effects,
European Journal of Operational Research 188,315–329.
8. Britton, B. K., and Glynn, S. M. (1989). Mental management and creativity: A
cognitive model of time management for intellectual productivity. In J. A. Glover & C.
R. Reynolds (Eds.), Handbook of creativity (Vol. 24, pp. 429-440). New York: Plenum
Press
9. Cheng, T.C.E., Kang, L.Y., Ng, C.T. (2004). Due-date assignment and single machine
scheduling with deteriorating jobs. Journal of Operational Research Society 55, 198–
203.
10. Cheng, T.C.E., Kang, L.Y., Ng, C.T. (2005). Single machine due-date scheduling of
jobs with decreasing start-time dependent processing times. International Transactions
in Operational Research, 12, 355–366.
11. Crawford, S. ( 2000). A field study of schedulers in industry. Understanding their work,
practices and performance, Ph.D. Thesis, Nottingham University, Nottingham.
12. Crawford, S. (2001). Making sense of scheduling: the realities of scheduling practice in
an engineering firm. In: MacCarthy, B.L., Wilson, J.R. (Eds.), Human Performance in
Planning and Scheduling. Taylor & Francis, London, 83–104.
13. Cropley, A. J. (1999). Definitions of creativity. In M. A. Runco & S. R. Pritzker (Eds.),
Encyclopedia of Creativity (Vol. 1, pp. 511-524). London: Academic Press
14. Dragut A.B. (2003) A Markovian Approach To The Mathematical Control Of NPD
Projects, PhD Thesis, Department of Industrial Engineering and Innovation Sciences
,TU/e The Netherlands.
15. Dragut A.B. (2004). A Weakly Monotonic Backward Induction Algorithm on Finite
Bounded Subsets of Vector Lattices. Journal of Computational and Applied
Mathematics, 164-165(1), 295-306.
163
16. Dragut, A.B. (2006). Monotonic Robust Optimal Control Policies for the Time-Quality
Trade-Offs in Concurrent New Product Development (NPD), Computers and
Mathematics with Applications, volume 51, issue 2, p. 229-238
17. Dragut A.B. , Bertrand J.W.M. (2008) A representation model for the solving time
distribution of a set of design tasks in new product development (NPD). European
Journal of Operations Research. 189(3), 1217-1233
18. Dragut, A.B. (2011) Structured MDP Policies, Markov Decision Processes, in James J.
Cochran (Editor-in-Chief) Wiley Encyclopedia of Operations Research and
Management Science, vol.8 , Wiley, to appear in February 2011
19. Eyrolle, H., Cellier, J. M. (2000). The effects of interruptions in work activity: field and
laboratory results. Applied Ergonomics, 31(5), 537-543
20. Gardner, D. G. (1986). Activation Theory and task design: An empirical test of several
new predictions. Journal of Applied Psychology, 71(3), 411-418.
21. Gardner, D. G., Cummings, L. L. (1988). Activation Theory and job design - Review
and reconceptualization. Research in Organizational Behavior, 10, 81-122.
22. Gordon, V.S., Potts, C.N., Strusevich, V.A., Whitehead, J.D. (2008). Single machine
scheduling models with deterioration and learning: handling precedence constraints via
priority generation, Journal of Scheduling 11, 357–370.
23. Jackson, S., Wilson, J.R. and MacCarthy, B.L. (2004). A new model of scheduling in
manufacturing: Tasks. Roles, and Monitoring, Human Factors 46 (3), 533–550.
24. Lodree, E.J, Geiger, C.D. (2010). A note on the optimal sequence position for a rate-
modifying activity under simple linear deterioration. European Journal of Operational
Research 201,644–648.
25. Melchior,M.,Caspi, A., Milne, B., Danese, A., Poulton, R., and Moffitt,T.E.. (2007).
Work stress precipitates depression and anxiety in young, working women and men.
Psychological Medicine,37,1119-1130.
26. McKay, K.N., Buzacott, J.A., Charness, N., Safayeni, F.R.,1992. The scheduler’s
predictive expertise: an interdisciplinary perspective. In: Doukidis, G.I., Paul, R.J.
(Eds.), Artificial Intelligence in Operational Research. Macmillan, Basingstoke, pp.
139–150.
27. Min Ji, Cheng,T. C. E. (2010). Scheduling with job-dependent learning effects and
multiple rate-modifying activities. Information Processing Letters 110 (11), 460-463
28. Nepal, M. P. , Park,M. and Son, B. (2006). Effects of Schedule Pressure on
Construction Performance, J. Constr. Engrg. and Mgmt., 132(2), 182-188
29. Oorschot, K.E. van, Bertrand, J.W.M., and Rutte, C.G. (2005). Field studies into the
dynamics of product development tasks, International Journal of Operations &
Production Management, 25 (8), 720 – 739
164
30. Ormerod, T. C. (2005). Planning and ill-defined problems. In R. Morris & G.
Ward(Eds.), The Cognitive Psychology of Planning (pp. 53-70). London: Psychology
Press.
31. Puterman,M. (1994), Markov Decision Processes: Discrete Stochastic Dynamic
Programming, John Wiley and Sons
32. Sio, U. N. and Rudowicz, E. (2007). The role of an incubation period in creative
problem solving. Creativity Research Journal, 19(2-3), 307-318.
33. Smith, S. M. and Blankenship, S. E. (1991). Incubation and the persistence of fixation
in problem-solving. American Journal of Psychology, 104(1), 61-87.
34. Sonnentag, S. (2003). Recovery, work engagement, and proactive behavior: A new look
at the interface between nonwork and work. Journal of Applied Psychology, 88(3), 518-
528.
35. Sternberg, R. J. and Davidson, J. E. (1999). Insight. In M. A. Runco & S. R. Pritzker
(Eds.), Encyclopedia of creativity (Vol. 2, pp. 57-69). London: Academic Press.
36. Topkis D.M. (1998). Supermodularity and Complementarity. Princeton University
Press, Frontiers of Economic Research series
37. Zijlstra, F. R. H., Roe, R. A., Leonora, A. B., Krediet, I. (1999). Temporal factors in
mental work: Effects of interrupted activities. Journal of Occupational and
Organizational Psychology, 72, 163-185.
Appendix
Proof Proposition 1.
We recall that in this control system the probability that after taking the action at in the state xt, a
transition from state xt to state tttt axxx +≤≤ +1 occurs is given by the probability of solving in
one unit of time (which is actually the length of a review period [t,t+1)) the activities belonging
to the performance levels newly achieved in state xt+1. More precisely, the transition probability
is given by probability that the remaining solving time of the tasks in state xt up to the
performance levels from xt+1 is contained in an interval of length ε around the length of a
review period [t,t+1).
Let ( )
( ))(
1,
1,1nnk a
tnl
tnli
N
n
+
+==∑∑=
denote the total number of activities to be solved in order to make a transition from state
( ) )),(,...,,1( tNltlxt =
165
to the state ( ) ))1,(,...,1,1(1 ++=+ tNltlxt .
Let Cmax be the makespan of a queuing system of s parallel servers with a common queue of k
activities. Let tai be the number of nonempty components in the action at in A(t), corresponding
to concurrent tasks. Let ),(taikΞ be the random variable giving the remaining solving time of
the tasks in state xt up to the performance levels from xt+1. According to Dragut and Bertrand
2008 ),(taikΞ 's cumulative distribution function )))(,(( hikF
taΞ equals:
( ) )()1()()),(Pr()1(max
hfphfphiktaiErlangt YkCa −
−+⋅=≤Ξ .
However, in the state xt, the same number k of planned activities can be obtained for different
equiprobable xt+1. Therefore in order to obtain ),,( 1+ttt xaxp we have to divide the solving time
by the number of possible ways of arriving at these equiprobable xt+1. Let χ(xt+1 - xt) is the
number of partitions of the total sum of activities from xt to xt+1 into multiples of the number of
activities per level.
Thus ( ))(
)()1()(~),(,,
1
111 xx
ippxxppixxpxaxp
t
atttattttt
t
t −
⋅−+−⋅=−=
+
+++ χ
,
where tttt axxx +≤≤ +1 and dzzfip kCat t)()(~
)(2/12/1 max∫=
+−εε , dzzfip
taiErlangt Yat )()()1(
2/12/1 −∫=
+−εε .
These transition probabilities can be fast numerically approximated by the Euler Inverse
Laplace Transform method (see Dragut and Bertrand 2008).
In order to prove that ∑+≤≤
++
≤tttt axxx
ttt xaxp1
1),,( 1 we remark that both the arrival of new activities
per task and the interruption process stop when the last of the k planned activity is solved. Thus,
after a random finite number Q of solved activities, the queuing system stops. The system
without interruptions was an M/M/s system, and the interruptions delay the solving process
without adding any solved activity, thus we have a M/G/s queuing system, which has the
probability of two or more events in a time frame (∆ , ∆+h) equal to )(ho .
We have that
)()},( {Pr}1)({Pr
)()},( ,1)({Pr}{Pr
inevent one
inevent one
εε
εεε
oq£oqt£Sq
++∆∆−=∆=
++∆∆−==+∆<<∆
where qS is the time of the q -th event of the queuing system (the q -th activity solved) and
)(∆£ is the total number of activities that have been solved up to time ∆ . Thus we can write
the following equation:
166
==+∆≤Ξ<∆∑ }|),(Pr{ kKiKta
k
ε
= •=−=∑ }|1)(Pr{ qQQt£k
Pr{ one event in )()},( εε o++∆∆ .
We can conclude by taking with ∆=1 since the length of our time frame is [t,t+1) and by
dividing both sides with small ε. Then for small ε : 1)(),(2/12/1 ≅∫ Ξ
+−∑ dzzf
taiKk
εε .
Proof Theorem
The rewards properties are obvious by construction.
To prove that the sum of transition probabilities taken over any increasing subset of X(t) is
nondecreasing and superadditive we need first describe in detail the action set )( tt xA for
( )tXxt ∈ (which will be done in Proposition 2). Using this description we are able to
decompose an arbitrary increasing set into elementary ones (which will be done in a Corollary).
Then we prove all the required properties on the elementary increasing sets, and we extend them
to the general case of an arbitrary increasing set (which will be done in Proposition 3 and
Proposition 4)
Proposition 2 ( ) ( ) ttttt AxAxA == ~ , for ( )tXxx ∈~, .
Proof: )( tt xA is included in the increasing set }|{: minminaxxKa ≥= . The states are feasible
so they lead probabilistically to the rewarded region in tT − stages. Solving times are
independent, then n
na
i
N
nS
t )(
11
max,
==∑∑ is distributed Erlang ),( µk , where
)(:)(
11
max,
nnk a
na
i
N
n
t
==∑∑= .
Thus, for tamax, , the action space is equivalent to
[ ] )(1
!)( )(
1
0te
jMtT MtT
jk
jβµ µ −≤
−∑ −−−
=
which implies a maximal number of activities to be done. Thus, the maximal number of
possible levels to be achieved in )( tT − review periods is constant as well. There exists as
well a maximal number of possible levels of the set of actions which decides that the engineers
should perform more than the minimal requirement )(min ta . If ),(max tnL is large enough we
167
have no preset bounding of the action spaces at any decision point . Thus, they are only
naturally bounded by the engineers finite capacity and we have )~()( tttt xAxA = .
Corollary The increasing subsets of )(tX are
} s.t. },...,1{|)({,...,1 iss sxkitXxKq
≥∈∃∈= ,
where qss ,...,1 are not comparable, and N∈q , and js
q
jKtX
1)(
== , for some q .
Proof The partial order transitivity implies that the top of )(tA is a nondecreasing surface. A
bounded set in a finitely-dimensional lattice has only a finite number of incomparable elements.
Thus the proof is clear.
Proposition 3 The function ),,( 11
+∈+
∑ ttxtKx
xaxpt
is nondecreasing in )(tXxt ∈ , for any K
increasing subset of )1( +tX .
Proof The points in )1( +tX with 0),,( 1 ≠+tttt xaxp belong to the rewarded set
}|{def
, tttN
xax xaxxxRttt
+≤≤∈=+ N . We will prove for the elementary increasing subsets
since the the translation arguments are the same for the general case. Consider xx ~≤ then
),,~(),,( 1111
+∈
+∈ ++
∑≤∑ tttKx
tttKx
xaxpxaxpStSt
.
Case 1. SKxx ∈~, . Then we have Stt Kaxax ∈++ ~, . Thus, Sxaxxax KRRtt
⊂++ ~,~, , , so both
sums are equal being taken over all terms.
Case 2. tax + , St Kax ∈/+~ . Then SKxx ∈/~, both sides zero.
Case 3. SKxx ∈/~, , but St Kax ∈+~ . Since xx ~≤ , tax + may/may not be in SK . If
St Kax ∈/+ , =∩+ Sxax KRt, Ø and
),,~(0),,( 11~,~11
+∩∈
+∈
∑∑+++
≤= tttKRx
tttKx
xaxpxaxpSxtaxtSt
since =/∩+ Sxax KRt
~,~ Ø and 0),,~( 1 ≥+ttt xaxp . If St Kax ∈+ for each point in the
intersection Sxax KRt
∩+, we can construct a unique point of equal probability in the
intersection Sxax KRt
∩+~,~ . And since the terms of nonzero probability in ),,( 11
+∈+
∑ tttKx
xaxpSt
are the ones belonging to Sxax KRt
∩+, . For 2=N there is an easy graphical description.
168
Proposition 4 ),,( 11
+∈+
∑ ttttKx
xaxpt
is superadditive in
)()(),( tttt xAtXax ×∈ , for any K increasing subset of )1( +tX .
Proof Consider xx ~≤ s.t. )(~, tXxx ∈ , )~()(~ xAxAaa tt =∈≤ . We finished if we can
prove that the following holds for an arbitrary sK :
≥∑−∑ +∈
+∈ ++
),,~(),~,~( 1111
ttKx
ttKx
xaxpxaxpStSt
),,(),~,( 1111
+∈
+∈ ++
∑−∑≥ ttKx
ttKx
xaxpxaxpStSt
The transition probabilities separate the dependency on tt xx −+1 and ai . Then
0)()~( 11
,\~,1
1
~,~\~~,~1
1
≥−∑≥−∑ ++
++∈+
∈+
++∈+
∈+
xxPxxP tttt
axxRaxxRtx
SKtx
axxRaxxRtx
SKtx.
We reduce the proof to similar translation arguments for several cases as in the above
proposition.
Case 1. SKaxax ∈/++ ~~,~ and Case 2 SKxx ∈~, , where for each point in axxaxx RR ++ ,~, \
we have a correspondent in axxaxx RR ++ ~,~~~,~ \ .
Case 2. SKax ∈/+ ~ . Since aa ~≤ we have that SKax ∈/+ , and then
=∩=∩ ++ SaxxSaxx KRKR ,~, Ø and the right-hand is zero. We have xx ~≤ , both x~ ,
ax ~~ + may or may not belong to SK , but the inequality holds if the lefthand side is positive
due to the transition probabilities.
Case 3. SKx∈/ but SKax ∈+ ~ . Because aa ~≤ we have that ax + may or may not
belong to SK . Because SKaxxx ∈+≤ ~~,~ , but axx +~,~ may or may not belong to SK . We
will use the form of the transition probabilities as well as for SK increasing set of )1( +tX the
set Saxxaxx KRR ∩++ ,~, \ contains less points than Saxxaxx KRR ∩++ ~,~~~,~ \ . In the intersection,
we construct points of equal probability by translation.
169
Productiebeheersing: groei naar volwassenheid
Prof. Bertrand
Rede uitgesproken ter gelegenheid van de aanvaarding van het ambt van hoogleraar in de
productiebeheersing aan de faculteit Bedrijfskunde van de Technische Universiteit Eindhoven
op 29 september 1989.
170
Dames en Heren,
In deze rede wil ik een drietal onderwerpen aan de orde stellen die te maken hebben met de
groei naar volwassenheid van het vakgebied Produktiebeheersing. Deze onderwerpen hebben
betrekking op de organisatie van het onderzoek, de inhoud van het vakgebied en het
professionele gebruik van onze kennis in de praktijk.
De organisatie van het onderzoek Het vakgebied Productiebeheersing heeft de laatste decennia een onstuimige ontwikkeling
doorgemaakt. Het is opmerkelijk te moeten constateren dat de impulsen tot deze ontwikkeling
meestal niet afkomstig zijn uit de wereld van het wetenschappelijk onderzoek, maar grotendeels
het gevolg zijn geweest van gebeurtenissen in de bedrijfspraktijk. Op een paar uitzonderingen
na is tot nu toe geen enkele grensverleggende ontwikkeling het gevolg geweest van
researchwerk gedaan aan universitaire onderzoeksinstellingen. Integendeel, we zien een
omgekeerde gang van zaken. Onder invloed van ontwikkelingen in de praktijk ontstaat een
werkwijze die in bepaalde opzichten zeer succesvol blijkt te zijn. Dit wordt door managers en
consultants waargenomen en uitgedragen, en ook in de wetenschappelijke wereld vindt de
werkwijze op een zeker moment acceptatie. Aan de wetenschap is dan de taak te analyseren
waarom de werkwijze succesvol is, wat haar beperkingen zijn, waar nog verdere verbeteringen
mogelijk zijn, en onder welke omstandigheden deze realiseerbaar zijn.
Deze positieschets staat in nogal schrille tegenstelling tot de doorsnee-opvatting die er in
academische kringen heerst over de verhouding tussen onderzoek en praktijk. Deze opvatting
houdt in dat de wetenschap de bedrijfsproblemen en de ontwikkelingen observeert en
analyseert, en kennis produceert die gebruikt kan worden om betere productiebeheersings-
systemen te maken. Het zal duidelijk zijn dat dit verschil tussen bedoeling en werkelijkheid leidt
tot frustraties zowel bij de wetenschappelijk onderzoekers als hij de potentiële gebruikers in de
praktijk. Ik zal dit illustreren met twee historische voorbeelden.
MRP-I
De wetenschappelijke wereld werd in de jaren zeventig verrast door de ontwikkeling in de
praktijk van MRP-I als hulpmiddel om integraal materiaalbehoeften door te rekenen, over
meerdere productiefasen heen.
Deze techniek was in principe al lang bekend. Een artikel van Vazsonyi in het eerste nummer
van Management Science in 1954 behandelde de techniek Stuklijstexplosie ten behoeve van de
Materiaalbehoefteberekening1. Alleen het benodigde rekenvermogen was toen nog niet
beschikbaar. Tot dat moment werd voor grootschalige productie uitsluitend gewerkt met ketens
171
van bestelsystemen. De gebreken van deze werkwijze waren bekend en zijn op uitermate
inventieve wijze geanalyseerd door Forrester in zijn Industrial Dynamics2. De remedie was in
principe ook bekend, nl. snelle informatiedoorgave, korte reactietijden en integraal beheersen
over de keten heen, maar was nog niet praktisch uitvoerbaar (afgezien van een aantal
eenvoudige varianten van integrale beheersing zoals Base Stock Control, waar we hier niet
nader op ingaan). Met de ontwikkeling van snelle computers en grote geheugencapaciteit
ontstond ook de mogelijkheid om snelle informatiedoorgave en integrale beheersing te
realiseren. Er werden softwarepakketten gebouwd, en de kruistocht voor integrale
goederenstroombeheersing nam een aanvang.
De houding van de wetenschappelijke wereld tegenover deze ontwikkeling kenmerkte zich door
twee fasen. De eerste fase bestond daaruit dat men constateerde dat er niets nieuws onder de zon
was. De mooiste illustratie hiervan vernam ik onlangs van een Amerikaanse collega. Een MRP-I
deskundige gaf aan het begin van de jaren zeventig een lezing aan zijn universiteit over de
werking van MRP-1. Toen de spreker ophield keken zijn toehoorders hem verwachtingsvol aan:
de essentie van het verhaal zou nog moeten komen; dit kon niet meer zijn dan de inleiding.
Niets was echter minder waar: de spreker had zijn boodschap al gebracht.
Deze anekdote illustreert dat MRP-I in eerste instantie op inhoudelijke gronden niet serieus
werd genomen: men vond het niet de moeite waard om zich hiermee bezig te houden en er
werden vele, vaak terechte, bezwaren geformuleerd tegen het wat naïeve beeld van de
werkelijkheid dat onder MRP-ligt. De praktijk hield zich echter wel bezig met MRP-1. De
APICS zag de grote praktische mogelijkheden van integrale goederenstroombeheersing en nam,
gestuwd door consultancybureaus, het voortouw bij de MRP-I kruistocht. De vaak terechte
bezwaren van wetenschappers tegen de rigide, deterministische werkwijze van MRP-I werden
aanvankelijk genegeerd, en bij de ontwikkeling van de softwarepakketten werden geleidelijk
lapmiddelen ingebouwd om de gevolgen van bepaalde gebreken te ondervangen. De MRP-I
systemen werden vervolgens op grote schaal in de praktijk ingevoerd. Voorraadpunten werden
geëlimineerd, de productieorganisaties werden aangepast, en Materials Management werd
ingevoerd.
Voor dit harde feit geplaatst, zien we vanaf het einde van de jaren zeventig dat een deel van de
wetenschappelijke wereld zich stort op formele probleemstellingen ontleend aan de nieuwe
besturingssituatie. Weer is een van de meest populaire onderwerpen het bepalen van de optimale
seriegrootte, maar nu binnen de MRP-I context. Het is alsof de wetenschappers opgelucht
constateren dat ze weer op vertrouwd terrein zijn. De beperkingen en gebreken van MRP-I, die
iedere vakman op het gebied van productiebeheersing onmiddellijk duidelijk zijn, worden niet
of nauwelijks meer naar voren gebracht; er wordt ook niet of nauwelijks gewerkt aan de
ontwikkeling van betere integrale systemen. Een gunstige uitzondering hierop is het werk van
Monhemius en Timmer, gericht op de integratie van Base Stock Control en MRP-13, dat later
172
onder leiding van Wijngaard en Wortmann is voortgezet door Van Donselaar4. Beide
onderzoeksprojecten werden uitgevoerd in bedrijven. In het algemeen echter kunnen we zien dat
er van het wetenschappelijke onderzoek weinig praktisch gerichte vernieuwing uitgaat, en dat is
zeer te betreuren.
KANBAN
Het tweede voorbeeld is ontleend aan de ontwikkelingen rond Just-in-Time. Het begrip Just-in-
Time heeft betrekking op alle activiteiten die gericht zijn op het tot stand brengen van een
organisatie die in kleine series, met lage voorraden, efficiënt en op tijd kan produceren en
leveren. Seriegroottes en voorraadhoogtes zijn altijd populaire onderwerpen geweest van
wetenschappelijk onderzoek. Als er dus één gebied is waar de beschikbare wetenschappelijke
kennis een nuttige bron van vooruitgang voor de praktijk kan betekenen, dan is het wel dit
gebied. Echter, als we terugkijken naar de gebeurtenissen van de afgelopen twintig jaar, dan is
niets minder waar. De parallellen met het voorafgaande voorbeeld zijn tekenend. In de jaren 70
werden in het westen de eerste verhalen bekend over het KANBAN-systeem. Wie als
productiebeheersingsdeskundige voor het eerste de werking van het KANBAN-systeem
bestudeert, heeft als primaire reactie dat dit systeem alle problemen negeert, die je in praktijk
tegenkomt, en dus uiterst onpraktisch en theoretisch van aard is.
Er wordt niet of nauwelijks rekening gehouden met seriegrootte-effecten, doorlooptijdvariaties,
beperkingen in de capaciteitsbeschikbaarheid en instabiliteit van de afname. Het KANBAN-
systeem is een zeer eenvoudig en effectief systeem, maar kan alleen werken in een ideale
omgeving.
In 1979 nam ik in Amsterdam deel aan de Fifth International Conference on Production
Research. Een tweetal medewerkers van Toyota Motor Co., die meegewerkt hadden aan de
ontwikkeling van het KANBAN-systeem, gaven daar een presentatie over de dynamische
eigenschappen van het KANBAN-systeem, en constrasteerden deze met de dynamische
eigenschappen van een klassieke besturing5. Tot mijn verbazing gingen ze in hun studie voor
het KANBAN-systeem uit van stationaire afname van de producten, en lieten ze resultaten zien
die uitsluitend onder stationaire afname golden. Later is mij duidelijk geworden dat het
KANBAN-systeem geacht wordt toe te leveren aan een of meer montagebanden die zelf over
een zekere termijn een constant tempo hebben. Bij wijziging van het tempo wordt dit vooraf
besloten, en voor het hele productiesysteem integraal ingevoerd. Aan twee ogenschijnlijk
strijdige eisen, integrale besturing en operationele eenvoud, wordt hiermee in één keer voldaan
door organisatorische afspraken te maken over de momenten waarop het productieniveau zich
mag wijzigen, en vooruitlopende op die wijziging, de nodige maatregelen te treffen om die
niveauwijziging op het geplande tijdstip te laten ingaan. Op deze manier hoeft men zich in elke
173
schakel van de keten enkel bezig te houden met het vervangen van de werkelijk verbruikte
producten, en worden opslingerverschijnselen en nervositeit vermeden zonder dat hiervoor
gecompliceerde informatieverwerkingssystemen nodig zijn. De besturing van de productie-
niveaus is de verantwoordelijkheid van een hoger gelegen besturingsniveau: wij zouden dit de
goederenstroombesturing noemen.
Bekijken we de aanvankelijke opstelling van de wetenschappelijke wereld ten opzichte van deze
ontwikkeling dan zien we weer hetzelfde patroon. Artikelen over integrale goederenstroom-
beheersing bestonden er al in de jaren vijftig (b.v. Clark en Scarf6), meestal gericht op de vraag
hoeveel totale ketenvoorraad er nodig is op elk niveau in de keten, welke besturingsregels er
moeten worden gebruikt, welke bestelseries gebruikt moeten worden en welke veiligheden er
nodig zijn. Het was algemeen bekend dat dit probleem relatief eenvoudig is als de vraag
constant is, de seriegroottes overal klein zijn, en de productiedoorlooptijden kort zijn, maar met
eenvoudige problemen hield men zich niet bezig. Het was weer Burbidge7 die in zijn publicaties
over Production Flow Analysis en Period Batch Control al vroeg een lans brak voor het kiezen
van eenvoudige oplossingen voor beheersingsproblemen, en voor het zodanig (re)organiseren
van de uitvoering dat het probleem eenvoudig wordt Een ander frappant voorbeeld van het
kiezen van een eenvoudige organisatorische oplossing is de ontwikkeling van het Perioden
Planningssysteem voor de enkelstuksfabricage dat door Geraerds8
in de jaren 60 bij de
Koninklijke Luchtmacht is ontwikkeld, in een tijdvak dat helemaal in de geest stond van het
onderzoek naar prioriteitsregels voor stochastische wachtrijproblemen. De aanpak van Geraerds
was er ook een van eliminatie van een aantal complicerende factoren, stabilisering van de vraag
naar capaciteit in verhouding tot het capaciteitsaanbod, en daarna werken met vast planperioden
per bewerking voor de uitvoering.
Deze aanpak schept zoveel organisatorische duidelijkheid dat het voor het uitvoerende
personeel mogelijk wordt naar het gewenste resultaat toe te werken; anticiperen wordt mogelijk,
waardoor tijdig maatregelen kunnen worden getroffen om detailproblemen tijdig op te lossen.
Voor complexe fabricageprocessen die sterk afhankelijk zijn van menselijk ingrijpen, zoals in
de kleinserie en enkelstuksfabricage, betekent deze keuze van de beheersingsstructuur dat
zelforganisatie van de uitvoerende werkzaamheden mogelijk wordt gemaakt. De tien condities
voor zelforganisatie die Kuiper9
recentelijk noemde in zijn intreerede, worden onder deze
omstandigheid realiseerbaar. De aanpak gaat uit van de mens als probleemoplosser en niet van
de mens als uitvoerder van een voorgeprogrammeerde beslisregel. Zowel Period Batch Control
en Production Flow Analysis, alsook het Periode Planning Systeem beschouw ik als voorlopers
van de JIT-aanpak. Geen van deze ontwikkelingen nam echter in het wetenschappelijk
onderzoek een hoge vlucht.
174
Toen het succes van het KANBAN-systeem duidelijk werd, is de JIT-aanpak breed
uitgewaaierd over de praktijk, en is de wetenschappelijke wereld hard aan het werk gegaan om
het functioneren van KANBAN-systemen te analyseren, en na te gaan waar de essentiële
verschillen liggen met andere vormen van besturing. Weer is het de wetenschap die reageert op
autonome ontwikkelingen in de praktijk en er in eerste instantie niet mede vorm aan geeft,
terwijl toch de ingrediënten voor het mede bepalen van de ontwikkelingen beschikbaar waren.
Dames en Heren,
Een volwassen vakgebied kenmerkt zich door vermogen om alle nuttige kennis op deelgebieden
te kunnen integreren in diagnose en in ontwerp.
De voorgaande voorbeelden laten zien dat het vakgebied Productiebeheersing nog niet overal op
een volwassen manier beoefend wordt. Welke conclusies kunnen we nu trekken met betrekking
tot de organisatie van het onderzoek naar het ontwerp van productiebeheersingssystemen? Mijn
stellingname is de volgende.
- Productiebeheersing is een functie binnen de bedrijfsvoering waarbij alle aspecten van de
bedrijfskunde een rol spelen. Het ontwerp van goede productiebeheersingssystemen vereist
dan ook de inbreng van kennis uit alle bedrijfskundige disciplines.
- De kwantitatieve analyse van processen en beheersingsregels is een van de grondslagen van
het vakgebied productiebeheersing.
- Zolang in de uitvoering van de productiefunctie mensen een rol spelen, zal de besturing
altijd verlopen via beslissingen die genomen en uitgevoerd worden door mensen. Het
ontwerp van productiebesturingssystemen moet derhalve uitgaan van de mens als
probleemoplossend element in het geheel. Deze mens zal zijn rol spelen, afhankelijk van de
positie die hij inneemt in de organisatie, de wijze waarop hij wordt beoordeeld, de
technische hulpmiddelen die hij ter beschikking heeft, de informatie die hij krijgt of die hij
zich weet te verschaffen en zijn vermogen om die informatie te gebruiken .
- Productiebeheersing is een proces dat zich afspeelt in bedrijven. Het onderzoek naar
productiebeheersingssystemen dient dan ook plaats te vinden in bedrijven. Het is
vanzelfsprekend dat hiermee de ruimte voor experimenteel onderzoek beperkt is; het
afbreukrisico voor bedrijven is groot.
In laboratoriumsituaties kunnen natuurlijk wel bepaalde aspecten van een probleem worden
onderzocht zoals het functioneren binnen een bepaald model van een bepaald type
beslisregel, of de effectiviteit van het aanbieden van informatie in bepaalde vorm b.v. op
een beeldscherm, op een print, etc., of het functioneren van bepaalde geautomatiseerde
informatiekoppelingen tussen machines. Echter binnen de bedrijfskunde is dat alleen dan
175
zinnig onderzoek, als duidelijk is binnen welke functiegerichte probleemstelling het
resultaat van dit onderzoek past.
In de faculteit Bedrijfskunde kristalliseert het onderzoek zich de laatste jaren rond een viertal
onderzoekthema's. Ondanks het feit dat de ent van dit kristallisatieproces telkens gelegen was in
een der disciplines, zien we dat het ontwerpgerichte karakter en de daarmee samenhangende
multidisciplinaire samenwerking zich hoe langer hoe sterker begint af te tekenen. Monhemius is
jarenlang de drijvende kracht is geweest achter deze organisatie van ons wetenschappelijk
onderzoek en heeft ervoor gezorgd heeft dat we nu langzamerhand binnen de faculteit toekomen
aan integratie van de beschikbare disciplinaire kennis. Als actueel voorbeeld wil ik hier noemen
het onderzoek naar integratie van logistieke en administratieve beheerssystemen, dat momenteel
onder leiding van Theeuwes wordt uitgevoerd10
, en waarin wordt samengewerkt vanuit de
vakgebieden informatiesystemen, bedrijfseconomie en productiebeheersing. Ik verwacht dat er
de komende jaren meerdere van dit soort projecten zullen worden gestart. Voorwaarde hiervoor
is echter wel dat er een geschikte infrastructuur is voor het uitvoeren en managen van dit soort
onderzoek en dat we in staat zijn ervaren onderzoekers gedurende hun meest productieve jaren
vast te houden. Naar mijn mening is een bedrijfskundig onderzoeksinstituut voor dit doel
onontbeerlijk. Voor sommige delen van ons onderzoek kan de benodigde infrastructuur
gevonden worden binnen het ITP.
De inhoud van het vakgebied
Dames en Heren,
In het voorgaande betoog ben ik ingegaan op de positie van het productiebeheersingsonderzoek,
ofwel de industriële logistiek, in relatie tot de ontwikkelingen in de praktijk. Ik heb hierbij over
productiebeheersing in algemene termen gesproken. Dat mag de suggestie wekken dat in mijn
optiek de productiebeheersing een eenvormig kernobject is, en dat ik b.v. zou kunnen behoren
tot de school die van mening is dat de gereedschapskist waarop de naam Manufacturing
Resources Planning ofwel MRP-II staat, voldoende concepten en instrumenten bevat om alle
typen produktiebeheersings-problemen aan te kunnen. Niets is echter minder waar.
De diversiteit waarin productiebeheersingsproblemen zich voordoen, is enorm en het is zelfs de
vraag of deze ooit uitputtend in zijn relevante factoren beschreven kan worden. De
productiebeheersingsproblemen evolueren als onderdeel van de evolutie van de bedrijven en de
bedrijfsvoering. Als voorbeeld noem ik de aandacht die er tegenwoordig is voor coördinatie
tussen de klant en zijn toeleverancier(s). Was het 15 jaar geleden nog vanzelfsprekend om de
systeemgrenzen van het probleem te leggen bij de ontvangende en de verzendende afdelingen,
176
nu zien we dat de inkoop-en verkoopfunctie, en de structurele afspraken die deze afdelingen met
hun tegen-spelers maken, vaak onderdeel zijn van de probleemstelling.
Het is tekenend voor de alertheid van het onderzoekmanagement van de faculteit Bedrijfskunde,
dat er reeds geruime tijd een onderzoekproject loopt op dit gebied.
Als dan de werkelijkheid zo divers en zo dynamisch is, zijn er dan geen constanten, en vaste
ontwerpregels die de ontwerper kan gebruiken? Naar mijn mening zijn die regels er wel
degelijk, maar zijn ze dermate abstract dat ze nietszeggend zijn voor wie niet beschikt over veel
en gevarieerde praktijkervaring. Voor de ervaren ontwerper daarentegen vormen deze regels een
praktisch stramien dat richting geeft aan zijn ontwerp. Een uitstekend voorbeeld voor zo'n set
ontwerpregels vinden we in de publicatie van MeaP1 in de Harvard Business Review, getiteld
'Putting Production Decisions Where They Belong', en in eerste aanzet, in de intreerede van
Wijngaard12
. Ik prijs me gelukkig dat ik de laatste jaren de gelegenheid heb gehad om samen
met Wijngaard en Wortmann deze ontwerpregels verder uit te werken. Het gebruik van deze
regels is echter zeer situatie-afhankelijk, en de ervaren en deskundige ontwerper kenmerkt zich
daardoor, dat hij de vaardigheid bezit om per situatie de juiste vertaalslag te maken naar
concrete beslisfuncties en beheerssystemen. Een volwassen vakgebied kenmerkt zich o.a.
daardoor, dat ontwerpers zodanig kunnen worden opgeleid dat ze na enige jaren ervaring in de
praktijk deze vaardigheid bezitten. Het vakgebied productiebeheersing staat in dit opzicht op de
drempel van volwassenheid. De oprichting van een Nadoctorale Opleiding tot Logistiek
Ontwerper aan deze universiteit is hiervan het resultaat.
Zoals gezegd is de toepassing van de ontwerpregels sterk situatieafhankelijk. Ondanks de grote
diversiteit in productiebeheersingsproblemen kunnen we toch een aantal factoren onderkennen
die ten dele bepalend zijn voor de inrichting van de beheersing. Deze factoren zijn direct
gerelateerd aan de twee dominante aspecten van de productiebeheersing: de materiaal-
coördinatie en de capaciteitscoördinatie. We zullen op beide aspecten kort ingaan.
Materiaalcoördinatie
Met betrekking tot de materiaalcoördinatie kunnen we onderscheid maken in productie-
processen met een sterk divergente materiaalstructuur en productieprocessen met een sterk
convergente materiaalstructuur. Bij een divergente materiaalstructuur wordt uit een gering
aantal uitgangsmaterialen een groot aantal verschillende eindproducten vervaardigd; er is dus
met betrekking tot de eind productie sprake van een grote materiaalgemeenschappelijkheid.
Materiaalcoördinatie is hier niet zo'n groot probleem. Dit kenmerk vinden we bij de onderdelen-
fabricage. Bij de convergente materiaalstructuur zien we dat er per eindproduct een groot aantal
verschillende materialen en halfproducten nodig is. Meestal is het aantal verschillende
177
eindproducten op hoofdtypeniveau aanmerkelijk kleiner dan het aantal verschillende uitgangs-
materialen. Dit kenmerk vinden we bij assemblageprocessen.
Capaciteitscoördinatie
Met betrekking tot de capaciteitscoördinatie kunnen we onderscheid maken in productie-
processen met een eenvoudige capaciteitsstructuur en die met een complexe capaciteits-
structuur. Er is sprake van een eenvoudige capaciteitsstructuur als elk van de producttypen in
dezelfde volgorde evenveel capaciteit vraagt van elk van de capaciteitstypen. In dat geval kan
de capaciteit uitgedrukt worden in aantallen producten per tijdseenheid en kan een van de
capaciteitstypen als structurele bottleneck worden aangewezen.
Dit soort capaciteitsstructuur treffen we aan in de massafabricage, waar de productie sterk
gespecialiseerd is en de productiecapaciteit ontwikkeld is voor een speciale range producten.
Er is sprake van een complexe capaciteitsstructuur als er grote variaties zijn in de bewerkings-
volgorde en de hoeveelheid capaciteit die elk producttype vraagt van elk der capaciteitstypen. In
dit geval kan de capaciteit van een productiesysteem voor besturingsdoeleinden niet uitgedrukt
worden in aantallen producten/tijdseenheid, maar moeten we werken met hoeveelheid
beschikbare uren per tijdseenheid per capaciteitstype. De bottleneck hangt af van de producttype
mix in het orderpakket en wisselt daarom voortdurend. Dit soort capaciteitsstructuur treffen we
aan in de enkelstuks- en kleinseriefabricage.
Dames en Heren,
Het zal duidelijk zijn dat ik hierboven extremen heb geschetst en dat maar weinig
productiesituaties zich met een der extremen laten karakteriseren. In een materiaalstructuur die
in principe divergent is, komt, zeker in de laatste productiefase, vaak enige convergentie voor.
In een in principe convergente materiaalstructuur komt vaak enige divergentie voor, zeker in de
eerste productiefase waar zoveel mogelijk naar het gebruik van gemeenschappelijke onderdelen
zal worden gestreefd. Een soortgelijke kanttekening geldt voor de capaciteitsstructuur. Dat
neemt echter niet weg dat het denken in extremen een helder licht werpt op de essentieel
verschillende situaties die er zich vanuit besturingsoogpunt voordoen. We bereiken dit door de
extremen van deze twee factoren paarsgewijs met elkaar te combineren. We krijgen dan het
diagram in Fig. 1.
178
Fig. 1
Als we proberen deze indeling toe te passen op complete bedrijven, dan zullen we vaak
constateren dat dat niet kan. Dat komt doordat binnen de bedrijfsgrenzen veel typen producten,
tussenproducten en Fabricageprocessen voorkomen die elk een andere karakteristiek hebben.
Dat lijkt verwarrend, maar is het niet. Het betekent enkel dat de typologie in Fig. 1 enkel
toegepast mag worden op productieafdelingen, of productie-eenheden. Binnen de
bedrijfsgrenzen kan er een aantal productieafdelingen bestaan die ieder hun eigen karakteristiek
hebben, en die ieder op een andere wijze bestuurd moeten worden.
De coördinatie van de goederenstromen tussen de afdelingen binnen het bedrijf, en
tegenwoordig zelfs voor een deel buiten het bedrijf, noemen we de goederenstroombeheersing.
Deze houdt zich niet bezig met de interne afdelingsbeheersing. In principe krijgen we hierdoor
de besturingsstructuur in Fig. 2.
Het goederenstroombeheersingsprobleem kan complex worden door de configuratie van
verschillende afdelingen die voor de productie nodig zijn, ieder met hun eigen karakteristiek, en
de complexiteit van de materiaalstroom tussen de afdelingen. Het zal duidelijk zijn dat er vele
combinaties van verschillende type afdelingen binnen een goederenstroom kunnen voorkomen.
179
Fig. 2
Dit maakt meteen duidelijk waarom het gebruik van standaardsoftware in bedrijven telkens
weer op grote weerstanden stuit. Niet alleen is er sprake van een historisch bepaalde wijze van
werken, waar mensen maar moeilijk van af te brengen zijn, maar ook is de bestaande werkwijze
vaak haarscherp afgestemd op de typische configuratie van afdelingssituaties in dat bedrijf, en
op de typische productmarkt verhouding die voor dit bedrijf geldt; maar daar wil ik hier niet op
ingaan.
Uit het diagram in Fig. 1 kan een aantal conclusies worden getrokken. Deze wil ik kort
samenvatten, en ook per situatie een onderwerp noemen waarop het productiebeheersings-
onderzoek zich de komende jaren moet concentreren.
Procesgewijze Fabricage
- Voor de procesgewijze fabricage is het productiebeheersingsprobleem op afdelingsniveau
tamelijk eenvoudig. Als we produceren, produceren we vaak snel, en het onderhanden werk
is relatief laag ten opzichte van de voorraden voor en achter de productieafdeling. Er is een
geringe volumeflexibiliteit en de besturing van de afdeling is vaak geïntegreerd in de
goederenstroombesturing op het centrale bedrijfsbureau.
Interne afdelingsbeheersingsproblemen komen hier vaak voort uit combinatie- en omstel-
problematiek, opbrengstvariaties en onbeheerste processen.
In dit gebied is veel kwantitatieve kennis voorhanden met betrekking tot de
hoofdproductieplanning en de seriegrootte- en volgordeprogrammering. Lacunes zijn de
korte termijnbesturing, en met name het reageren op onverwachte gebeurtenissen in
productieopbrengst en materiaalkwaliteit, gekoppeld aan de orderpositie en de
grondstofpositie, met behoud van de geprogrammeerde afdelingsdoelstelling. Ik verwacht
dat hier de komende jaren veel onderzoek naar gedaan zal worden.
GOEDERENSTROOMBEHEERSING
AFD AFD AFD
180
Grootserie-assemblage
- Voor de grootserie-assemblage is de materiaalcoördinatie de dominante factor. De hele
besturing is gericht op het tijdig beschikbaar hebben van de vele componenten. Het
beslissen over omstellen naar een ander producttype is ook hier vaak geïntegreerd in de
goederenstroombesturing. Onderzoek is hier nodig naar de coördinatie van product-
ontwikkeling, inkoop en verwerving, fabricage, en service/ onderhoud. Kern van dit
onderzoek is de vraag naar de integratie van de diverse stuklijsten die in deze verschillende
functionele gebieden gebruikt worden. Ik verwacht veel van het onderzoek dat op dit gebied
onder leiding van Wortmann wordt uitgevoerd13
.
Enkelstuks/kleinserie fabricage
- Voor de enkelstuks/kleinseriefabricage is de interne capaciteitscoördinatie de dominante
factor; materialen zijn met enkele uitzonderingen meestal eenvoudig en met korte
levertijden verkrijgbaar. Volumeflexibiliteit wordt hier gevonden in het uitbesteden van
werk. Hier vinden we relatief zelfstandige productieafdelingen met een complexe interne
besturing, waarover veel kwantitatieve kennis beschikbaar is. Veel van deze kennis wordt
echter nu niet gebruikt, omdat de benodigde gegevens niet in de juiste vorm, op het juiste
tijdstip en bij de juiste man beschikbaar zijn. De invoering van real-time order-
voortgangssignaleringssystemen zal een verdere delegatie van bevoegdheden mogelijk
maken. Er zal voor dit type situatie de komende jaren veel organisatorisch onderzoek nodig
zijn, ten einde de organisatievormen te ontwikkelen of te vinden waarin het best gebruik
gemaakt wordt van de beschikbare informatie.
Projectgewijze Fabricage
- Voor de projectgewijze fabricage is zowel capaciteitcoördinatie als materiaalcoördinatie van
belang. In de projectgewijze assemblage zien we vaak een grote capacitatieve flexibiliteit;
naast een aantal montage- en testspecialisten zijn er vele productiemedewerkers wier
vaardigheden op ruime schaal voorhanden zijn. Capaciteitscoördinatie houdt hier in de
beheersing in het werkaanbod voor de specialisten, en het managen van de capaciteit voor
de algemene montage. Materiaalcoördinatie is hier een ingewikkeld probleem, omdat er
vaak meer projecten gelijktijdig lopen, elk project in een andere fase verkeert en de
materiaalbehoeftes per project slechts geleidelijk aan duidelijk worden gedurende de
looptijd van het project.
In de projectgewijze fabricage bestaat er vaak een grote flexibiliteit met betrekking tot
volgorde waarin activiteiten moeten worden uitgevoerd, en ook de definitie van de
activiteiten is vaak niet eenduidig en vast. De huidige plannings- en besturingstechnieken
181
gaan echter uit van bekende vaste entiteiten en relaties, en behandelen die entiteiten en
relaties ook als zodanig.
De grofplanning, de netwerkplanning en de detailplanning zijn daarom in deze situatie niet
of slechts met grote moeite consistent te houden.
Het is zelfs moeilijk om te definiëren wat hier onder het begrip consistentie hier moet
worden verstaan. Dit vormt een ernstige handicap voor de beheersing van dit soort
productieprocessen; het uit de hand lopen van productontwikkelingsprojecten en
bouwprojecten is vaak hierop terug te voeren. Het ontwikkelen van nieuwe concepten,
organisatievormen en hulpmiddelen voor de beheersing van de projectgewijze fabricage, die
toelaten dat van de beschikbare flexibiliteit wordt gebruik gemaakt, en het tegelijk mogelijk
maken op elk beheersingsniveau de werkelijke beperkingen en beïnvloedingsmogelijkheden
in grijpbare en hanteerbare vorm weer te geven, beschouw ik als een belangrijk onderzoeks-
project binnen de productiebeheersing.
Goederenstroombeheersing
Wat is, in de bovenstaande visie op productiebeheersing op afdelingsniveau, de positie van de
goederenstroombeheersing? De goederenstroombeheersing is gericht op het realiseren van het
Hoofdplan. De goederenstroombeheersing doet dit door met de afdelingen op termijn afspraken
te maken over het orderniveau, ten behoeve van de capaciteitsplanning, en door het plaatsen van
concrete orders of productieopdrachten. Het doet er hierbij niet toe of de betrokken
productieafdelingen behoren tot de eigen onderneming, of dat het externe bedrijven zijn, die ook
produceren voor andere productondernemingen. Het kan zelfs voorkomen dat een
productieafdeling produceert voor een aantal verschillende productgroepen die elk hun eigen
goederenstroombeheersingssysteem hebben. De koppeling naar de afdelingsbeheersing verloopt
via de capaciteitsplanning en de het plaatsen van productieorders.
Dames en Heren,
Uit het voorgaande betoog kan worden afgeleid dat productieafdelingbeheersing en
goederenstroombeheersing twee geheel verschillende problemen zijn. Dit verklaart de beperkte
toepasbaarheid in de praktijk van de standaard MRP-systemen, die immers primair ontwikkeld
zijn voor de goederenstroom beheersing ten behoeve van complexe producten die in
middelgrote tot grote series worden geproduceerd. Dit houdt in dat standaard MRP in principe
niet geschikt is voor de goederenstroom beheersing van productgroepen die zich kenmerken
door procesgewijze fabricage en enkelstuks-fabricage. Daarnaast kunnen er vraagtekens gezet
worden bij de koppeling naar de afdelingsbeheersing. De goederenstroombeheersing staat of
valt met betrouwbare productieafdelingen die dus moeten beschikken over goede
182
afdelingsbeheersingssystemen. In de praktijk blijkt de afdelingsbeheersing en de koppeling
tussen afdelingsbeheersing en goederenstroombeheersing vaak een zwak punt te zijn van de
standaard MRP-systemen. Naar aanleiding van een tweedaagse conferentie over dit onderwerp
in Rochester, USA, met als titel 'Beyond MRP-I; Evolution to a new Standard?' werd door een
aantal deelnemers de voorzichtige hypothese geformuleerd dat het huidige integrale standaard
MRP-concept wel eens een doodlopende straat zou kunnen zijn.
Als alternatieve weg werd gezien het ontwikkelen van krachtige afdelings-beheersingssystemen
die op zichzelf kunnen werken en voorzien zijn van standaard koppelingen ten behoeve van de
capaciteitsplanning en het plaatsen van orders. Deze ontwikkeling wordt mogelijk gemaakt door
het beschikbaar komen van real-time besturingssystemen op de werkplek.
Ten behoeve van de goederenstroombeheersing zijn er dan enkel nog systemen nodig die zich
bezig houden met capaciteitsmanagement en materiaal·· coördinatie en orders, met daarnaast
natuurlijk de koppeling met de commerciële activiteiten in de Hoofd Planning per productgroep.
Ik verwacht dat de ontwikkeling zich in deze richting zal voortzetten en ben van mening dat het
productiebeheersingsonderzoek in de faculteit hierop moet inspelen. De toenemende
specialisatie van bedrijven naar hun kernactiviteiten, en de tendens om delen van de niet-
kritische productieactiviteiten uit te besteden bij gespecialiseerde bedrijven, werkt in dezelfde
richting. Een belangrijk vraagstuk in dit verband is de vraag naar de aard van de relaties tussen
de productieafdelingen en de productgroepen. Dit dient mijns inziens een van de kernvragen te
zijn van een onderzoek naar de beheersing van industriële ketens; een zeer actueel onderwerp
voor de Nederlandse en Europese industrie.
Het gebruik in de praktijk
Dames en Heren,
Het derde en tevens laatste thema van deze lezing vergt het minste tijd, maar ligt mij zeer na aan
het hart. Tot nu toe hebben we gesproken over de aard van productiebeheersingsonderzoek, en
over de beschikbare kennis en een aantal lacunes in deze kennis. Tot slot wil ik even stilstaan bij
de praktijk van de productiebeheersing en hiervoor put ik uit mijn subjectieve waarnemingen als
projectmedewerker en organisatieontwikkelaar in het kader van een aantal logistieke
verbeteringsprojecten in de praktijk. Productiebeheersingsonderzoek is Multi-disciplinair en
derhalve moeilijk te realiseren. Dat geldt veel minder voor het realiseren van logistieke
verbeteringen in de praktijk, zoals ik tot mijn verbazing een aantal keren heb moeten
constateren. Ik zal deze stellingname illustreren aan de hand van een grootscheeps logistiek
verbeteringsproject waarbij ik een aantal jaren geleden betrokken was. Het betrof hier een
183
onderneming bestaande uit een aantal relatief zelfstandige bedrijven. Het verbeteringsproject
werd top-down aangepakt, met centraal vastgestelde doelstellingen op termijn, gesteld in termen
van doorlooptijdverkortingen en verhogingen van leverbetrouwbaarheid. Het project werd
geleid vanuit een centrale stuurgroep waarin alle bedrijfsleiders zitting hadden, per bedrijf een
lokale stuurgroep, en lokale werkgroepen waarin de benodigde veranderingen en de
invoeringsmethodiek werden uitgewerkt. De nadruk in het project lag op de lokale
verantwoordelijkheid van de bedrijven, en het bood veel ruimte voor het realiseren van een
eigen aanpak van de problematiek. Er ontstonden dan ook grote tempoverschillen tussen de
bedrijven, en in elk bedrijf lag het accent van het verbeteringsproces op andere aspecten,
ondanks grote overeenkomsten in de problematiek. De decentrale aanpak hield verder in dat er
vanuit de centrale staforganisatie kennis en informatie en capaciteit beschikbaar gesteld werd
aan de bedrijven. Een van mijn functies in het geheel was dienen als kennisbron en vraagbaak,
en de bedrijven te helpen bij het uitwerken van procedures.
De verschillende bedrijven deden in zeer verschillende mate een beroep op de centraal
beschikbare kennis en capaciteit. Toen na 2,5 jaar de tussenbalans werd opgemaakt kon worden
geconstateerd dat bij een aantal bedrijven de levertijd en betrouwbaarheid sterk waren
verbeterd. Daaronder was ook een klein bedrijf dat geen beroep had gedaan op de centrale
stafcapaciteit, en dat ook verder geen externe ondersteuning had aangetrokken.
Het bleek dat de bedrijfsleider in zijn eentje de essentiële boodschappen met betrekking tot de
benodigde veranderingen had opgepikt uit de centrale stuurgroepvergaderingen en deze in zijn
bedrijf met kleine werkgroepen van direct verantwoordelijken had ingevoerd. Dat ging, naar hij
vertelde, niet zonder slag of stoot, maar door vasthoudend en goed communicerend met zijn
mensen aan de veranderingen te blijven werken, kreeg hij het voor elkaar.
Het gedrag van deze bedrijfsleider illustreert de uitspraak: productiebeheersing is niet moeilijk,
productiebeheersing moet je gewoon doen.
Dames en heren,
De les die ik uit deze ervaring getrokken heb is dat benodigde voorwaarden voor
organisatorische veranderingen ten minste zijn:
1. een leiding die belang heeft bij de verandering;
2. een leiding die weet dat de nieuwe taakstelling realiseerbaar is;
3. een leiding die in grote lijnen weet hoe de verandering tot stand moet worden
gebracht.
184
Voor diegenen die wetenschappelijk of professioneel actief zijn in het vakgebied
productiebeheersing, volgen hieruit drie taakstellingen. Ten eerste: het doen van onderzoek naar
de strategische relatie tussen enerzijds een goede productiebeperking, en anderzijds het
bedrijfsbelang op langere termijn. Het is opmerkelijk hoe weinig empirische kennis er hierover
beschikbaar is. Ten tweede: het onderzoek naar de vraag onder welke omstandigheden welke
logistieke prestatie realiseerbaar is, en welke organisatiemodellen hiervoor nodig zijn. Dit dient
voor verschillende bedrijfstypen te worden uitgewerkt. Ten derde: bevorderen van
kennisoverdracht naar leidinggevenden op alle niveaus in de organisatie.
De Nederlandse Vereniging voor Logistiek Management, de NEVEM, mede opgericht en toe
bloei gekomen onder leiding van Botter, heeft de afgelopen 10 jaar een uitstekende basis gelegd
voor de kennisoverdracht op deze drie gebieden. Met name is er een compleet cursusbouwwerk
tot stand gebracht. Er worden momenteel stappen ondernomen om ook de eerste twee
taakstellingen te realiseren. Ik prijs me gelukkig in de gelegenheid te zijn binnen de NEVEM
hieraan op bestuurs- en op uitvoerend niveau mee te kunnen werken. Ik ervaar deze combinatie
als zeer effectief. De beschikbare capaciteit voor ontwikkeling en kennisoverdracht binnen
Nederland is beperkt en verdeeld over vele instituten en universiteiten. Het zou een goede zaak
zijn als het gebruik van deze capaciteit door de instituten enigszins zou worden gecoördineerd.
Dat blijkt niet vanzelf te gaan. Naar mijn mening ligt hier een uitdaging voor de NEVEM. Een
volwassen vakgebied kenmerkt zich mijns inziens o.a. door het bestaan van
beroepsverenigingen die mede richting géven aan de ontwikkeling van het vakgebied.
De in gang gezette ontwikkeling van deze relatie tussen de professionele gebruikers en de
professionele ontwikkelaars kan worden gezien als de afsluitende fase in de ontwikkeling van
het vakgebied productiebeheersing naar de status van volwassenheid.
Ik dank u voor uw aandacht.
Literatuur
1. Vaszonyi, A., 'The Use of Mathematics in Production and Inventory Control',
Management Science 1955,1, pp. 70-85.
2. Forrester, J., Industrial Dynamics, Cambridge, Mass., M.LT. Press, 1961.
3. Timmer, J.P.J., Monhemius, W. en Bertrand, J.W.M., 'Production and Inventory
Control with the Base Stock System', Report EUT / BDK/12, Eindhoven University of
Technology, 1984.
4. Donselaar, K. van, 'Material Coordination under Uncertainty', proefschrift TUE, 1989.
185
5. Kimura, O. en Terada, H., 'Design and Analysis of 'Pull system'. A method of Multi-
stage Production Control', full papers of Vth IC&PC Conference, Amsterdam, 1979.
6. Clark, A.J. en Scarf, H., 'Optimal Policies for a Multi-Echelon Inventory Problem',
Management Science 960, 6, pp. 475-490.
7. Burbidge, J.L., 'The Principles of Production Control', McDonnald & Evans Ltd.,
London.
8. Geraerds, W.M.J., 'Produktiebeheersing en Bedrijfskunde: naar samenhang',
Inaugurale rede, THE, 1973.
9. Kuipers, H., 'Zelforganisatie als Ontwerpprincipe', Inaugurale rede, TUE,1989.
10. Theeuwes, JAM., 'Naar een Bedrijfskundige Bedrijfseconomie', Inaugurale rede, TUE,
1987.
11. Meal, H.C., 'Puttihg Production Decisions where they Belong', Harvard Business
Review, 1984, pp. 102-'111.
12. Wijngaard, J., 'Kwantitatieve Methoden bij de Bedrijfsvoering', Inaugurale rede, THE,
1983.
13. Wortmann, J.C., 'Bedrijfskunde en Informatica: naar samenhang', Inaugurale rede,
TUE, 1987.
186
187
THE STRUCTURING OF PRODUCTION
CONTROL SYSTEMS3
J. W. M. Bertrand and J. Wijngaard
ABSTRACT This paper presents a qualitative methodology for designing hierarchically structured
production control systems for complex production situations. The methodology is based on the
assumption that complexity should be reduced by defining self-contained PU-systems with clear
and well-defined operational characteristics. Furthermore the interactions between the
subsystems should be simple and restricted. We introduce the Production Unit (PU) as a basic
control entity. From the perspective of goods flow control the PUs are black boxes having
certain operational characteristics. The objective of goods flow control is to realise a certain
delivery performance, taking into account the PU-operational constraints. The main elements in
the goods flow control structure as developed here are Master Planning, Material Co-
ordination, Workload Control and Work Order Release.
Introduction
Production control refers to the co-ordination of production and distribution activities in a
manufacturing system to achieve a specific delivery reliability at minimum costs. In many
customer oriented production situations the manufacturing activities have developed in such a
way that manufacturing is specialised according to product-type and/or to manufacturing
technology. The result is a production structure with a number of production units, where each
unit takes care of a separate part of the production, and where the goods flow in and between
these production units can be quite complex. In such a production system, each of the
production units will have its own short-term and long-term goals, whereas each product-type
delivered to the market may require materials and capacity from a number of different
production units.
In order to realise the required delivery performance in the market, co-ordination of the
3 Overgenomen uit International Journal of Operations and Production Management, Vol. 6 (1966), No. 2
188
activities of the production units is therefore necessary. These co-ordination activities,
however, should not conflict with reaching the production economics objectives for each of the
production units. On the one hand, realising production economics objectives is in the interest
of the system as a whole. On the other hand, however, the production units should show high
flexibility with respect to reacting to changing market conditions, demand forecasts, and actual
demand. Lack of flexibility may lead to high and unbalanced stocks, poor delivery perfomance,
and possibly loss of market position. This conflict between short-term interests of operation
units and goods flow control is well-known in the literature. A structural (hierarchical)
approach is needed to resolve this conflict [l].
In the past decade a number of studies have been published on the design of hierarchical
production control systems. Many of these studies reported on the principles underlying
particular design projects in practice [e.g., 2, 3]. Other research used mathematical analysis to
investigate specific types of aggregation and decomposition [e.g., 4-6], or used systematic
computer simulation for this purpose [e.g., 7, 8]. In this paper we stress the general problem of
how to structure the complete production control process. We study the subject from the point
of view that the control sub-problems at any level should be defined such that the controllability
of the problem is guaranteed and the actual control performance can be measured and therefore
can be monitored. This approach has been used in a previous research project on production
control in a production unit [9] and is now applied to a more complex production problem
including goods flow control. The concepts used in this approach are partly based on certain
concepts from Manufacturing Resources Planning (MRP-II).
This conceptual framework for production control deals with the problems of short-term and
long-term inflexibilities in the production units [l0, 11]. Specifically, the concept of a Master
Production Schedule (MPS) has been introduced as a device to reconcile the conflicts between
market needs and production possibilities (see, e.g. [12]).
The MRP-II concepts were a substantial step forward in the design of goods flow control
systems. However, a real quantitative basis for the operational design of such systems is still
lacking, as can be concluded from the many difficulties encountered when MRP-I1 is being
used as a basis for design in practice. This paper aims at filling a part of this practicality gap. It
introduces some basic concepts for designing production control systems which achieve high
flexibility while still enabling the production units to realise their production economics.
For this purpose, we first introduce definitions of basic concepts for describing a production
system, such as items, materials, capacity and operations, then we introduce the concept of
Production Unit which is used as a homogeneous logistic entity. The co-ordination of the
production units is referred to as Goods Flow Control (GFC). Next we consider how to
structure GFC and the relationship of GFC to Production Unit Control (PUC).
189
Basic Concepts
We assume that for any production control problem in practice a system boundary can be
determined. The system boundary defines what "part of the production world" is considered and
what part is beyond our scope. The system boundary should be established operationally by
specifying the inflows and outflows from the environment into the system considered and vice
versa. This system boundary concept is illustrated in Figure 1.
We assume the manufacturing processes to be given. Thus for each finished product (end
item) the following is known:
• the end product structure, which is the way in which the product is composed of
materials, parts or components, and subassemblies;
• the capacity types needed;
• the manufacturing steps which are needed for each of the components, subassemblies
and final assemblies in the product; and
• the amount of capacity required for each manufacturing step.
Figure 1. The Boundaries of the Production System
With this general specification of the system we can roughly define the production control
problem to be that given certain consistent objectives regarding customer delivery performance
and manufacturing costs, how should we:
(1) accept customer orders;
(2) place procurement orders;
(3) vary the capacity;
(4) allocate available capacity to manufacturing steps?
Capacity decisions Capacity changes
Production system
considered
Deliveries
Procurement orders
Shipments
Customer orders
190
Depending on the system boundary chosen, the complexity of the problem can vary
substantially. We restrict our research to production systems with many complex end items,
where interactions and relationships between products and their timing stem from the following
factors:
• the products use shared capacity resources, with restricted availability;
• the products use shared types of materials and subassemblies;
• information on market demand is limited;
• work orders are released with batch sizes which may be larger than the immediate required
amount;
• short term capacity variations are possible to a limited extent and with a certain lead-time.
Materials, Resources
We assume that the manufacturing process for an end-item can be defined as a related set of
transformations. Each transformation may require materials and/or resources. As materials we
define other items which are absorbed in a finite discrete amount during the transformation
step. As a resource we define objects which are not used up during the transformation step, but
which are in use (machines, space, etc.).
Operations
A manufacturing process is a network of manufacturing steps. For the purpose of production
control the manufacturing steps are aggregated into operations. The specification of operations
should be related to the scope of the production control problem at hand. The operations
generally do not follow straightforwardly from the description of the manufacturing steps, but
must be based on the aspect of the system that is addressed by production control. Now as
production control addresses the timing of the allocation of resources and materials, a natural
criterion for the grouping of manufacturing steps into operations is their relative independence
in time. Thus if there is little freedom in relative timing within a group of manufacturing steps,
it would be natural to consider this group as one operation, which requires the resources and
materials of all the manufacturing steps in the group. From the production control point of view,
an operation is a black box with specific properties, and which is not subject to internal
manipulation.
Production Units and Goods Flow Control
In order to simplify the total production control it is necessary to distinguish production units
191
(PU). A PU is a specific set of capacities performing a specific set of operations using a specific
set of materials. The total control is decomposed to PU Control (PUC) and Goods Flow Control
(GFC). The control variable which constitutes the interface between PUC and GFC is the
release of new work orders to the PU. The PU has to realise the work orders according to certain
norms. GFC has to co-ordinate the release of work orders to the various PUs, to each other and
to customer orders and to provide the required materials.
An ideal PU is self-contained from a manufacturing point of view. Also from a production
control point of view the PU is self-contained, but it is generally constrained with respect to the
amount and timing of its production. These constraints constitute the operational relationships
of a production unit. The constraints are basically generated by its limited capacity and by the
operation processing times required for the manufacturing of the items. However, additional
constraints can be generated by the way in which a PU organises its production process in
order to realise specific objectives regarding product quality and production efficiency. For
instance, if set-up times are an important part of the operation processing time, then the PU
may want to work with specific batch sizes. Moreover, if machine set-up times are sequence
dependent, the PU may want to maintain a certain working stock of work-in-process at that
machine in order to be able to create an efficient production sequence.
The creation of a PU requires a relatively stable environment for that unit with respect to the
availability of resources and the demand for product items produced by that unit. This stability
is required because the PU will operate in a relatively independent way, and therefore it will
need a number of environmental invariabilities to base its internal structure on.
From the perspective of Goods Flow Control, the PUs are black boxes, which have specific
operational characteristics, and which can only be influenced under certain conditions via
specific inputs. These conditions reflect the agreements regarding the environmental
conditions. For instance, an agreement could be that GFC can release work orders to a PU, on
the condition that the work load of that PU never exceeds a specific limit. On that condition,
the PU may promise average delivery times of started work orders according to specific pre-set
norms. A very different agreement might be that GFC could release any work order to the
production unit, and that the PU promises to deliver the orders according to variable due dates,
specified at the time of release, which takes into account the actual workload at that time.
Many more examples can be given of possible sets of agreements regarding performance and
environmental conditions.
It will be clear that generating stable environmental conditions for a PU will be quite easy if
the environment of the production system itself is rather stable. In fact, the "difference" between
the actual stability of the system's environment and the stability implied in the agreed
environmental conditions, of the PUs has to be accounted for by GFC. For instance, if the
agreed conditions per PU imply more stability than the system's environment shows, then the
192
GFC system should be designed so that it can absorb the difference. Goods Flow Control could
then hold and use buffer stocks of finished goods or components to allow for the PU to adapt to
the changes in the environment in a smooth way. On the other hand, if the agreements with the
PU imply much flexibility, then GFC can just pass the variations in the system's environment
to the PU.
Each possible transformation that can be realised by a PU can be defined on the production
network of operations. Ideally, a PU should be defined such that the subnetworks per PU can
be considered from the GFC viewpoint as single production phases. Then, for GFC, the
production processes for end-items can be expressed in terms of this set of production phases,
and a set of relationships between these phases. In fact these sets constitute an aggregate
production structure, which allows the GFC to use a rather aggregate production model
showing much less detail. GFC does not deal with operations, but with production phases.
Thus GFC controls the behaviour of the stock levels of the items at specific points in the
operations network of the products. These specific points are the manufacturing states in
between the PUs. We refer to these manufacturing states as controlled stock items of the
production process.
The GFC problem therefore can be defined as the control of the levels of the controlled stock
items, by means of the release of work orders for production phases, within the constraints set
by the PU agreements, in order to realise a specific delivery performance at minimum costs.
Bertrand [13] and Bertrand and Wijngaard [14] give more details about production units, the
interface of PUC and GFC and the location of controlled stock-points.
Goods Flow Control Structure
Considering the complexity of Goods Flow Control, it will be clear that designing a general
optimal control is impossible. We propose to decompose GFC into the following four parts:
(1) Master Planning
(2) Material Co-ordination
(3) Workload Control
(4) Work Order Release.
Figure 2 gives an outline of the relationship between these levels of control. The Master
Planning level forms the connection with the higher levels of control in the production
organisation, where the various systems of the organisation (logistics, quality, finance,
193
manpower, etc.) are integrated into a production objective (Management Control [15]). This
production objective (the Master Plan) is formulated in terms of production level plans for the
PUs and a delivery schedule for the Master Production Schedule (MPS) items. Material Co-
ordination translates the MPS into priorities for purchasing orders and production work orders
per controlled stock item, taking into account the actual production progress. Finally, the
production level plans per PU are combined with the work order priorities to determine which
work orders are to be released to the shop floor if the actual
Figure 2. Sketch of Goods Flow Control Structure
workload allows such a release. The operational constraints of the PU are accounted for in this
phase. From this description it will be clear that Workload Control and Work Order Release are
decision functions located at the interface between Goods Flow Control and Production Unit
Control.
Depending on the specific production situation this global control structure can be detailed in
many different ways. Two factors are particularly crucial in this respect:
• the co-ordination of sales and manufacturing;
• the interference of capacities and products.
These two elements will be discussed next.
Priorities
Capacity usage
Aggregate release pattern
Master planning
Material co-ordination
Workload control
Workorder release
MPS
Workorders released
194
Co-ordination of Sales and Manufacturing Within each production system one has to recognise the influence of the sales and
manufacturing functions. Sales generates demand and accepts commitments with respect to
customer deliveries. Manufacturing has to realise the required deliveries. Co-ordination between
these activities is required. There have to be aggregate agreements with respect to service
performance, stability and reliability of required delivery patterns, total accepted demand, etc.
This we call structural co-ordination, which affects aggregates and averages.
Structural and Operational Co-ordination
If there is only structural co-ordination sales determines required delivery patterns, taking into
account the aggregate agreements, but neglecting the actual state of manufacturing. In such a
situation the required deliveries can be interpreted as autonomous demand. In many cases,
however, improvements can be expected from also taking into account the actual state of
manufacturing in the co-ordination process. This implies operational co-ordination. In the
MRP-II framework this type of co-ordination is incorporated in the concept MPS [12].
However, in that framework the MPS serves two purposes. It is not only a realistic (potential)
delivery plan based on the structural and operational co-ordination of sales and manufacturing,
but It is also the basis for co-ordinating the various production units in the system by netting and
offsetting. In current MRP literature these two functions of the MPS are not well distinguished.
In this paper we will formalise this distinction and interpret the MPS as (potential) delivery plan
based on state information with respect to both sales and manufacturing.
Sales Flexibility
Including state information with respect to manufacturing in the MPS makes sense only if sales
can use the information. This is the case if there exists a certain flexibility at the sales side.
Such sales flexibility could be for instance the possibility of influencing customer order due
dates, the possibility of stinting short term sales promotions, etc. Another source of sales
flexibility exists if sales controls to some extent the inventories in the distribution stages.
However, if sales flexibility is very limited it can be beneficial to have an MPS including
information on the state of manufacturing. For instance if, due to some manufacturing problem,
shortages are inevitable, it makes it possible for sales to distribute the shortages over the
various product types or customers.
In situations where sales flexibility is very small the MPS can be interpreted as "demand",
that is, as an objective for manufacturing. In cases where there is much short term sales
flexibility the MPS also represents the state of manufacturing and can be interpreted as an
objective for sales as well.
195
Form of the MPS
An MPS is a sequence of vectors or a matrix (quantities/period of the MPS items). It will be
clear that such a matrix is a rather poor representation of the combined state of the
manufacturing and sales process. A better state representation would be a set of trajectories.
For manufacturing these trajectories would pertain to the set of realisable delivery patterns; for
sales they would pertain to the set of acceptable delivery patterns, where the term acceptable
refers to the possibility of realising certain sales patterns. The objective of both sales and
manufacturing would then be to realise a non-void intersection of these two sets of trajectories.
However, we should keep in mind that these state variables can only be realistically described
in stochastic terms, as the real future deliveries and sales not only depend on the current state
and the future control decisions, but also on a number of uncontrolled stochastic variables.
Therefore we accept the convention of using one single delivery pattern as state representation
of both manufacturing and sales. This pattern may (informally) be interpreted as a more or less
arbitrary pattern in the above mentioned intersection of sets of realisable and acceptable
delivery patterns. From this discussion it will be clear that the relationship between this MPS
and the real state of the system cannot be formalised. However, in situations where the
flexibility of manufacturing and sales and the procedures to co-ordinate both are quite stable, we
may expect that the participants in this co-ordination process will generate implicitly stable and
consistent models of this relationship.
Generally the MPS is established periodically and is intended to be valid during the following
period. The discussion above makes clear that each new MPS may differ from the old one. This
is because of all kinds of uncontrolled stochastic variables, with respect to sales as well to
manufacturing, which have occurred during the period, and which are accounted for in the new
MPS. By not adapting the MPS, a high sales or manufacturing flexibility would be presupposed
and this flexibility would be used in a very rigid and probably non-optimal way.
Interference of Capacities and Products
To illustrate the relationship between capacities and products in Goods Flow Control we
consider a production system with independent demand (no sales flexibility). The difficulty of
having an MPS instead of responding to demand has been discussed in the previous section. In
the next section both difficulties are combined and complete control structures are described.
In situations with independent demand the objective of GFC is to realise a certain service
performance. This has to be done by controlling work order releases to PU’s and by adjusting
capacities and capacity usage of PU’s. The decision freedom for GFC to manipulate these
196
variables and to vary the inventories in the controlled stock points have to be "budgeted" by
higher levels of control. This also includes the restrictions on the variables imposed by the
batching and sequencing constraints.
In controlling the goods flow we face disturbances and fluctuations with respect to:
• procurement lead times
• production lead-times
• capacities
• yield
• demand
• registration of inventory and work in process.
In general it is important to put effort into reducing these fluctuations and disturbances or into
making them more predictable. The current drive in industry for realising short lead-time, high
quality (zero-defect) and just-in-time production therefore should be highly valued. However, in
many situations the possibilities are limited, sometimes because of technical constraints,
sometimes because of economic constraints. On top of that, the capacity flexibility may be too
small to cope with the remaining variability, be it predictable or unpredictable. For many
situations inventory buffers are necessary to absorb the state variations due to the gap between
flexibility and variability. Using inventory buffers in the right way can be an effective and
efficient way to absorb short term variability.
Inventory as Stored Capacity
A weak point of using inventory buffers to absorb production and demand fluctuations is that
inventories can only be realised as quantities of specific items, while most types of fluctuations
are directly or indirectly related to capacity availability. Fluctuations in production lead-times,
for instance, are partly due to disturbances in capacity availability. But fluctuations in demand
for a certain product also have a capacity dimension. To some extent, inventories of other
products can be used to absorb these fluctuations. In short, inventory of specific products has
the property that it also can be used as stored capacity.
We will clarify this point with a simple example. Consider a make-to-stock situation with one
production stage and two products which have identical production and demand characteristics.
Compare a state with inventories [I1,I2] with a state with inventories [I1+ x, I2 - x].
In the short term these states are different with respect to the risk of stock-out. On a somewhat
longer view, however, the states are equivalent in this respect. The term at which the states are
equivalent corresponds to the term at which for both products production has taken place. At
that term the effectiveness of the buffer for absorbing capacity and demand fluctuations depends
197
only on the total inventory. Also, fluctuations in demand for product 1 or product 2 are
equivalent at that term. One can cope to the same extent with both types of fluctuations
irrespective of the detailed state of the inventory.
Two Control Levels
The above example shows that in the somewhat longer term the inventory of products with
regular demand can be considered as being just stored capacity. This term is roughly equal to
the production cycle time, that is, the term at which all products with regular demand have
been produced at least once. In many situations the production cycle time is much shorter than
the term at which capacities can be changed. That makes it possible to distinguish two levels of
control. Capacity decisions can be decomposed from detailed inventory control and work order
release. In many cases, i.e., where constant work order throughput times are a prerequisite for
adequate control, capacity usage is determined by capacity [9]. In other situations it is possible
in principle to vary capacity usage without varying the capacity. In such situations work orders
are directly coupled to customer orders and internal work order throughput times maybe related
to customer order due dates. However, in such situations, if the PU’s operate at a high
utilisation rate, such independent variations of capacity usage have a long lasting impact on
mean work order throughput times. For that reason, in these cases it makes sense to integrate
capacity decisions and capacity usage decisions and to decompose this from detailed inventory
control.
This kind of decomposition has been investigated and discussed [16-18]. At the first level only
capacities and capacity aggregates are used as variables. Demand, production and inventory are
all aggregated to capacity. At the second level, results with respect to production levels are used
as budgets. Next, the production level has to be allocated to the various products. A good
objective for this second level of control is to keep the expected run-out times of the individual
products equal as far as possible (taking into account, of course, the batching and sequencing
restrictions). At this level detailed short term information has to be used. Because of production
and demand variations, and because of the operational constraints of the PU’s it is not possible
to realize completely equal run-out times. The remaining degree of imbalance in run-out times
can be considered as the control performance of the second level and has to be taken into
account at the first control level: at the first level extra inventory (slack) should be provided to
allow for this imbalance.
The effectiveness of such a decomposition depends on the extent to which the performance of
the second level decision process is independent of the decisions made at the first level.
Independence is high in case of rigid capacities and high utilisation rates.
It should be noticed that slow moving products should be excluded from this hierarchical
method of control. This is because slow mover inventory is not very effective as stored capacity.
198
Including slow movers would increase the production time significantly. Thus the problem
remains of how to deal with slow movers in this approach. A straightforward and very effective
possibility is to give slow movers high priority at all levels of decision making and to adjust the
capacity availability by reducing it for the capacity required for the slow movers (for details of
this approach see [19]).
Decomposition can also be applied to multi-stage production situations [20], although in this
case the relationship between the capacity aspect and the product aspect is more intricate. In
this case, for each product-item of a PU the horizon for the second (detailed) level of Goods
Flow Control has to be increased with the total production lead-time (the stacked item lead-
time).
The Product-oriented Approach
An attractive alternative to the capacity-oriented hierarchical approach discussed above is the
product-oriented approach. In this approach the capacity usage decisions are not integrated
with the capacity adjustment decision. Capacity usage is integrated with capacity allocation.
These decisions are decomposed along the products; all products are controlled separately. The
interference with other products because of restricted capacities is modeled as a stationary
extension of the production lead-time in the PU. This interference is revealed at the work order
release level. In situations with a stable utilisation rate such a product-oriented approach works
just as well as the hierarchical approach [18]. It is possible then to use queueing type analyses
to estimate the delays due to the interference of products because of restricted capacities [21].
But if there is no stable utilisation rate the product-oriented approach may not be expected to
work well. Decisions with respect to capacity changes have to be based on the effect of these
changes at the lower level of control. The only effect which can be taken into account easily in
this product-oriented approach is the influence on utilisation rate and via utilisation rate on
interference delays. This only makes sense if each capacity change leads to a new stationary
situation. However, if a PU works at a high utilisation rate, the transient times are long and it
will take much time before a new situation is established. Thus if capacities are changed
frequently it is necessary to control (aggregate) inventories and capacities simultaneously. The
product-oriented approach does not then fit.
It has to be mentioned that in the hierarchical approach slow movers have to be controlled in a
product-oriented manner. However, the capacity usage of slow movers is generally so small that
there are hardly any delays due to interference with the restricted capacities. This implies that
with the capacity-oriented hierarchical approach we have none of the problems that are typical
for the product-oriented approach.
199
The Integral Detailed Approach
Up to this point we have discussed two ways to decompose the complete problem of controlling
capacities, capacity usage and individual product inventories. Our premises have been that the
complexity and the stochastic nature of the problem make a decomposition approach
unavoidable. However, in simple and more deterministic cases the use of an integral approach
may be realistic, possibly restricted to the main products. An integral model of the production
control problem, including decision variables, state variables and goal variables, can then be
built, and the solution of this model can be realistically implemented in practice. In that case it
is generally necessary to use a kind of decomposition but this time the term decomposition
refers to techniques to solve the integral model (i.e., decomposition techniques applied to large
scale mathematical programming models). See Billington et al. [22] for an outline of such an
approach. This type of decomposition (problem decomposition) should be distinguished
carefully from the decomposition approach applied to decision making processes which we
discuss in this paper.
The Complete Control Structure
The Goods Flow Control structure has to depend on the:
• flexibility of the system
• objectives
• characteristics of the environment.
The flexibility of the GFC system is determined by the operational relationships of the
Production Units and by restrictions (budgets) with respect to inventories, make or-buy
decisions, capacity changes, etc. The objectives of the GFC system are norms and restrictions
with respect to service performance. Characteristics of the environment refer to procurement
characteristics, customer behaviour, yield (product and production quality) behaviour of
capacities (as far as not controllable by GFC).
It has to be mentioned that the distinction between an objective and a restriction of the system
is rather arbitrary. For instance an inventory budget could be interpreted as an objective.
However, we choose not to do so. We consider timely delivery as the objective function of
GFC and the possibility of varying inventories as one of the means for realising this. To
guarantee an adequate control structure and to realise consistent restrictions and objectives it is
necessary (at design level) to monitor and control flexibility, objectives and environmental
characteristics. It is important to notice that flexibility is not given but is partly the result of the
200
definition of production units and the design of Production Unit Control. At GFC level PU’s
are characterised by operational constraints and norms. Flexibility within the PU’s provides the
possibility of keeping the characteristics rather simple [23]; this contributes to GFC simplicity.
The introduction of PU’s with rather simple operational constraints and performance norms
reduces the variety of GFC systems. However, the variety is still too high to allow for one
uniform structure to be applied to all possible situations in practice. Therefore, we have
developed two closely related structures which cannot be used in all situations directly, but
which can, in most situations at least, be used as a starting point in the design process. The dif-
ference between the structures is due to different PU characteristics. The first structure is
intended for cases where the production levels of the PU’s are difficult to change (e.g. complex
job-shop production units). The second structure is intended for situations where the
production level can be varied at the same term as the capacity allocation. In both structures,
however, we distinguish four levels of control:
• Master Planning
• Workload Control
• Material Co-ordination
• Work Order Release.
Here we will only present the first structure. For a description of the second structure and a
discussion of the relationship of the two structures we refer to [14]. From now on we restrict our
attention to cases where, in the short term, the production level of the PU’s is inflexible. To
keep the performance of such PU’s predictable, the mean throughput time and utilisation rate
should be fixed [9]. This means that the available capacity determines the production level. The
basic structure we propose for this kind of situation is given in Figure 3. The function of Master
Planning is to control capacity variations (and hence variations in production level) and the
aggregate MPS. The capacity aggregates determined by Master Planning serve as restrictions
for Material Co-ordination. The function of Material Co-ordination is to disaggregate (allocate)
the aggregate flows so as to get a good balance of the individual inventories. The relationship
between Master Planning and Material Coordination is as previously described. The
performance of Material Co-ordination in such a structure is not expressed in terms of customer
order delivery performance but is related to the balance of the individual final inventories (or
backlogs). At Master Planning level an estimate of this performance should be available to
allow realisation of aggregate inventories such that the delivery performance for each of the in-
dividual products is sufficient.
Two important factors influencing the performance of Material Co-ordination are batching
restrictions and sequencing restrictions.
201
Figure 3. Goods Flow Control Structure in the Case of Inflexible Production Levels
Workload Control and Work Order Release are functions at the interface of GFC and PUC.
The output of Workload Control is the aggregate release pattern. Work Order Release
concerns work orders of individual products. Co-ordination with PUC is necessary because of
preferences within the PU. We will consider each of these functions now in more detail.
Master Planning
The aggregates to use at Master Planning level are capacity aggregates. This is easy as long as
Work orders
Aggregate
Detailed
Workorder priorities
Aggregate release pattern
(aggregate) MPS
Production level
Goods Flow state (capacities, inventories
Master planning
Sales + Marketing Information (aggregate)
Workload control
Material Co-ordination
Work order release
Production Unit state
Sales information (detailed)
202
there is only one relevant capacity dimension and the products have almost identical
characteristics. In such cases production level patterns and an aggregate MPS are sufficient to
project future expected aggregate inventory patterns.
It may become more difficult if there are more relevant capacity dimensions. It is necessary then
to estimate the capacity content of the inventories for all relevant capacity dimensions. This can
be done by controlling the MPS in a more detailed way, or by using fixed ratios of capacity
requirements on different capacity dimensions. This is equivalent to aggregating the bill of
capacity.
The possibilities of using such aggregate capacity bills in the Master Planning process have
been investigated by Axsater and Jönsson [8]. The stability of the MPS over the various
products is important in using such aggregate bills.
Another difficulty with respect to aggregate Master Planning is due to the existence of slow
movers or more general products with irregular production. As has been explained, such
products should not be included in the aggregate decision variables. A possible solution is to
aggregate only over regularly produced products and give the other products a high priority at
all levels of decision making. The slow movers are then completely controlled by Material Co-
ordination, and the control performance of Material Co-ordination with respect to these products
is the delivery performance itself. The aggregate release pattern for fast movers has to be
adjusted for the releases of work orders for slow movers.
It is difficult to construct suitable decision support models to support the Master Planning
function. An important reason for this is that it is not easy to formalise sales flexibility. The
MPS is the output of a co-ordination process of sales/marketing and manufacturing. Given the
current definition of a Master Plan, the procedures of this co-ordination process can be
formalised, but not its content. As long as sales and marketing flexibility cannot be formalised
in an operational way it is only possible to support the production level determination of Master
Planning by models in which the MPS is treated as independent demand. And even then the
models may become highly complex because of the behaviour of the MPS as a function of time
(stability, reliability) and because of the multi-stage, multi-capacity aspect of the production
situation. However, one could try to use HMMS-rules, linear programming based or control
theory type of decision rules, and to evaluate, by means of numerical analyses or simulation,
how such methods work for certain typical situations. Evaluation has to be based on variation in
production level and inventory variation [24]. The influence of the planning horizon could be
one of the points of interest here (for reviews of models which can be used to support the Master
Planning function see [25-27]).
203
Material Co-ordination
As already mentioned the function of Material Co-ordination is to balance the finial inventories
of the various products. The capacity usage determined by Master Planning serves as a budget.
Slow movers, however, are excluded from this hierarchical approach and are controlled
completely at Material Co-ordination level.
It is important to notice that at the Material Co-ordination level it is possible in general to use
more actual and more detailed sales information than at the Master Planning level. Master
Planning and Material Co-ordination may have different review periods and the sales
information used by Material Co-ordination is detailed and updated compared to the aggregate
sales information used by Master Planning.
Models to support Material Co-ordination are infinite capacity safety stock models [28] as far
as products with irregular production are concerned. For products with regular demand one
needs models to estimate the effect of certain release policies on the imbalance of final
inventories. Constructing useful models of this last kind is fairly easy because the imbalance of
the final inventories is rather insensitive to capacity variations and changes in predictability
[29]. This even makes it possible to use small scale simulation models to get complete results
for this control aspect.
Workload Control
In the type of situation considered here the throughput time is assumed to be determined at a
higher level of control. That means that capacity, production level and workload norm are
strictly coupled variables. Applying the workload norm means that the aggregate release
pattern depends also on the way the PU is controlled. The aggregate release depends on the
state of the PU.
Work Order Release
Material Co-ordination determines detailed (dynamic) release priorities taking into account the
(static) restrictions with respect to batching and sequencing. However, the preferences of the
PU cannot be described completely by such static restrictions. Think for instance of the actual
availability of specialised personnel. This means that in general the actual work order releases
will be the result of a co-ordination process of Material Co-ordination and Production Unit
Control. The actual state of the production unit will influence what is going to be released. This
will affect the performance of Material Co-ordination (the balance), just as the sequencing and
batching restrictions affect this performance. At the Master Planning level this effect has to be
taken into account as well.
204
Conclusions
Production control is complex. Many decisions interfere with each other and the production
control in total interferes with the control of other aspects in the organisation (quality,
manpower, ...). Structuring is necessary to reduce the complexity. The structure should be
chosen so that the loss of potential flexibility is minimised.
The precise structure to be chosen should depend on the characteristics of the organisation.
However, there are elements with respect to structure which have a much wider generality. The
following elements have been discussed in this paper:
• the definition of basic elements as capacities, materials and operations as a first step in the
design of the production control structure, instead of considering them as externally given
items;
• the introduction of production units and the decomposition of the total production control
to Goods Flow Control and Production Unit Control; and
• the relationship of sales and manufacturing and the interference of products and capacities
as two main determining factors of the Goods Flow Control structure.
The generality of these elements makes it possible to develop a small, but relatively complete
set of reference structures. For Goods Flow Control in a repetitive manufacturing situation
(multi-stage, multi-product) we have discussed one such reference structure.
During recent decades the contribution of Operations Research (OR) to production control has
been quite restricted. On the one hand OR could contribute to solving specific instances of
problems in practice. On the other hand OR could contribute to solving rather general but
artificially simple theoretical problems. The existence of (more or less) standard production
control structures makes it possible to exploit OR models and techniques much better. The
existence of such standard structures also reduces the number of relevant OR models. Relevant
OR models are models which fit some standard structure or which can be used to choose
between different standard structures.
References
1 Meal, H. C., "Putting Production Decisions where they Belong", Harvard Business Review,
Vol. 84, 1984, pp. 102-111.
2 Bitran, G. R. and Hax, A. C., "On the Design of Hierarchical Production Planning
Systems", Decision Sciences, Vol. 8, 1977, pp. 28-55.
205
3 Hax, A. C. and Meal, H. C., "Hierarchical Integration of Production Planning and
`Scheduling", in, M. A. Geisler (Ed.), Studies in Management Sciences, Vol. 1, Logistics,
North-Holland-American Elsevier, Amsterdam, 1975.
4 Axsäter, S., "On the Design of the Aggregate Model in a Hierarchical Production Planning
System", Engineering and Process Economics, Vol. 4, 1979, pp. 89-97.
5 Zipkin, P., "Exact Approximate Cost Functions for Product Aggregates", Management
Science, Vol. 28, 1982, pp. 1002-12.
6 Wijngaard, J., "On Aggregation in Production Planning", Engineering Costs and Production
Economics, Vol. 6, 1982, pp. 259-65.
7 Jonnson, H.. "Simulation Studies in Hierarchical Systems in Production and Inventory
Control", PhD Thesis, Department of Production Economics, Linkoping University, 1983.
8 Axsäter, S. and Jonsson, H., ''Aggregation and Disaggregation in Hierarchical Production
Planning", European Journal of Operational Research, Vol. 17, 1984, pp. 338-50.
9. Bertrand, J. W. M. and Wortman, J. C., Production Control and Information Systems for
Component Manufacturing Shops, Elsevier Scientific Publishing Co., Amsterdam, 1981.
10. New, C. C., Managing the Manufacture of Complex Products, Business Books, London,
1977.
11. Plossl, G. W. and Welch, W. E., The Role of Top Management in the Control of Inventory,
Reston Publishing Co., Reston, VA, 1979.
12. Berry, W. L., Vollmann, T. E. and Whybark, D. C., Master Production Scheduling;
Principles and Practice, APICS, 1979.
13. Bertrand, J. W. M., ''A Hierarchical Approach to Structuring the Goods Flow Control in
MultiPhase Production Systems", IFIP Working Group 07 Workshop, Copenhagen,
August 29-31, 1984. (To be published in Proceedings.)
14. Bertrand, J. W. M. and Wijngaard, J., "The Structuring of Production Control Systems",
Report ARW 03 THE BDK/ORS/84/10, Eindhoven University of Technology, 1984.
15. Anthony, R. N., Planning and Control Systems: A Framework for Analysis, Harvard
University, Cambridge, Massachusetts, 1965.
16. van Beek, P., ''An Application of Dynamic Programming and the HMMS Rule on Two-
level Production Control", Zeitschrift fur Operations Research, Vol. 21, 1977, pp. B133-
BI41.
17. Meal, H. C., ''A Study of Multi-Stage Production Planning", in Hax, A. C. (Ed.), Studies in
Operations Management, Chapter 9, North-Holland, Amsterdam, 1978.
18. Bemelmans, R. P. H. G. and Wijngaard, J., ''A Capacity Oriented Approach of Production
and Inventory Systems", Proceedings of the Second International Symposium on
Inventories, 1982, pp. 595-607.
206
19. Bemelmans, R. P. H. G., "On the Capacity Aspect of Inventories", PhD Thesis,
Department of Industrial Engineering, Eindhoven University of Technology, 1985.
20. Wijngaard, J., ''Aggregation and Decomposition in Logistic Control of Multi-Stage, Multi-
Product, Production/Inventory Systems", in J. P. Brans (Ed.), Operational Research' 84,
1984, pp. 551-65.
21. Williams, T. M., "Special Products and Uncertainty in Production/Inventory Systems",
European Journal of Operational Research, Vol. 15, 1984, pp. 46-54.
22. Billington, P. J., McClain, J. O. and Thomas, L. J., "Capacity Constrained MRP Systems",
Management Science, Vol. 29, 1983, pp. 1126-41.
23. Simon, H. A., The Sciences of the Artificial MIT Press, Cambridge, Massachusetts, 1981.
24. Bertrand, J. W. M., "Balancing Production Level Variations and Inventory Variations in
Complex Production Systems", Report BDK/KBS/84/04, Eindhoven University of
Technology, 1984.
25. Bemelmans, R. P. H. G., ''A Single-Machine Multi-Product Planning Problem under
Periodic Review", Report BDK/ORS/83/02, Eindhoven University of Technology, 1983.
26. Silver, E. A., "Aggregate Production Planning: State of the Art", Production and Inventory
Management, 1st quarter, 1972, pp. 15-40.
27. Boskma, K., "Medium Term Planning of Production and Sales", PhD Thesis, University of
Groningen (in Dutch), 1979.
28. Wortmann, J. C. and Wijngaard, J., "MRP and Inventories", Report BDK/ORS/84/08,
Eindhoven University of Technology, 1984.
29. Wijngaard, J., "Capacities in Inventory Control", Report BDK/ORS/84/06, Eindhoven
University of Technology, 1984.
207
A NOTE ON THE DESIGN OF LOGISTICS CONTROL SYSTEMS
J.W.M. Bertrand 1. Introduction
In this paper we present a systematic approach to the process of designing logistics control
systems. Logistics has been described as "...the managerial problems raised by production,
inventory and distribution... " (preface of: Graves et al., 1993).
Since a long time logistics is a widely researched area. Among the first published results in this
field were the determination of the economic order quantity (Harris, 1913) and the
determination of the number of servers needed in order to achieve a certain service rate (Erlang,
1917). Since then, and in particular after World War II, a rich flow of research has emerged in
fields such as inventory control, production control, capacity control, and supply chain control,
resulting in an impressive number of methods and techniques for logistics decision making for
all kinds of decision problems. In parallel, research has been done on the characteristics of
logistics control situations, leading to production typologies such as the product-process matrix,
and tables that show what type of logistics techniques is suited for solving what type of decision
problem in what type of production situation. Overviews of these results can be found in
textbooks such as Silver et al. (1998), and Hopp and Spearman (2000).
A lot of models and techniques are available nowadays to support the design of logistics control
system in real life production situations. However, the logistics control system design process
itself has received much less attention over the last decades. A thorough discussion of the
general logistics control design problem can be found in Bertrand, Wortmann, Wijngaard
(1990). The authors provide a design typology, a generic hierarchical control framework, and an
explicit account of the design questions that need to be answered in the design process.
However, no guidance is given regarding how these design questions should be answered.
Examples of detailed designs for the logistic control of specific real life situations can be found
in Hax and Meal (1975), Bertrand and Wortmann (1981), and Schneeweiss and Schröder
(1992). However, the design process discussed in these reports is largely implicit and specific
for the real life situation at hand. No generic guidance is given. In Schneeweiss (2003), building
on concepts from general systems theory (see Mesarovic et al., 1970), a detailed account is
given of the anatomy of hierarchical production control systems, providing a concise
208
terminology for describing production systems and their control. However, this account does not
discuss the problem of how to design the control system.
Research on the process of product design is ample, and has a long tradition. It has resulted in
different types of output streams, ranging from explanatory research into the organization of the
design process, and into the conditions that lead to an effective and efficient design process (e.g.
Allen, 1987), to normative research about how to structure the design process (e.g. Suh, 1990),
and to descriptive research about how design approached have played a role in the evolution of
products and industries (Baldwin and Clark, 2000).
In this paper we take the formal prescriptive rules for structuring the design process in Product
Design as used in Suh (1990) and reported in Baldwin and Clark (2000) as a starting point for
developing similar formal prescriptive rules for structuring the process in Logistics Control
Systems Design. We start in section 2 with a short presentation of design principles for product
design. Then, in section 3, we present a formal description of the logistics control design
problem. Next in section 4 we give characteristics of the logistics control design problem. In
section 5 we discuss the Production Unit Control design problem, and in section 6 we discuss
the Goods Flow Control design problem.
2. Design rules for product design.
Many papers and books have been published that give rules for how to design a complex
product. In this paper we use the terminology and rules presented in Suh (1990), and Baldwin
and Clark (2000). What follows is a short summary of their descriptions of the design process.
Design starts with specifying the Functional Requirements (FR) for the product to be designed.
The design process involves the search for a product, specified by the Design Parameters (DP),
that satisfies these functional requirements. This process consists of the execution of one or
more design tasks, a design task being the task to specify one or more of the design parameters.
From a design perspective, an artifact (or product) is characterized by an individual set of the
design parameters. The categories in the set, such a material, length, width etc., make up the
dimensions of the design space of the product. A particular point in this design space therefore
specifies a particular product.
The space of all possible designs can be extremely large, ill-structured, and difficult to search.
Therefore Baldwin and Clark (2000) distinguish hierarchical design parameters. For these
parameters the designer's choice has the character of a logical switch, and a "yes" brings a new
209
set of design parameters into existence. Decisions on a hierarchical design parameters precede
decisions on the non-hierarchical design parameters, and are used to delimit and bound the
space of designs. As a result the design problem gets manageable, given the knowledge and
resources available.
Design would be relatively easy if design parameters would be independent of each other, and
each design parameter would only influence one functional aspect. However, design parameters
often depend on each other and have interacting effects on functionality. Thus, the success of
the design process largely depends on the knowledge in the design team about these
interdependencies and interactions.
Hierarchical relationships and interdependencies among design parameters can be mapped in
the Design Structure Matrix (DSM), (Steward 1981, 1981a, Ulrich and Eppinger, 1995, Smith
and Eppinger, 1997).
In a DSM, the individual design parameters are assigned in the same order to the rows and
columns of a square matrix. For each row, one puts a mark (x) in the element (a, b) of the matrix
if the specification of design parameter a requires knowledge about design parameter b.
Specifying the value of a design parameter is a design task. Thus, the DSM of a design problem
corresponds one-to-one with the task structure Matrix (TSM), (Smith and Eppinger, 1997). The
TSM reveals the hierarchical relationship between design tasks and the interdependencies, that
can be modelled as precedence relationships. The TSM shows how to organize the design
process. Often groups of design tasks can be identified where design tasks within a group are
mutually interdependent, and interdependencies between design tasks belonging to different
groups are sparse or don't even exist. Design tasks belonging to different groups can be
performed relatively independently. However, for design tasks within a group, some form of
interdependence-breaking coordination is required to avoid endless iterations. This can be done
by assigning each group of strongly interdependent design tasks to a separate design team, with
short communication lines and high information exchange within the design team. The
remaining interdependencies between groups of design tasks can be resolved by, for instance,
adding constraints for design rules to individual design parameters, based on design knowledge
about the interactions between them. However, since design knowledge generally will be
imperfect, interactions in the design of complex products cannot be entirely avoided.
Baldwin and Clark (2000) introduce the concepts of modularity and design rules in their
discussion of how to manage the complexity of the design process. Modules are units in a larger
system that are structurally independent of one another, but work together to achieve to the
overall functional requirements. The first step in the design process is therefore to define a
210
framework (or a conceptual design) that allows for independence of structure and integration of
function. They state that ".....A complex system can be managed by dividing it up into smaller
pieces and looking at each one separately. When the complexity of one of the elements crosses a
certain threshold, that complexity can be isolated by defining a separate abstraction that has a
certain interface. The abstraction hides the complexity of the element; the interface indicates
how the element interacts with the larger system.....".
In the process of defining modules, the systems architect, overseeing the interdependencies, first
introduces design rules that are imposed on each of the design teams as constraints to create
independence among the modular design tasks. Modular independence is further achieved by
information hiding during the modular design process.
Information hiding is making sure that each modular design team realizes its design without
using any information about how any of the other design teams solved its design problem.
Information hiding guarantees that the design modules can perform independently from one
another. If furthermore guarantees that, in the future, the design of each module can be
improved independently, as long as the systems design rules are respected. By respecting the
design rules and the defined interfaces, modules can evolve individually and independently to
improve the functioning of the product as a whole. Design rules, modularity, information hiding
and interfaces are therefore important principles for developing products that can relatively
easily be further improved in future (re)designs.
3. The Logistics Control System Design problem
In this section we give a short description of the problem of designing Logistics Control
Systems. We start by defining what we consider to be given at the start of the design project.
Then we define the functional requirements and the design parameters and discuss
characteristics of the Logistic Control Systems design problem.
For a given Logistics Control System design problem, we assume that the scope of the total
system to be controlled is known. The boundaries of the system are set defining by:
- the set of end-products that are produced by the system
- the set of raw materials that are input to the system
The internal structure of the system to be controlled is given by:
211
- the bills of material that relate each end-product via a number of specific intermediate
products to the raw materials used by each end-product
- the processes, that convert (raw) materials into higher value states (including the
intermediate products stated in the bills of materials) and finally into the end-product
state.
- per process, the set of resources types needed to execute the process, the setup time, the
change-over costs, the amount of resource needed, the processing yield, etc.
The functional requirements regarding the performance of the Logistics Control System can be
expressed in variables such as:
- the delivery performance relative to demand for end-items, possibly conditioned on
specific patterns in end-item demand,
- total logistics related costs; these are costs that vary as function of logistics decisions
taken in the control system.
The design parameters of the Logistics Control System design problem are the following:
- for each of the resource types, the decision function that determines the availability of
the amount of resource of this type, as a function of time.
- for each process, the decision function that determines (directly or indirectly) the time
during which the process is executed to bring materials into higher value states.
For real life production systems, solving the Logistics Control System design problem can be a
formidable task because:
- the design space, having as many dimensions as the sum of different resource types and
processes, can be very large.
- many interdependencies exist between these design parameters.
Fortunately enough, very many combinations of possible decision functions lead to evidently
poor performance and can therefore be easily ruled out. However, design parameters in
Logistics Control System design do not pertain to features for which a choice must be made
from a well defined range of values, but pertain to decision functions, operating on the state of
the system. Very many different decision functions may be reasonable candidates for each
particular decision problem (design parameter). Thus for complex design problems the
remaining space of all decision functions that are reasonable candidates will still be
overwhelmingly large. It follows that, like in product design, the first phase in the design
process should the determination of the control system architecture. The control system
212
architecture serves to break many of the interdependences between design parameters and create
(modular) sets of design tasks that can be solved independently, and which, when solved, also
solve the overall design problem. In the next section we will show how the principles behind,
modularity, abstraction, information hiding and interfaces could be applied to the design of
Logistics Control Systems.
4. Modularity, information hiding, abstraction, and interfaces
Like in product design, the person that develops the logistics control system's architecture must
be very knowledgeable in the field of logistics control systems design. He/she must have
detailed knowledge of the interdependencies that may exist between design parameters as a
function of the characteristics of the products, the processes, and the resources. A design
structure matrix can be used to map the dependence relationships between the design
parameters. In production systems, the dependence relationships between resources and
processes mainly stem from the material-requirements relationships between the items produced
by the processes and the capacity-requirements relationships between processes and resources.
Listing the processes and the resource types in the same order along the rows and columns of a
square matrix, a design structure matrix can be determined, the elements of which indicate
whether the decisions function in the row depends on the decision function in the column. Such
a matrix will reveal that each process decision function depends on all process decision
functions upstream in the material flow (because the materials produced by these upstream
processes must be available for executing the process), and on the resource decision function for
all the resource types needed for its own execution, and for the execution of all these upstream
processes. This points towards a very high level of interdependency between the design
parameters (decision functions) in Logistics Control System design. However, in most
production systems material flows are unidirectional, and in most production systems
specialized resource types are used for different process types. The control system's architect
can take advantage of these special types of relationship between design parameter when
developing the system's architecture.
Modularization The special structure of the interdependencies in production systems suggests to form modules
by grouping specific sequences of process into sets { } 1..., ,iP i N= and to group specific
resource types into sets { } 1,..jR j N= such that the processes in set iP can be executed with
213
the resources in set iR , and the resources in set iR are only used for executing processes in iP . A
combination of processes and resources { },i iP R can be defined as a production unit (PU).
Using its resources, a PU executes a sequence of processes on raw materials, in order to produce
the PU-end items. A PU therefore can be considered as a subsystem of the production system.
Decomposing the overall production system in N such sub systems { }, , 1,..., ,i iP R i N= is the
first step in creating a modular design. Other modules may be necessary, for instance the system
that coordinates the Production Units to achieve the overall systems performance could also be
defined as a module. We refer to this systems as the Goods Flow Control System. We first
concentrate on modular PU's.
Information hiding requires that each PU functions independently from the internal state or
functioning of any other part of the system. Interactions between PU's are confined to the
interfaces and are based on abstractions that describe the performance of the PU in response to
inputs. For instance common abstractions regarding the performance of a PU are its lead time
and its capacity. Defining these abstractions precedes the specification of the PU design task,
and allows for the design task to be expressed as, for instance:
- specify capacity decision functions and processing decision functions that guarantee that
each order to produce up to X items of any PU-end item is completed within Y days after
release, on the condition that total number of orders released per day does not exceed Z,
and that raw materials are always available.
This description also reveals the nature of the interfaces of this (to be designed) production unit
with its environment:
- an interface with the system that release production orders to the unit
- an interface with the system that supplies raw materials to the production unit
- an interface with the system to which completed products are delivered.
Modular functional requirements
We have seen that abstraction plays an important role in defining the interfaces and the
performance requirements for each of the PU modules. The architect has to define interfaces
between the Production Units, and the interface between the logistics control function of each of
the PU's and the Goods Flow Coordination System. The Goods Flow Control System thus
coordinates performance of the Production Units to achieve the logistics performance targets for
the overall production system, and can only do so via the interfaces. Thus, if N Production Units
214
are defined, then at least 1N + Control Systems must be designed and integrated in order to
solve the overall Logistics Control System design problem. Integration is the process of
connecting the (modular) subsystems, testing their combined performance, and fine-tuning the
separate designs in order to achieve end-item performance according to specification.
The functional requirements of each PU are expressed in the volume and timing of delivering
the of PU-end items, and the total logistics related costs in the PU. Before specifying functional
requirements the systems architect first has to identify the constraints on the possible
performance of a PU.
Knowledge about constraints can be obtained from the reference models of production systems
and their control available in literature. Reference models characterize specific types of
production systems such as job shops, flow shops, flow lines, assembly lines, project shops, etc.
The architect can use these reference models to select appropriate global control models for
each PU, and to make initial decisions on the interfaces between the modules and Goods Flow
Control, the performance variables per module, and the logistics related cost budget per module.
We can distinguish three types of interfaces:
- material supply interfaces
- control interfaces
- product delivery interfaces
Material supply and product delivery interfaces are physical and are given as soon as the
boundaries of the PU's are defined. Also PU-end-items result from this choice.
Control interfaces are determined thereafter. Control interfaces are related to the way in which
the required performance of a PU is specified. Different options are available here. For instance,
the performance of a PU could be expressed as an agreed time to produce an order for the PU-
end items issued by Goods Flow Control, (if the PU works on a make-to-order basis), or in a
percentage of demand for each of the PU-end items to be delivered out-of stock, (if the PU
works on a "manufacturer managed inventory" basis), under a constraint on the volume. The
two choices lead to different interfaces between modules in a design. With the make-to-order
interface, Goods Flow Control controls the stock of PU-end items by issuing production orders
to the PU; with the PU-managed inventory, the PU itself directly controls the stock of PU-end
items in response to demand PU-end items, in order to achieve a performance target specified
by Goods Flow Control.
215
5 The Production Unit Control design problem
We next discuss the design of Production Unit Control Systems. The PU design space consists
of the decision functions regarding the resource types and the processes in the PU. The PU
functional requirements specifications consist of the target values specified by the architect
regarding volume and timing of production of PU-end items in response to specific patterns in
demand for PU-end items, under a constraint on logistics related costs assuming raw materials
are ample. We assume that the design principle of information hiding is used. This implies that
each PU can achieve its performance without having information about how the design problem
of any other PU is solved, and how the design problem for Goods Flow Control is solved.
The actual output of a PU, however, may be affected by the output of other PU's, for instance if
a PU involved in preceding processes does not deliver sufficient raw materials. Such
interdependencies are managed at the Good Flow Control level.
An important performance measure for a PU control system is its logistics related costs. The
major components of logistics related costs are:
- costs of capacity variation
- costs of capacity slack
- costs of work-in-process
- costs of setting up a process
- costs of running the production unit system.
Depending on the costs allowed, a production unit can realize very different performances. In
the limit, if there is no constraint on costs, a production could produce any amount of each of its
PU end items with a throughput time that is equal to the raw processing time. On the other hand,
if there exist hard constraints on costs and capacity utilization, the throughput time would
include waiting times for resources to be available, and constraints would be put on production
batch sizes. It follows that, when specifying the PUC design problems, the system architect
must estimate what combination of performance and costs are possible (without of course first
solving each of the many design problems that could be specified), and specify a design
problem that he/she estimates to be solvable. It is very likely therefore that the design problem
specified contains slack, that is, there will exist solutions to the design problem that deliver the
specified performance at lower costs then specified by the costs constraint. If this would not be
the case, the system architect would demand the design problem to be solved to optimality, and
we know that practically all real life design problems are computationally intractable. Design
therefore is not about finding optimal solutions but about finding satisfactory solutions, which
of course can be improved gradually over time.
216
To realize his design, the PUC designer must know in detailed terms how, for each of the
processes and resource types in the PU, the above costs components depend on certain aspects
of the decision functions. For this design, the designer builds detailed models of the Production
Unit and searches the PU design space for finding decision functions that satisfy the functional
requirements under the given costs constraint. Depending on the type of production system (job
shop, flow shop, flow line, etc.) the number of design parameters may be large or small. The
designer will try to reduce design complexity by decreasing the degrees of freedom. This can be
done by for instance by coupling the processing times of processes such that equal amounts of
items are produced (same production batch size for consecutive processes) or by coupling
processing times of processes to resource availability (continue processing as long a resource is
available), and control amounts produced by controlling resource availability.
Depending on the total costs constraint, certain constraints may imposed on the decision
functions. Examples of constraints that the designer might put on the decision functions are:
- this process should never be performed for less than x time units,
- process I on resource A should never be performed directly after process J,
- availability of resource K should never be increased with more than y units per week,
- on resource R, at maximum z units of time per week should be spent on changing over,
- average overall utilization of resources per period should at least be equal to r % of
available capacity.
Such internal constraints are translated by the designer into operational constraints on the
performance of the PU level, for instance constraints on production volume, on volume changes,
on production order release patterns, and on production order throughput times. The operational
constraints per PU specify the conditions under with the PU can realize its performance and are
input to the design of the Goods Flow Control System.
6 The Goods Flow Control Design Problem
Goods Flow Control coordinates the logistics performance of the PU's such that the overall
logistics control targets are obtained within the given budget. The design parameters (decision
functions) for Goods Flow Control pertain to the output requirements of PU-end items. These
are artificial design parameters, that is, they result from the definition of the PU's. Goods Flow
Control does not directly influence the availability of resources, nor the execution of the
processes. These decision variables are controlled by the PU's. Suppose that for each of the PU's
the output that can be obtained under specified conditions (the operational constraints) is known
217
at the start of the design of the Goods Flow Control System. Then, the output of a PU is
specified in terms of the PU-end items. Thus the function of Goods Flow Control is to select,
from among all possible PU's-end items output values, the ones that together realize the system's
overall output requirements, while respecting the operational constraints of the PU's and the
constraint on total logistics related costs. Various options are available for specifying output
requirements per PU. One option, which is used in a make-to-order setting, is to specify orders
for PU-end items that have to be completed within a specific lead time. In this case the orders to
be delivered would be the decision variable of Goods Flow Control. Another option, which is
used in a manufacturer managed inventory situation, is to specify a demand forecast for PU-end
items and require that the PU can deliver out of stock on the conditions that the actual demand is
within a certain bandwidth around a specific demand forecast. In this case the demand forecast
would be the decision variable of the Goods Flow Control. Under both options, Goods Flow
Control must respect the constraints per PU, and must check for the validity of the conditions
under which these outputs within the constraints can be achieved. Since the output of each PU
may constrain future outputs of other PU's, Goods Flow Control must take into account the
material requirements relationships between PU-end items. Moreover it must take into account
the operational constraints per PU. The main difficulty in the Goods Flow Control design
problem results from the operational constraints specified by the various PU's in the system.
Different PU's may specify very different operational constraints; for instance one PU may
formulate a constraint of variations in the volume of total production overtime, while another
PU may allow for practically unlimited volume fluctuations but impose a serious constraint on
the production batch size. The Goods Flow Control design problem is to specify a decision
function that specifies output requirements for the PU's such that, if realized by the PU, together
lead to production system output that satisfies the output requirements. The output of a PU is
constrained by the availability of the raw materials needed for its processes. These raw materials
are produced by other PU's. Since all PU's in a production system can differ in their
responsiveness (operational constraints), Goods Flow Control may need to build in some slack
in the time-phasing of the materials inputs of a PU and the materials outputs of the PU's feeding
this PU. Then, on the short term, inputs and outputs of PU's must be decoupled and stocks of
PU-end items act as decoupling points in the material flow in the production system.
Decoupling in the material flow may also be needed for two other reasons. First going upstream
the material flow, uncertainty about demand for items increases. Goods Flow Control may want
to build in some slack in availability of items to compensate for this (increasing) uncertainty in
demand. A well-known demand uncertainty decoupling point is the customer-order-decoupling
point, where downstream all production processes are executed to satisfy customer orders
already placed and upstream all production processes are executed in anticipation of future
218
customer orders. Decoupling may also occur for items which result from processes with
stochastic yield.
Second, items may exist which are common components for different items downstream. For
such items, Goods Flow Control may want to allocate the amounts produced only after
production, resulting in a decoupling of production of the common component from production
of the items that consume that common component. For a more detailed account of decoupling
points we refer to Bertrand et al. (1990).
For the Goods Flow Control design problem, the designer may evaluate specific control models
based on knowledge of reference models available in literature. For instance, if, for each of the
PU's defined in the system, capacity can always be made available such that output of each PU
is only constrained by available raw materials and by a deterministic manufacturing lead time,
echelon stock control or MRP-I might be considered as a Goods Flow Control mechanism.
As another example, suppose that some PU's are seriously constrained in the output volume that
can be obtained, while others are not, or that PU's strongly differ in the costs involved in
changing the output volume over time. In such a case, the designer might use a chase strategy
for PU's with high resource-flexibility and a level strategy for PU's with low resource-
inflexibility. This would require temporary advancing of output of resource-inflexible PU's,
relative to the output of resource flexible PU's, and would lead to the temporary buildup and
builddown of stocks for specific PU-end items. Such PU-end items would temporary store
"capacity".
Similar control mode choices may follow if PU's differ in production order batch size
requirements (leading to cycle stocks for specific PU end-items), or if certain PU's operate
processes with uncertain processing yields (leading to or the need for safety stock of specific
PU-end items).
References - Allen, T. (1987). Managing the Flow of Technology, Harvard University Press, Cambridge,
Mass.
- Baldwin, C.Y. and K.B. Clark (2000). Design Rules Volume 1. The Power of Modularity.
The MIT Press, Cambridge, Mass.
- Bertrand. J.W.M., J.C. Wortmann, and J. Wijngaard (1990). Production Control: A
Structural and Design Oriented Approach, Manufacturing Research and Technology 11,
Elsevier, Amsterdam.
219
- Bertrand, J.W.M. and J.C. Wortmann (1981). Production Control and Information Systems
for Component Manufacturing Shops. Studies in Production and Engineering Economics 1,
Elsevier, Amsterdam.
- Erlang, A.K. (1917). The solution of some problems of significance in automatic telephone
exchanges. P.O. Electric Engineering Journal, 189.
- Graves, S.C., A.H.G. Rinnooy Kan and P.H. Zipkin (1993). Logistics of Production and
Inventory, Handbooks in Operation Research and Management Science, Vol. 4,
Amsterdam, North Holland.
- Harris, F. (1913). How many parts to make at once. Factory, The Magazine of Management,
10, pp. 135-136.
- Hax, A.C. and D. Meal (1995). Hierarchical integration of production planning and
scheduling. In: Geisler (ed.). Logistics, TIMS Studies in the Management Sciences, North
Holland, Amsterdam.
- Hopp, W.J., and M.L. Spearman (2000). Factory Physics, second edition, McGraw-Hill,
Boston.
- Mesavovic, M.D., Macko, D., and Y. Takahara (1970). Theory of Hierarchical Multilevel
Systems. Academic Press, New York.
- Schneeweiss, C. (2003). Distributed Decision Making, second edition, Springer.
- Silver, E.A., Pyke, D.F., and R. Peterson (1998). Inventory Management and Production
Planning and Scheduling. J. Wiley and Sons, New York.
- Smith, R.P. and S.D. Eppinger (1997). Identifying Controlling Features of Engineering
Design Iterations, Management Science 43 (3): 276-292.
- Suh, N.P. (1990). The Principles of Design. Oxford University Press, New York.
- Steward, D.V. (1981). System Analysis and Management: Structure, Strategy and Design.
Petrocelli Books, New York.
- Steward, D.V. (1981a). The Design Structure System: A Method for Managing the Design
of Complex Systems. IEEE Transactions in Engineering Management 28, 3, 71-84.
- Ulrich, K.T. (1995). The Role of Product Architecture in the Manufacturing Firm,
Research Policy 24: 419-440
- Ulrich, K.T. and S.D. Eppinger (1995). Product Design and Development. McGraw-Hill,
New York.
220
221
Promovendi Prof. W. Bertrand (1e Promotor):
Corné Dirne (1990)
Production control for flexible automated manufacturing stations in low volume
component manufacturing.
Jan Fransoo (1993)
Production control and demand management in capacitatedf low process
industries
Petra de Groot (1993)
Decision support for admission planning under multiple resource constraints..
Ton van de Wakker (1993)
Throughput time control and due date reliability in tool and die shops.
Jan Vissers (1994)
Patient flow bases allocation of hospital resources.
Werner Rutten (1995)
The use of recipe flexibility in production planning and inventory control.
Harrie de Haas (1995)
The coordination of initial stock and flexible manpower in repairable item
systems.
Paul Stoop (1996)
Performance management in manufacturing; a method for short term
performance evaluation and diagnosis.
Henny van Ooijen (1996)
Load-based work-order release and its effectiveness on delivery performance
improvement.
222
Wenny Raaijmakers (1999)
Order acceptance and capacity loading in batch process industries.
René Hanssen (2000)
Concurrent engineering vanuit beheersingsperspectief.
Joris Keizers (2000)
Subcontracting as a capacity management tool in multi-project repair shops.
Kim van Oorschot (2001)
Analyzing radical NPD projects from an operational control perspective.
Bogdana Dragut (2003)
A Markovian approach to the mathematical control of NPD projects.
Cristina Ivanescu (2004)
Order acceptance under uncertainty in batch process industries.
Pieter van Nyen (2005)
The integrated control of production-inventory systems.
Gergely Mincsovisc (2008)
Studies on tactical capacity planning with contingent capacities.
Cagdas Büyükkaramikli (2011)
Preliminary title: Cost Saving Prospects of Capacity Flexibility in a Maintenance Service Provider Environment
223
Afstudeerders Prof. W. Bertrand (1e begeleider):
Geenen A.L.J. (1970) Het bouwen en gebruiken van een simulatiemodel voor een deel van het TL-buizen fabricageproces ten behoeve van een investeringsbeslissing. Plante E.A. (1971) Toepassing van een planning-simulatiemodel. Abercrombie A.A. (1971) Simulatie van regels voor afstemming tussen hoogovens en staalfabrieken. Ramaekers J.M.G. (1973) Defectenonderzoek. Bakker A.C.M. de (1974) Voorraadbesturing en programmabesturing in een eenvoudige situatie. Strijp P.W.A.C. de (1974) Productiebeheersing in de halfgeleidersfabriek. Vlerken P.M.W.H. van (1975) Onderzoek productie besturing. Beekman P.R. (1976) Onderzoek betreffende productiebesturing in een complexe productie situatie d.m.v. simulatie. Henselmans J.H.M. (1976) Het ontwerp van productiebesturingsprogrammatuur binnen een database. Pol F.X. van de (1976) Produktiebesturing: een analyse van een productiebesturing van het depot vliegtuigmateriaal van de Koninklijke Luchtmacht. Steffens E.J.A. (1976) Productienivoregeling in een halfgeleiderfabriek. Brouwer A. (1977) Toepassing van optimale besturingstheorie op een financieel ondernemingsmodel. Weegen E.M. van de (1977) Job Shop Simulatie. Kessels E.G.M (1978) Aggregaatplanning. Lebbink G.J.M (1979) Aggregaatplanning. Koninkx H.J.J (1979) Onderzoek naar een systeem voor het kunnen beheersen van de doorlooptijden op een productieafdeling van halfgeleiders.
224
Govers J.G.J (1980) Mogelijke verbeteringen bij een Offsetdrukkerij. Peter J.M. van de Griendt (1982) Produktiebesturingssysteem voor de tachtiger jaren bij DAF speciale produkten . Broeke A.S van de (1982) Mogelijke verbeteringen van de productiebesturing in een projectorganisatie. Hoftijzer R.W. (1983) Ontwikkeling van het Hoofd Produktie Programma voor de ELMO-divisie van Fokker B.V.. Prins F.J.B. (1983) Een onderzoek naar een verbeterd shop Floor control systeem voor Oldelft te Delft. Candel J.G.M (1983) Op weg naar korte en betrouwbare doorlooptijden binnen de Euroline. Jonge A.H. de (1984) Analyse en verbetering van de productiebesturing bij Salland Offsetdruk. Kempen M.H.G. (1984) Fundament voor de magneetkoppenproductie. Hecker G.H.A.H.J. (1984) Een productiebeheersingsconcept bij de Verenigde Buizenfabriek BV. Dirne C.W.G.M. (1984) Een productiebesturingsconcept voor de Mini Digital Cassette Recorder. Paumen P.H.J. (1985) Workload Control, the foundation for production control of assembly factories of I.C.’s at Philips. Sande J.C.M. van de (1985) Beheersing werklast in de machinefabriek Acht. Donk H.M.P. van de (1985) De randvoorwaarden van een doorzichtig logistiek beheersingsconcept.
Geurts G.P.G.M. (1985) Master Planning; een eerste aanzet tot de beleidsmatige beheersing van materiaal en capaciteit. Woezik B.M.T. van (1985) Beheersing van de produktie in een autonome cel bij DAF Trucks. Swinkels M.P.T.M. (1985) Een aanzet tot prioriteitsbepaling voor een flexibele produktiemodule in een job shop omgeving mbv simulatie.
225
Muller M.A. (1986) Een kleine (serie) stap op weg naar JIT-produktie. Gubbels M.A.G.M. (1986) Werklastbeheersing bij DAF Special products. Giesen M.T.M. (1986) Productiebeheersing in een kleine machinefabriek. M.P.L. van der Meulen (1987) Ontwerp en invoering van een afdelingsbeheersings- systeem; een pilot-studie naar werklastbeheersing in de afdeling “Klein Vlak” van de Philips Machinefabriek-M. P.B.A. Gosselink (1987) Doorlooptijdbeheersing: goed bestuurde doorlooptijden geeft flexibiliteit.
Jeroen J.M. Wortelboer (1988) Beter beheren door strukturen: het ontwerpen van een integraal voorraadbeheersingssysteem, geoptimaliseerd naar kosten en servicegraad.
R. Middelkoop (1988) Beheersing werklast in een groepentechnologische georiënteerde fabrikageafdeling van Stork Brabant. Tromp J.W.T. (1989) Doorlooptijd verkorting binnen een gereedschapmakerij. Foppe Atema (1989) De invoering en evaluatie van OPT in de metaalwaren van Philips Leeuwarden: de ontwikkeling van de bijbehorende performance indicators.
J. Verdier (1989) Industriële afnemers en leveranciers: samen apart?
T.V.M. Heerkens (1989) Doorlooptijdbeheersing van een reproduktieproces.
J.C. de Bruin (1990) Logistieke verbeteringen bij het Consumer Electronics Overseas Suport Center (CE-OSC) in Veldhoven.
R.A. Frijlink (1990) Globale capaciteitsplanning.
J.A. van der Burg (1990) Voorraadbeheersing in een make-to-order assemblage- bedrijf. Groot P.L.A.M. de (1990) “Goederenstromen in ziekenhuizen”?. S.J. Paauwe (1991) Crystal stock control in assembly centres. Wim Provoost (1991) Doorlooptijdbeheersing bij LIPS Sloten.
226
M.J. Verweij (1991) The limits of control; a new control structure for the viton-plant. Nicole Rutten (1991) Opslag en voorbewerken van ruwe olie op Europoort; onderzoek naar verbeteringen voor planning. J.M.J. van den Bosch (1991) Slagvaardigheid door productiebeheersing in een make- to-order “sellling capacity” organisatie. Marco van Duijnhoven (1992) Van logistieke geldstroomdiagnose naar een reductie van het onderhanden werk; aanzet tot een logistiek beheersingsconcept. Gilbert M. Geurtjens (1992) Verbetering van de produktieprestatie in de motoren- assemblage bij Scania Nederland B.V. R.F. Balk (1992) Een haalbare planning voor de smelterij. Arjan van Bragt (1993) Logistieke parameters in de middellange termijnplanning: een case-study naar de mogelijkheden om de leverbetrouwbaarheid van een sourcing unit te verhogen. C.S. van Daal (1993) Productiebeheersing bij de Barneveldse Drukkerij en Uitgeverij. Jack E.L. van Lieshout (1993) Flexibel reageren: onderzoek naar een nieuwe methode van productiebeheersing voor Van Geel Kanalisatiesystemen. J.W.J. Verhoeven (1993) Productieplanning op basis van verwacht aanwezige mancapaciteit. B.J. Dirkx (1993) Ontwerp van een nieuw logistiek besturingsmodel. R. Hoornstra (1993) Voorraadbeheer bij Quaker Chemical B.V.; een alternatief beheersingssysteem. J.M. Sonke (1994) Uitbesteden van onderhoud: de ontwikkeling van een beslissingsmethode. L.M. Klootwijk (1995) Productiebeheersing bij Interpress B.V. Niels van Bladeren (1998) Ijskoud de beste?: eisen aan consumer service en flexibiliteit in een ijsfabriek. J.A.C.A. Aerts (1998) Logistieke beheersing in de semi-procesindustrie: een nieuw logistiek concept voor MBI.
227
School J.M. (1998) Beheersing van de productontwikkeling bij Stork Digital Imaging; Analyse en verbetering. J.P. de Visser (1999) Beslissingsondersteuning bij de productieplanning in de semi- procesindustrie. Jitse Marree (2000) Een herontwerp van de logistieke besturindsstructuur. Houdt M.R. van (2005) Is flexibility the key to better performance? Distributing the skills for Nashuatec Technical Services E.J.A. Jansen (2007) Task valuation and resource allocation to optimize value delivered in product and process improvement projects. N. Karabash (2008) Impact of learning-by-doing on efficiency of multi-skills employees performing complex tasks. F.E.V. Lievens (2009) Factors affecting the learning effect in the production environment of ASML suppliers. Rooy J.C. van (2010) Logistical information sharing and collaboration between ASML and its supply chain.
228
229
Ontwerpers LMS opleiding, 1e begeleider Bertrand
Abspoel S.J., Faster Payback; a logistics control structure for investments in network capacity at KPN Telecom.
Acker V.M.V. van, SHARLI: Stockage en Hauteur des Aciers et Réorganisation de la Logistique Interne.
Albers E.I., Ontwerp van een communicatiestructuur over capaciteitsmanagement in de keten.
Bollen G.L.L., Doorlooptijdverkorting van ASM-L.
Boven D. van, Patiëntenlogistiek binnen het Máxima Medisch Centrum.
Bressers C.H.P., Herontwerp van de bepaling van de netto benodigde capaciteit t.b.v. de lange termijn capaciteitsplanning van rollend materieel bij de NS.
Budel M.L.H.J., Een logistiek herontwerp voor de bijsturing van de treindienst.
Chen H.Y., Improve the design of forecasts and safety stock
Claus C., Een informatie-architectuur voor de productdivisie Licht.
Cornellana A., Supply chain re-engineering at PMS-XRD; Integrating modular product planning and vendor Managed inventory to enable an upstream shift of the customer orrder decoupling point.
Dekkers K.J.P., A design for KLM Aircraft services’ planning-rostering-indeling chain.
Delnooz R., Bijsturing anders bekeken; Een proces-gericht herontwerp van de bijsturing.
Dijk W.G.M. van, Capaciteitsplanning rijdend personeel bij NS Reizigers.
Driel R.J. van, Voorstellen ter verbetering van de goederenstroombesturing ten behoeve van de ABS-organisatie van DSM.
Du W., Functional design of production planning and control system for Dalsa PI.
Egbers J.M., Bottlenecks at Railway Nodes (Knelpunten op Knooppunten).
Elisa E.B.A., Logistics design project at Assembléon.
Ero W.E., Beslissingsondersteunend model voor de lange termijn capaciteitsplanning voor rollend materieel bij NS.
Graste R.M., Planning and control at Aircraft Services.
Groenewoud R., Herstructurering in de projectindustrie.
230
Habets M.H.J., Introductie van “shippable items” bij Vanderlande Industries.
Hakan K., Online scheduling of towing process to facilitate on-time completion of Gereedstelling Processes.
Hartwijk D., Ontwerp van een logistiek besturingsconcept bij Philips Machinefabriek Acht.
Heribanova E., Redesign of central planning system at NS Cargo NV.
Heuvel J.T.W. van den, Herontwerp besturingssysteem voor Kijfhoek en Havenspoor- lijn.
Hofland M.M., Aandacht voor prestatie: logistieke besturing in een hartkatheterisa- tielaboratorium.
Houben N.N.P.H., A logistics redesign of Schiphol Brede Services to improve the punctuality of the KLM.
Huiskamp E.F., Een ontwerpmethode voor de specificatie van inhaalconstructies op baanvakken.
Kang Y., Logistics design project; MRP (l) simulation environment.
Kapys Y., Detailed design and implementation of a one-piece flow assembly process.
Lavrijsen A., Herontwerp besluitvormingsproces voor de lange termijn planning voor rollend materieel bij NS.
Lierop F.L.G. van, Logistieke besturing in de procesindustrie.
Linssen Ch., Het ontwerpen van een besturingsconcept voor de materieelbijsturing bij de Nederlandse Spoorwegen NV.
Man H. de, Control of repairable service at Philips Medical Systems.
Marín Aznárez P., Decision-making process and advice on outsourcing of electronics production engineering and production activities in Thales Nederland b.v., Hengelo site.
Mikkers B., Ontwerpen van een logistiek model.
Moltzer M.J., Planningsmethodiek reservecapaciteit vloot; Tools voor het beoordelen en plannen van flexibiliteit voor de dienstregeling.
Muller M., Herontwerp van de planningsmethodiek voor allocatie van personeel buffers.
Omachel K., Design of a management control model for cabin quality at KLM.
Peeters M.J.P., Ontwerp en implementatie van een logistiek besturingsconcept voor metaalbestellingen bij Hoogovens Aluminium N.V. (Sidal).
231
Perdaen J.A.G., Implementation design for better coordination between towing and preparation processes.
Pluijmakers A.P.M., Logistiek ontwerp voor de productie van Specifieke Dagen.
Potoms D., Toetsing van de bestuurbaarheid van het NSReizigers vervoersproduct als functie van materieelreserves.
Pottinga P.H., Voorraadallocatie model voor reserve-onderdelen van medische systemen.
Reinartz C.J.M., Afstemming NS Railinfrabeheer-NS Reizigers; Besturing van benutting en onttrekking van railcapaciteit.
Rongen J.M.J., Herontwerp van het optieplanningsproces bij DAF Trucks N.V.
Rooden M.A.H., Herontwerp van de logistieke organisatie in het voortraject van Schlumberger Bladel.
Schiebaan H., Integral fleet service control; Design project at Schiphol Fleet Services.
Simons P.H.W., Rapper rapporteren. Een ontwerp van het kosteninformatieproces.
Slagt A.W., Ontwerp van een planningsmethodiek voor de KLM-dienstregeling.
Spijk M.A.J.C. van, Flexibilisering en doorlooptijdverkorting van het planningsproces voor materieel bij NS Reizigers; Een herontwerp.
Sprenger J., A logistic redesign for the department of Radiation Oncology of the University Hospital Rotterdam.
Tany T.W., Online scheduling of towing and preparation processes to increase the departure punctuality.
Tauber R., NSR logistiek dagplan: beheersing van het planproces.
Verhaegh P.A., Ontwerp supply chain model voor herontwerp ketenbeheersing in het toelevernetwerk van ASM Lithography.
Versluijs A.J., Herontwerp van het besturingsconcept in de tactische planningsfase bij NS Reizigers.
Wiltschek J.T.F., Logistiek in de “bouw”.
Yunita T., Managing effective and efficiënt hub operation; A study of ground time management at KLM Aircraft Services.
Zajaczkowski P.A., Computerized planner support tool for mid-term planning.
Zuijderwijk M., Material coördination and order acceptance under product variety.
232
233
AUTEURSGEGEVENS.
Peter Bos studeerde werktuigbouwkunde (kandidaats) en bedrijfskunde (doctoraal) aan de TU
in Delft. Hij heeft van 1984 tot 1996 in diverse functies bij Fokker Aircraft BV gewerkt en
werkt sinds 1996 bij de KLM. Hij was daar onder meer hoofd maintenance, vice president
operational integrity management, vice president Aircraft Services (divisie ground services) en
is sinds 2011 algemeen directeur van KLM Equipment Services.
Onno J. Boxma holds the chair of Stochastic Operations Research in Eindhoven University of
Technology, and is scientific director of the European research institute EURANDOM.
His main research interests are in queueing theory and its applications to the performance
analysis of computer-communication and production systems.He is honorary professor in
Heriot-Watt University, Edinburgh (2008-2013) and in June 2009 he received an honorary
doctorate from the University of Haifa.
Corné W.G.M Dirne is Lecturer Industrial Engineering and Curriculum Co-ordinator Industrial
Engineering of the Avans University of Applied Sciences.He holds an MSc in Industrial
Engineering and a PhD in Operations Management & Logistics from Eindhoven University of
Technology. After his PhD he had the following positions: Assistant Professor Logistics
(Eindhoven University of Technology); Co-ordinator Part-time Programme Industrial
Engineering (Eindhoven University of Technology); Director of Education Industrial
Engineering (Eindhoven University of Technology).
Andreea Dragut is associate professor of discrete mathematics at the University Aix-Marseille
II. She received her PhD in 1999 from the Institute of Mathematical Statistics and Applied
Mathematics of the Romanian Academy of Sciences and a second one in 2003 from the school
of Industrial Engineering of the Eindhoven University of Technology. She held positions as
post-doctoral researcher at the EUT, assistant professor at the University Aix-Marseille and
visiting researcher at the State University of New York. Her research interests are data stream
analysis, geometrical structures in mathematical programming and multimodularity and
submodularity applications for Markov decision processes.
234
Jan C. Fransoo is a professor of Operations Management and Logistics in the School of
Industrial Engineering at Eindhoven University of Technology in the Netherlands. He holds an
MSc in Industrial Engineering and a PhD in Operations Management & Logistics from
Eindhoven University of Technology. His dissertation “Production Control in Flow Process
Industries” was completed in 1993 under supervision of Professor Will Bertrand and Professor
Jacob Wijngaard. He currently serves as Program Director of the Master Program in Operations
Management & Logistics at TU/e, as Research Director of the European Supply Chain Forum,
and as vice-president of Dinalog (Dutch Institute for Advanced Logistics). Professor Fransoo
held various visiting appointments at US universities, including Clemson University, the
University of California at Los Angeles, Stanford University, and MIT. He currently serves as
Senior Editor of Production and Operations Management and is editorial board member of five
other journals. He has published over 60 papers in academic journals and presented at many
academic and industry conferences. As part from his academic activities, Professor Fransoo has
conducted dozens of projects with industry, many with the more than 50 Master students that he
has supervised.
Twan Geenen studeerde Technische Bedrijfskunde aan de Technische Universiteit Eindhoven
van september 1967 tot juni 1972. Van juni 1972 tot februari 1974 was hij werkzaam als
organisatieadviseur en van 1974 tot 1989 als directeur van verschillende verpleeg- en
ziekenhuizen.Vanaf 1989 tot eind 2009 was hij werkzaam op de Technische Universiteit
Eindhoven in het vakgebied logistiek. Hij verzorgde inleidende colleges in de logistiek en
keuzecolleges op het gebied van materials handling. Daarnaast begeleidde hij vele studenten bij
groepsprojecten en afstudeerprojecten. Van oktober 1995 tot november 2007 was hij tevens
parttime programmamanager van de ontwerpersopleiding International Program in Logistics
Management Systems.
Paul Gosselink studeerde technische bedrijfskunde aan de Technische Universiteit Eindhoven.
Vervolgens was hij werkzaam als adviseur bedrijfsinterne logistiek bij IPL-TNO-TUE, adviseur
demand chain management en partner bij IPL consultants BV en programmamanager
procesindustrie bij NV BOM. Op dit moment is hij programmamanager nieuwe energie &
groene grondstoffen bij NV BOM.
Geert-Jan van Houtum is Professor of Maintenance, Reliability, and Quality at Eindhoven
University of Technology since 2008. Prior to that he filled positions as assistant/associate
professor at the University of Twente (1994-1998) and Eindhoven University of Technology
(1999-2007). He obtained his MSc and Ph.D. degree in Applied Mathematics from Eindhoven
University of Technology in 1990 and 1995, respectively. He is the scientific director of the
235
Beta Research School for Operations Management and Logistics. His research is focused on: (i)
Spare parts management; (ii) Maintenance and availability management of capital goods; (iii)
The effect of design decisions on the total cost of ownership of capital goods. A significant part
of this research is in collaboration with companies such as ASML, DAF, IBM, Nedtrain, Océ
Technologies, Marel Stork, and Vanderlande Industries.
Joris Keizers studied Econometrics and worked under supervision of Will Bertrand on a Ph.D
about Maintenance planning. In 2000, he joined Arthur D. Little, where he has worked for five
years on various assignments in the field of both Operations Management and also Strategy. In
2005, Joris left consulting and moved to SPG Prints, a former division of the industrial company
Stork. Within this company, he has been active in various roles, ranging from Manager
Logistics, Manager Strategy Implementation, (Master) Black Belt to his current job as
Operations Manager.
Ton de Kok holds a PhD in Mathematics from the Free University of Amsterdam. He worked at
Philips Electronics in Eindhoven, The Netherlands from 1985-1992as consultant in Operations
Research. In 1991 he was appointed part-time professor in Industrial Mathematics at Technische
Universiteit Eindhoven. Since 1992 he is a full time professor Operations Management at the same
university. He is the Director of the European Supply Chain Forum (eSCF). His main research
areas are Supply Chain Management and Concurrent Engineering with emphasis on quantitative
analysis. Ton de Kok is head of the capacity group Operations Planning, Accounting, and Control
(OPAC) in the Department of Industrial Engineering and Innovation Sciences since April 1, 2008.
Marc Lambrecht is Professor of Operations Management and Director of the Research Center
for Operations Management, Faculty of Business and Economics, K.U.Leuven. His research
interests include stochastic modeling, inventory management and queuing models. He is holder
of the Atlas Copco Research Chair in Service Systems.
Jan Vissers is professor of health operations management at the institute of Health Policy and
Management of the Erasmus University Rotterdam., and senior management consultant at Kiwa
Prismant a research & development company in the area of health care management in Utrecht.
He received his MSc in Industrial Engineering & Management Science from EUT in 1975 and
his PhD from EUT in 1994. He is member and past chairman of the Euro Working Group on
Operational Research Applied to Health Services, and member of the editorial boards of Health
Care Management Science and the International Journal of Care Pathways. His research focuses
on the analysis, design and control of operational health care processes and systems.
236
Guus de Vries is partner at DamhuisElshoutVerschure management consultants in 's-
Hertogenbosch. From 1993-2001 he was professor of Management Science in Health Care at
EUT, and from 2004-2011professor of health operations management at the institute of Health
Policy and Management of the Erasmus University Rotterdam. He received his MSc in
Industrial Engineering & Management Science from EUT in 1979 and his PhD from EUT in
1984.
Jacob Wijngaard studeerde Wis- en Natuurkunde aan de Vrije Universiteit (doctoraal in 1968).
Werkte van 1968 – 1991 als wetenschappelijk medeweker en hoogleraar aan de TUE, in de OR
groep bij de Faculteit Bedrijfskunde. Specialisatie Voorraadbeheersing en Productiebesturing.
Parttime advieswerk voor Philips en voor Berenschot. Vanaf 1991 hoogleraar Productiemanage-
ment aan de Faculteit Bedrijfskunde van de RuG. Van 1996 – 2001 decaan van de faculteit. De
onderzoek belangstelling is verschoven naar (productie) besturingsconcepten in het algemeen en
naar de rol van mensen daarbij. Een belangrijk toepassingsgebied van de groep in Groningen is
de gezondheidszorg. Lid van het bestuur van de Vereniging voor Logistiek Management (vLm).
Henk Zijm is a full professor in Production and Operations Management at the University of
Twente, and Scientific Director of the Dutch Institute for Advanced Logistics DINALOG, as
well as (non-voting) chairman of its International Scientific Advisory Board. He spent several
periods at Eindhoven University of Technology, as a PhD, as a part-time professor, and as a full
professor in Operations Management. He has been with Philips Electronics for eight years as a
consultant in Operations and Logistics Management. At Twente University, he worked with
numerous companies and organisations both in the Netherlands and abroad. He served in
various administrative positions and has been Rector Magnificus (Vice Chancellor) of the
University of Twente from 2005 until 2009. He was president of the International Society for
Inventory Research (ISIR), Budapest, from 2008 until 2010. Current ancillary positions include
those of Chairman of the Board of the NWO institute ASTRON on Radio Astronomy, vice-
chairman of the Board of Commissioners of OOST NV (the development agency for the East-
Netherlands provinces) and member of the Supervisory Board of the Roessingh Center for
rehabilitation care and research.