ESPANDA: For solving problems by applying the principle of similarity

  • Published on

  • View

  • Download

Embed Size (px)


  • Pergamon

    Expert Systems With Applications, Vol. 8, No. 2, pp. 249-254, 1995 Copyright 1994 Elsevier Science Ltd Printed in the USA. All rights reserved

    0957-4174/95 $9.50 + .00


    ESPANDA: For Solving Problems by Applying the Principle of Similarity


    INPRO, Berlin, Germany

    Abstract--ESPANDA is a software system that was developed for consultation on problems in con- nection with machining operations. The prototype of the system was implemented by INPRO ~ in collaboration with Mercedes-Benz AG and Volkswagen AG. The approach taken in this system is based on the assumption that problem classes, or so-called similarity classes, can be defined for machining problems. The solution to a given problem is derived inductively with the aid of a library of case data. This approach will be explained in greater detail within the context of the problem area in question.


    MACHINING OPERATIONS, for example, turning, dril- ling, and cutting, occupy a key position in production (Spur, 1979, p. 7). Due to steady advances in the de- velopment and improvement of materials, tools, and machines, moreover, the demands made on those re- sponsible for planning the manufacturing equipment and processes are continually growing. High tool costs and expensive machines are making it vital to identify the optimum machining conditions in each case. By switching to new cutting materials--for example, from high-speed steel to carbide--and taking advantage of the faster rotational speeds of new machines, the av- erage output values can be increased and production times thus shortened.

    Other restraints on the possibilities for shaping the production process must also be taken into account. For instance, other areas of corporate activity like cost calculation, ordering, purchasing, and controlling con- tinually impose new restrictions that can have tech- nological consequences. Because of environmental regulations, dry machining is gaining in importance over wet machining. In these areas, new experience must be continuously gained and made widely avail- able.

    Requests for reprints should be sent to Angelika Garben, INPRO, Nuernbergerstr. 68/69, 10787 Berlin, Germany.

    INPRO--Innovationsgesellschaft f'tir fortgeschrittene Produk- tionssysteme in der Fahrzeugindustrie mbH in Berlin is a jointly owned subsidiary of Mercedes-Benz AG, Krupp Stahl AG, SIEMENS AG, Voest-Alpine Steinel Ges.m.b.H., Volkswagen AG, HOECHST AG, and the Berlin Senate.


    This article is not intended to take account of all of the problems of this highly complex area within the framework of the ESPANDA project. Instead, a rele- vant and straight-forward area in connection with technological aspects was defined (Figure 1) with the aim of obtaining useful and informative system output.

    The already existing aids within this technological context have revealed themselves to be of little practi- cal use.

    The quantitative links between technological pa- rameters and machining results cannot be adequately captured in their entire complexity by analytical, mathematical models. Particularly because of the dy- namic nature of technological progress in the field of machining, such models cannot be modified quickly enough to take new developments into account. Be- cause they have been intentionally designed to provide generalized solutions, moreover, they are less suited for dealing with specific problems.

    The values recommended by tool manufacturers can only serve as general guidelines. They are virtually never applied in practice, because they do not (and cannot) give consideration to the special conditions existing in a given manufacturing operation: for ex- ample, the varying quality of cutting and other tools made in-house and by different manufacturers, of cy- cling times and tool-changing intervals, and so forth.

    Similarly, the existing data bases open to access by the public (e.g., INFOS in Germany, METCUT-MDC in the United States, and TRI in Japan), which were set up to support planners, also tend to contain only generally valid values, being unable to take account of plant-specific parameters. The data stored there are from machining laboratories or industrial production.

  • 250 A. Garben et al.

    Standards Purchasing Costs

    ManufaTam Further processing

    ..... / ./.,, ,, '/,., .: ..................... .......... Teciinlc, al expertiN ...... ...... ~ i~ ~lr i I ~. :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ,,, ,, .,, ., :::::::::::::::::::::::::::::::::::: ................

    I / ' :,i::iiiiiiiii ii!iii.. .............

    Machining operations ............................ L"~/'~,I I'~/~~~::iii!i!i]iiiiii!!ii~iiiiiiiiii::i]i]iiiiiii::i::(iiiil # ~F ," #~#~ -!:!:.:E::: i: !:~ :i :i:ii

    The problem i~ii~!~i~iii~!~!i]i!i~iiii~i~i~i~i~iii~i~:~i~!i~!~i!!~~!!i~[~]~i~]~ii~!!!!!!!!~]!i~i~.~i~i~]~iiii~!~ii~ FIGURE 1. The problem context and problem.

    Users criticize that these data cannot be checked, and their reliability is often called into doubt. There is also little or no support in the form of on-line availability or utilities that would permit direct utilization of data base resources for solving specific production problems.

    Here, the knowledge that machining experts have accumulated through many years of experience--in any case with tried-and-proven technologies--mainly provides the basis for deriving good solutions to ma- chining problems. It generally takes 5 to 6 years for a beginner to become a machining expert. This means that even normal personnel fluctuations can signifi- candy reduce the availability of this knowledge, making it all the more desirable to conserve it.


    The objective of our project was to place a tool at the disposal of production planners that would enable them to solve the problem of how to optimally allocate equipment and fix machining values (Figure 2).

    From the large range of different machining oper- ations, drilling was initially selected to be concentrated on. This operation has been largely ignored in the past, and there is consequently a large information deficit here. The problem at hand is sufficiently complex to serve as a test for validating the new approach.

    To this end, it was planned to incorporate results that had already been obtained. These results consisted


    I Work dt Choice of pr


    I Machir

    Support from expert system

    Machining solution

    FIGURE 2. Embedded problem.

  • ESPANDA and the Principle of Similarity 251

    ~.::.:.:-: ~

    :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ~ i i i i~ i~ i~ i i~ i~!~!~i~i : i~! : i~ i~ i~ i~ :~:i:i~:~:i~:~:!~i:~:~:i:~:~:i:i:i:~:~:i:i:i:i:i:~:~:~:~:i:i:i:;:~:i:~:~i:i:

    ~//~i~i~!~"!~i~i~i~iii~!~;~i~i~i~i~ i ',~,i',i~,M~hin;~i~,~,~,:~/,',~,~,i~i',~,',',',',',~,!~,:~i~i!i~;! ~ii~,i~,i::%:i~::::~ili;~'~iii',ii~,',~,~,i!;,;,i','~iiii!~: ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ~::::~ .... Feed .... ,,,,~!::;::!::::::~::;iiii::ili; i::i~i~i"e~b"sfi~"ii~i~ii::;~i::i~i::i~i::i

    iiii;iiiiii Power O~'"iiiiiiii!iii!iiiiiiill iiiiiiii;::~i~iiiiiiiii;

    ii~iiii!i~i~iiii~i~i~..Efi~y/w~d~awi~l..~iii !J!i::::iii:::::::: C amp ng ~;~;~:~:~:~:~:~:~~~,~[iii""i!ii::!::i::ii!::!iiiii ~i!::i!i!~iii!i!iiiiiiii!::ii ii::~ii[~::~::~iiiii!ii~i!!ii!iii!i ii~iii!i!iiii!~;ili!i~i~ii!i ~:!~i!ill i i'ii i Toolholder ' iii;iiii i ~;! i~!; ;! i ! i i i~~"; i i i i i i i i i l i ; i i ; i l i l ; ; :;:~:i:i:i:;:;:;:!:;:i:;:;:i:~:~ To lerance :;:;:;:!:~:i:!:;:!:;:~:i:!:i:~:;:~:~:i:i :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ,:~ :,.~::~:.".. ;:i:i:ii::i:ii;:!:;:~:~:!:!:;:;:;::~:i:~:i:;:i:;:;:i:~:i:i:~:i i!iii!i~iii~ iii.......,.............-.-.......::::iiiii~i;iiiiiii!iii!i!! ili!!~ Dri ;~ii~ili!;!i!~!;!;!i!~?~iiiil;~iii~ [: i:i:i:i:i:t!i!;!;!iE!~iii~;iii~i;;~ii~iiiiiiii~i!i!i! i:i:~:~:i:i:i:~:!:~:!:i:!:~:~:!: ~u~ace ::::::::::::::::::::::::::::::::::::::::::::::: i : i : i : ........!:~:i:~:i:i:i:i:~:i:i:::~:~:i:i:~:~:~:~:~:i:i:!:! I ~"~::t:::i:!:i:i:~::: :~:i:i: ::::::::::::::::::::::::::::::::::::::::::::: :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: :::::::::::::::::::::::::::::::::: " tYPe ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: : .~; :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ::.:-:-: 7.:.:. d r ," pr- ip" d i l l " ======================================== :::::: : : : : :::: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : , , ============================================================ ::i:i:~::~:i:i ' j , , , ' , ~, , , ,, ==================================== .:.:.:.:.:.:.:.:.:. Matedal :':':':':!i!!!!!!i!i!i!iii~i~:/~!i~i "~!.!.!.~:!:i:i:i:~:~:~:!:i:~:i:i:i:~:~:~:~:i:i:i:~i~:~:i:i:~ii!;::;~

    [::,:';. ii!iiiiiii ,iiiiiii!!':ii!i!' !ii' iii' il iiii i iii i iiiiii , i ii iiii ii iiiii iiii i iiiiii'iiiiiiiiii!!iii'i!iiii'ii ii iiiii:i: :ii:iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii iiiii:::::: i ::::::iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii iiiiiiiiiiii!iiiiiiiii iiiii::i::i::iiii Malei-iali~i~::~::i::~::iiii ::::::::i::::::::i::i::i::iiiii::i:::::::::: i::i::i ................. ~iiii::i::i::iiiii::i~ ~ i i : : i i li::i::i i::::ili ::::i::iii:: Prelimin~"~i;a:'~6ris :::::::::::: ~i~:~i~:~:~i~i~ii~:~:~;i~;~/~;:~i;~.~.:e~ ' [//~'~l:~:~ ~::;::~::~i~::~:/:?:~::

    i:ii:: ii! ii ::!:!/ili:ii!iiii!iiiiii!iiiii!i!iiiii!iiiiiii!!iii!i!ii!!ii! i iiii!iiiii!iii i ~/~.!~i~/~!~!~i~;~!~!!!~!!~/~!~!!~!~!~!~!!~!~!~!!~!!~!~!!~!~!~!!~!~ ~, ~,~,~!i!~,!~,!~,!~,!~,! ~, .~o~ ~ !',!~!i i~i~iii!!i!~!~!~!~!!~!!i~!~!~i~i~!!!!~i~!~i~i!!~!~i~!~i~i~i~!~i~i!~

    !iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii!iii!i!!!!!i ! i iiiiiiiiiii iiii!iiiiiiii iiiiiiiiiiiiiiiiii iiiiiiiiiiii iiiii iiiiiiii iiiiiiiii!iiii! ii

    . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . .=. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

    iii~iii!!!i!!!i!i!i~!!!!i!ii!i~i!iii!i~i!iiiiiii!i~i~i~i~i~ii!!i~i~i~i~i~iiiii~iii~i~i~i~!~i!!iii!~i !iiii~i!~i:#~i~i~i~ii~iiii~ii~#~!~!~!!~!!i~ii!i!iiiiii!i!ii!~!~!~i!iii~i~!iiiiii ~

    iiiiiiii!iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii!iiiiiiiiiiiiiiiiiiiiiiiiiiiiii iiiiiiii!i!iiii!ii!iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii

    FIGURE 3. Composition of the machining record.

    of information on machining cases that had been cap- tured but not yet written down or entered into in a data base. Thus, one of the principal subtasks involved in the project was to compile a case library (Figure 3). The primary goal of the project was to identify a pro- cedure that would permit the machining records col- lected in the case library to be utilized as directly as possible for dealing with new production tasks.

    In addition to its main function as a consultant, moreover, the system--if used and maintained regu- lar ly-also contributes to solving the problems asso- ciated with the fact that machining knowledge is nor- mally linked to specific individuals and therefore is only temporarily available to a company.


    The approach selected for ESPANDA involves search- ing a case library to find solutions to similar problems in the past. This approach departs from the assumption that problems can be collected together into classes, and that it is technologically feasible to apply the same solution to problems of the same class. The problems collected together in such a class are regarded as being similar to one another (thus forming a "similarity class"). Machining cases involving similar problems thus contain potential solutions (Figure 4).

    The first prerequisite (Hart, 1986, p. 110) that must be met in order for this approach to work is the exis- tence of a collection of machining records (case library). As used here, the term "machining record" refers to a description of a machining case, consisting of three parts: Description of the problem. The solution found for dealing with the problem; in

    the case of drilling, that means descriptions of the characteristic attributes of the machine and the tool

    used for a given drilling operation, as well as of the settings employed.

    The results of the machining operation, for example, the achieved machining quality, the service life of the tool used, and so forth, all of which can be used to assess the limitations of the solution.

    The second prerequisite that must be met is classifi- cation of the possible problems. This classification constitutes the link between a given problem and the case library. Since the comparisons to establish simi- larity make use of the problem description, it follows that the problem descriptions contained in the ma- chining records must at least contain all attributes that are relevant for purposes of comparison.

    For the field of application addressed by ESPANDA, the experts confirmed that the parameters chosen for purposes of comparison can be regarded as being in- dependent of one another. Consequently, it was pos- sible to define the problem classes by classifying the value ranges for their individual attributes. By means

    . ............................................................. ~.i~';iii~!~iiii

    Solution to the current problem

    iiii::i~#:~{~::i::i=:ii:::;i;i::i::i::i::i::i~.~#/:~:: .........

    I Case X: solution

    FIGURE 4. Schematic representation of the approach taken.

  • 252 A. Garben et al.

    of this intentional description of classes by applying formative laws, we were able to avoid explicit repre- sentation of these classes. This was very important for the success of the project, since otherwise it would have been necessary to describe more than 30,000 classes, even in our already-narrowed field of application.

    In this approach based on the principle of similarity (Figure 5), the problem class corresponding to the problem at hand is first identified. Because the problem classes were intentionally described, that means that an explicit representation of the class to which the problem at hand belongs must be generated. This is achieved by determining the characterization class corresponding to the characterization of each attribute of the problem. In the case of metrically defined attri- butes, for example, these characterization classes can simply contain intervals. For other attributes, same- ness--a special instance of similarity--is required. Here too, it is possible to speak of characterization classes, if the reference value is treated as a class con- taining just one element.

    After generating the problem class, this is used as an argument f...


View more >