AbstractA major faci

is being establisStandards (NBSprovide extremeemulating a wide varilsty Bf rntyplcal of a small machine job shop. The controlarchitecture adopted is hierarchical in natureand highly modular. The facility will be used forresearch on interface standards and metrologyin an automated environment.

Keywords: Automated Machining, Hier-archical Control, Manufacturing Research,Research Facility.

The Congressional Acred n g up the NationalBureau of Standards charges the Bureau with;

1.

2.

The custody, main&nce, and dcveiopmcntof the national standards of measurement, andthe provision of means and methods for makingmeasurements consistent with those standards.Cooperation with other government agenciesand with private organizations in the establish-ment of standard practices, incorporated incodes and sp&ifications.To perform these functions, the Bureau has,

over the yean, installcd numerous experimentalfacilities, including a nuclear march reactor, a

ponitsrneaswmmt andstandardsin the decades of the 1980s and

1990s. When completed in 1986, this facility will becapable of full-scale emulation of the flexiblemachining ctils in the automated faczory o f thefuture.

Purpose of the AMRFThe Automated ,Manufacturing Research Faci-

lity will reside in the Center for ManufacturingEngineering which was founded to supply to themechanical manufacturing sector the servicts des-cribed in the enabling legislation and to carry on aresearch program to develop 'means and methods"for making the measurements that willbe needed bythis -or in the future. The Center currentlyprovides a wide range of calibration services formechanical artrfact standards such as gage blocks,thread gages, and line scales as shown in Figure 1.

Thcse artifact standards, many of which weredeveloped by NBS in the first thm decades of thiscentury, arc idealized models of the products to

*Act of 22 July 19SO. W Sut. 371 (Pubiic Lav619.31 Congms)-An Act To amend seaion2 of the Act of March 3.1901 (31 Scat.IMP). IO provide borrc authority for the performance of certain funcuoru and activities of the Department of Commerc:. and for otherPUrp-.

17

which they are compared. The comparisons (cali-brations) arc organized according to statisticalquality control methods developed during andimmediateiy after Worid War11. Anifacts currentlya n the basis for the National MeYurrmenr Systemwhich provides nation-wide dimensional cornpati -bility by a chain of comparisons back to NationalStandards. The system has remained virruallyunchanged since the 194Os, except for the intro-ducion in the 1960s and 70s of the concepts ofMeasurement Assurance Programs (MAP;', whichemphasize the system aspects of measurement andintroduced the concepts of closed loop feedbackinto metrology management.

Manufacturing technology, however, bas notremained unchanged. The introduction of numcri-a l l y controlled machines, group technology con-cepts, and the first steps toward nexible Manufac -turing Systems (FMS) in the 1960s called attention

to the labor intensivenature,high s k i l l requirement.and time consumption of classic metrology.

The first effon of NBS to meet the upsomingchallenge was in 1968 when aresearch program wasmounted to investigate the possiblity of automatmgsurface plate metrology by the use of the then newcomputer controlled coordinate measuring machines(CMM). A decade o f work realized a measurementsystem based on such machines where the "product -like" artifact standards of the past were replaced,with measurement protocols based on laser inter-ferometer techniques for characterizing the mea-suringsystem (coordinate measuring machine) itself.Transfer standards were developed that permittedsuch machine or process characterization to beeconomically realized on machines of lesser butknown precision'. The three-dimensional ball piateon the table of the CMM in Figure 2 is one of thelatest. of such standards. These new measurement

18

methods are rapidly becoming the norm for certainpan familia; That familiesarc medium to large insize and compiex -prismatic in nature, and hencesimilar to the output of the first and secondgeneration FMS,

Even before this work was completed, it becameobvious that the= were many pan families that wereill-suited to measurement by CMM. Small pans,turned parts, and very simple partsarc all either verydifficult or uneconomic to:measure in this manner.Moreover, the rapid development of FMS, with theability to reduce inventory by shoner runs, castsdoubt on the continuing usefulness of any QCsystem which depends on statistical sampling. Alongwith others, NBS became convinced that the QCsystem of the future would increasingly depend oncharacterization of the process. monitoring of themachine parameters, and adaptive control ratherthan measurement of part parameters after theprocess, or a step in the process. was complete. Sucha development wil l require NBS to provide the-means and methods" o f measufement where themeasurements are deeply embedded in the process.

The totalBureau experience has amply demon-s t r a t c d that one Cannot learn to measure without'handson"experience, and every attempt to attackmeasurement problem on apurely theoretical basishas proved less than satisfaczory. Therefore, incooperation with the Bureau of Engraving andPrinting, an NC machining center was set up in theNBS Instnrrneat Shop to explore the measurementproblems involved in assuring pan dimensionalllceutacics by machine calibration. It was soonshown that the calibration techniques and softwarec o d o n algorithm for static e n o n developed oncoordinate measuring machines could be applied tomachine tools in a shop environment. A five-foldincrease in accuracy was demonstrated. Figure 3illuvates the de- of comction obtained for a

The comaion of dynamic er ron such asthermal distortion due to internally generated heator distonion due to cutting forces needs furthermearch, but appean to present no insurmountableobstacles. Certain complex dimensional measure-ments such as drill condition and tool setting areneeded, but modern microcomputer based tech-nology appears adequate to the task.

The AMRFwillallow research inmdsurernenttechnology to be expanded to include those systemelements at the ctlI(multiwork station) level. TheAMRF will provide a test bed where integratedmanufacturing system measurement research a n be

. The AMRF will provide a test bed for researchdirected toward the'establishing of standard prac -tices". If flexible manufacturing systems are tobecome widely atioptcd in the discrete pans Industrywhen 87% of the fi'ims employ less than 50 persons,they must become much more modular then theyare today. I t must become possible for a finn to stanwith an NC machine, add a robot, add anothermachine, and so on as capital is accumulated and asthe f i r m ' s business grows. Systems must also becapable of being tailored to various pan mixeswithout extensive engineering effort. However.before this degree of flexibility can beaccomplished.interface standards must be adopted so equipmentof diverse origin can be integrated incrementAlyinto the systems.

n e first steps in this direction have alreadybeen taken. Under Air Force Integrated ComputerAided Manufacturing (ICAM) sponsorship, theNBS coordinated the efforts of a consonium of 45

ldbgcCnkf.

. pdomled.

19

0 C-

f

pnvate firms 10 generace the Initial GraphicExchange Specification’ (IGES). IGES is a commonpublic domain data format which allows geometricdata to be exchanged between two different types ofcomputer aided design systems, or between acomputer aided design and a computer aided manu-facturing system. IGES thus allows access to thegeometric data bank of a computer aided designsystem without thenecessity of producing a drawing.IGES has recently b n n incorporated into a nationalstandard (ANSI Y14.26M). The development ofthis standard is important for its intrinsic value; butperhaps more important, it has demonstrated thatsuch Interface standards can be structured andgenerated ina manner which provides full protectionfor proprietary interests. The A.MRF will provide a

20

test bed for the development of similar intedacestandards for integrated manufacturing systems. Itwil l allow the test and verification of interface .standards in an open and nonproprietary atmo-sphere.

Description of the AMRFThe Automated ,Manufacturing Research Faci -

lity wi l l supeficially resemble a FMS designed tohandle the bulk of the part mix now manufacturedin the N5S lnstntment Shop, T’his part mix has betnstudied using Group Technology‘ concepts and IS

shown to be similar to a typical job shop. The pans

Jounul of Manufaccunng SystemsVolume I. \umber I

manufactured wii l fall within the following limita-tions:

1. Weight: Less than 50 kilogram (100 Ibs).2. Size, Prismatic: 300 mm cubes (12” x 12“ x

3. Size, Rotational: 250 mm diameter x 250 rnm

4. Parts Run: 1 to 1,OOO pieces.5. Complexity: Up to 4 axes prismatic.6. Mate& Steel, stainless s e i , aluminum,

brass, iron, lucitc.The AMRF and the mearch performed on it

wii l address only tne manufacture of individualpans by chop forming metal removal. Hence, theunit operations wil l include only: fixturing, milling.drilling, reaming, tapping. boring, turning, facing,threading, cicaning, debumng, and inspection. Suchproblems as automated assembly, welding, harden-ing, and finishing wi l l not beaddressed.

The AMRF hardware is structured around theconcept of singie selfxontained work stations, eachwith a well defined set of functions which can beuseful as a stand-aloneentity. The current plancallsfor the existence of eight such stations with varyingdegrees of complexity of function. They arc:

127.

length (IO”x IO”). .

I. Horizontal Machining Station.2. Venical Machining Station,3. Turning Station.4. Cleaning and Debumng Station.5. Inspection Station.6. Materials Inventory Statlon.7. Transfer System (station).8. Housekeeping System (station).

Items ’land 8, theTransferand the Housekeep -ing Systems arc not strictly stations since they arenonlocalized in the facility. From the point of viewof the control system, however. they wi l l be treatedas stations.

T h e Materials Inventory Station wiil be usedas a buffer to allow storage o f sufficient material forseveral days o f operation and an automatic inven-tory for much of the raw material requirements of aJob shop. Ifsuch a system were to serve sunply as abuffer, ony three or four days storage would berequired. that is. enough for automatic operationthrough a long weekend. Since It is not the purposeof this program to study such systems per se, oneweek of storage was chosen as a reasonable trade offbetween the requirements of the simple buffer (orInterface to the manual world) and a much more

elaborate total system inventory. At this stage weplan to use the inventory system for thcstorage ofraw material blanks, tools and tool holders(assembled), special fiitures. and finished parts orparu in-process. The inventory system will beloaded and unloaded manually while the facility isin operation.

The Materials Transport System will providethe means of moving parts, tooting, and fixtureswithin the facility. Two mechanisms will be used.One, a carousel, will also serve as the inventorysystem. The second, a robot can or automatedguided vehicle (AGV), will allow great flexibility inlayout and easy access to the machines’. Although,the transfer system itself is not seen as a primaryresearch area for NBS, the interfaces between thework stations and the transfer system wil l bedesigned to accoaunodate many different types ofsystcms as well as other options inorder to maintainmodularity.

Themachine tools were chosen to be represen-tative of the types of general purpose machine toolsincommon use throughout the U. S. The chotce alsomatches the speclftc needs of the NBS InstrumentShops as revealed by the Group Technology Study.Each of the machines will be configured into a workstation with a single industrial robot.

NBS’has chosen to use standard, modem.general purpose machine tools in the construction

.of the AMRF. This is a different strategy than thattaken by two other wciI known national programs inautomated batch manufacturing, the Bntlsh AS.P.plan’, and the Japanese MUM or FMC plan-. Botho f these other programs have assumed apriorr thatcurnnt machine tool designs are inadequate for anautomated research facility. However, based on aninternational study of the state-of-the-artinmachrnetool science’, NBS has decided that this assumptionis highly questionable. We have chosen to rely onthe engineering experience of a well -developedindustry rather than a radical new design. Shouldproblems arise in r ‘ ~ bdity, reparabllity. and chipremoval, we plan to subcontract any needed modi-fications to the same industry.

Cleaning and debumng was made Into aseparate function (and station) because of theimportance of this task for automatic Inspection. Asmany .debumng operations as possible will becarried out at the machining site. Neverthe!ess. thereappears to be no way to avotd cleaning and

21

Jounul of Manufacturing SystcrmVolumr I. Numkr 1

deburring as a separate operation in all cases.Studies haverevealedthat the cost for cieaning anddeburring in batch manufacturing is high and often

The Inspection Station wi l l be a modifmi fouraxes horizontal ann measuring machine tendedby arobot. It willbe very s@ilar to the machining workstations from the control pointsf -view. Thisconfiguration was chosen primarily for flexibility in

The Howkaping System will provide for theremoval o f chips during automated operation.Cleanliness during manufacturing and fituring,and the effects of cutting fluid and chips (dust) onsensors have been serious problems in many of theexisting FMS In the AMRF, chipmnovaiis expected to be complicated by the variety ofmaterial, the largenumber of sensors contemplated,and the decision to address the flexible fixwingproblem robotically at the rnrchincs. A pianregarding chip removal is being dweloped at thistime through both externai "

p

'2 andi 1studies.

~nfc~~gnizcd~.

w.

This system will be kept as simple as possible wlthlittle attempt to optimize for long unattended runs.

Layout for the facility is shown in Figure 4.This configuration allows easy access to themacbines, and the transfer mechanism can be eitherthe automatically guided vehicle system or thea n d carousel system. Coolant and cutting fluidarcrecycledat the machine. Buffering is provided tothe machines through the row of "file-cabinets "which make up the carousel shown in the center o fthe model.

The three robots in the lower right are part ofthe cleaning and deburring station. The separateroom at theupper right is the inspection station. Thefour machining stations each consist of an industrialrobot, amachine tool, a localized inventory of tools,fixtures, grippers (end effectors), probes, and inter-faces to the transfer and housekeeping systems.

The proposed operational scenario placessevere requirements upon the work station and its.subelenrenu. Some of these arc necessitated by thedccision not to palletize and others by the wide part

--k. .0

. E . . . . . . . . . . _. . . . . . _. _ . . . . . . . . . . . . . . . . . . .

Fig-4Modd of AMRF Skohng Loatioa in Insmtrnmt ShoQ.

The cmtnll. louted u rowr l rill be used to convq rnrtni.l.tooh and finished parts. For purposas of mvironmcnrd control

the meuunng station is, as shown. in a wpance room.

22

mix envisioned. The tools and material amving at awork station wil l not be precisely located ia space.This will require advances in the state-of-the-artover current industrial robot capabilities. The robotcapabilities wil l abo be stretched by the requiremento f futuring on the machine. The problems of chipbuildup and tool wear will be aggravated by thematerial and part mix contemplated. It is our beliefthat those requirements willbe the nom in secondgeneration FMS s y s t e m which willbe available inthe 1990s.

Table 1 gives a.paniai list of the functionsq u i d of the robots and machine tools in theAMRF. As can been seen. it is intended that theindustrial robot be able to locate pans, toob andfixtures, transfer these items to the machine tool,fixture the pan. and monitor the process whilemachining is carried out. Thus the robot wil l haveextended sensorycapabiliticsi the ability topreciselygrip and position variable shapes, and considerablemanipulative ability to fixture the pam upon themachine tool. In general, solutions to thae problemsare more difficult for prismatic than for cylindricalworkp ick .

To the best of our knowledge, no one hasaddressed the flexible fmturing problem. in anydepth though some very elaborate and expensive,solutions have been proposed by T~ffmsamrner'~.Tool setting on machining centers is in*a similariyundeveloped stage, as is generic tool wear/ breakagesensing". Our projcct plan has been to delineate ascarefully as possible those areas requiring development, research carefully the state-of-the-art in theseareas, and if required, initiate research dircctedtowards the appropriate goal(s).

At present the: Center. for - ManufacturingEngineering has two projects. one in robotics, theother in precision machining. These projects aredirected towards the development of the majorsubsystems required for the AMRF. The integrationof these two programs wil l take place first in theHorizontal Machining Work Station which will bethe tint work station to be assembled. T h e archi-tecture and control system hardware for this workstation will. s e r e as a model for the other fourgenerically similar work stations.

Although i t is recognized that there areimponant problems of CAD. CA,M integration tobe solved, the current plans do not include work inthis area. Production and process planning systems

Tab&IFunctions of the Work Station Subelements

Robots Arm Functions~

I. Part loading and unloading.2. Tool loading and unloading.3. Rough (k 50 mil) pan fixturing or fixture

4. Chip removal and control.5. Come visual inspection of faturn and pant.6. Initialpart and tool location.7. End effector sefmion.8. Deburring and deaning (oniy as needed for next

assembly.

opcration).9. safety.

10. Sdf-monitoring.

Machine T o d Functions

1.2.3.4.5.

6.7.

Machining.Part locations.Tool weart breakage sensing.

Proctss mcnitoring (cutting).a. Dynamics.b. Thermal.c. Hydraulics (ac.)Setf-monitoring..Deburring and cfeaning (as pan of the machiningoperation).

8. Adaptive control..needed to load the facility will be the larsely manualprocesses currently used for the NC station of theinstrument shop.

~ ~~ ~~~

In order for the A.MRF to serve as a researchfacility overthe next decade. it must exhibit a higherorder of flexibility than any cuncntly availableFMS. I t must not only be capable of very wide p a nmix, but must also be capable of easy reconfigura -tion to emulate work stations or small ce!ls operatingin the environment of a much larger and perhapsunmanned system. To accomplish .these goalsrequires a control system architecture of consider -ab12 sophistication. The conventional DirectNumerical Control (DYC) top down architecturewas judged to be unsuitable. primarily because o ithe inability o f such a system to react !o feedbackfrom sensors in real-time. In order for an entiremachine shop to completeiy operate automatically,ai l the machines must be equipped with sensors to

Jounul of Manufaauring SystemsVolumr 1. Numbcr 1

monitor their perfonnanct and compensate forirregularities and uncertainties in -the workenvironment. The sensor data must be processedand analyzed, and the rudts inuoduccd into themachine control system in real-time so that theresponse of each machine is goddirmed, relioble,and efficient.

A high degree of kory-intcnretivsbehavioron thepart of individualmachinuaeatuenormoussystem control problems for an entire shop. Theproblem of automatically controlling a number offeedback driven machine tools is much bigger thansimply the s u m of the contrcl problems for theindividual machines. The interactions among manysensory -interactive machines creates a systemcontrol problem in which complexity grows expolnentidy with the number of individual machinesand sensor systems. Once there are more than a fewnucfrines, each reacting to m o r data inreaI-time,the overall system control problem can bccomccornpleteiy unmanageable. This is the point atwhich most of the early attempts at building theagtomatic machine shop failed. The controi softwarefor such a system can bccorde enormously oomplexto write and virtually impossible to debug. Theclassical solution to control problems of thiscomplexity is to partition theproblem into modulesand introduce surne typeof hierarchical commandand control struaure. The advantage of hierarchicalcontrol is that it allows the.contro1 problem to bepartitioned so as to limit the complexity of anymodule in the hierarchy to manageabie'limiu,regardless of the complexity of the entire StruCture.

The use of hierarchical control for industrialapplications is not new. It has been employed incontrolling complex industriat plants such as steelmills, oii refineries, and giass works for years.However, such hierarchies are usually limited to twoor t h m levels and; most often represent falrlystralghtforward servo control applications. Theunique features of the control system being plannedfor the AMRF are the number of hierarchical levels(perhaps as many as seven or eight), and the amountof real-time computation and sensory -interaction at

' each level. Each hierarchical levef wil l perform asignificant amount of real-time computation andwil l Interact dynamically with the shop environmentin many different ways. The plan is to build a real-time sensory -interactive control system which at thelower leveis wil l respond to events o f millisecondduration (tight servo loops), and at the upper levels

will react to events of days or weeks duration(production planning and scheduiing problems).The levels in between these extremes will produceintelligent automatic rcsponsu to many differenttypes of shop floor conditions and situations.

contra1suucturc

On the left of Figwe 5 is an organizationalhierarchy wherein computing modules a narrangedin layers. The basic structure of the organizationalhierarchy is a tree. The flow of command andcontrol is vertical. Each node in the tree represents a

raxives input commandsd e (predecessorto one or more

rnode). There may8 sensory icputs and

cing data that flowhorizontally and/ or rise from lower levels in a cross-coupled netwark o f communication channels, buttheprimary command and control pathways form as t r i c t hierarchical tree.

At the top of the hierarchy is a single high-levelcomputer module. Here at the highest level, most&ob4 goals are decided upon and long-rangestrategy is formuiated. Feedback to this level is'integrated over an extensive time period and isevaluated against long-range objectives. Here long-range plans arc formulated to achieve the highestpriority objectives. Decisions made at this highestlevel commit the entire hierarchical structure to aunified and coordinated course of action whichwould result in the selected goal or goals beingachlevcd. At each of the lower levels, computingmodules decompose their input commands in thecontext of feedback information generated fromother modules at the same or lower levels, or fromthe external environment. Sequences of subcom-mands are then issued to e t s of subordinates at thenext lower level. Th is decomposition process isrepeated at each succtssiveiy lower hierarchicallevei, until at the bottom of the hierarchy there isgenerated a set of coordinated sequence of primitiveactions which drive individual actuators such asmotors of hydraulic pistons in generating motionsand forces in mechanical members.

Each cham-ofcommand in the organizational

Journal of Manufacrurtng SystemsVolume 1. \i;moer I

ORGANIZATIONAL COMPUTATlONALHIERARCHY HIERARCHY

BEHAVIORALHIERARCHY

.hierarchy consists of a computational hierarchy ofthe form shown in the center of Figure 5. Thiscomputational hiciarchy contains thm parallelhierarchies: (1) a task decomposition hierarchywhich decomposes high-level tasks into low levelactions, (2) a sensory .processing.hierarchy whichp ~ c c s s e ssensory data and extracts the informationneeded by the task decomposition modules at:achlevel and (3) a world model hierarchy which gene-rates expectations of what sensor data should beexpected at each level based on what subtask iscurrently being executed a l that level.

Each level of the task decomposition hierarchyconsists o f a processing unit which contains a set o fprocedures, functions. or rules for decomposinghigher level input commands into a string of lowerlevel output commands in the context of feedbackinformation from the sensory processing hierarchy.A t e v e y time increment each H module in the taskdecomposition hierarchy samples i ts inputs (com-mand inputs from the next higher level and feedback

'from the sensory processing module at :he samelevel) and computes an appropriate output. Adetailed description of such a system as applied torobots has been published

The sophisticated real-time use of sensor datafor coping with uncenainty and recovering fromerrors requires that sensory information be able tointeract with the control system at many differentlevels with many different constrLints on speed andtiming. Thris in general. sensov information at thehigher levels is mort abstract and requires theintegration of data over longer time intervals.However, behavioral decisions at the higher levelsneed tu be made less frequently. and therefore thegreater amount of sensory processing required canbe tolerated.

Attempting to deal with this full range ofsensory feedback in all of its possible combinationsat a single level leads to extremely complex andinefficient prosrams. The processing o f sensor data.panicularly vision data. is inherentiy a hierarchical

25

Journal of Manufacturing SystemsVolume 1. Number I

process. Only if the control system is also panitionedinto a hierarchy can the various levels of feedbackinformation be introduced into the appropriatecontrol levels in a simple and straightforwardmanner.

The world model hiexarchy contains priorhowlcdge about the uqk, thepuu,and the workenvironment. Typically, the type of feedmation required by the task decomposition modulesat each lwei depends upon what task is k ingperformed. As conditions change, different scasom,different * resolutions, and different processingalgorithm may be needed. Given the mtt of thetask exccution at each level, the world model canpredict what kind of sensory processing algorithmsshould k applied to the incoming data. Funher-more, sensor data can often be predicted from theactions king exccutd by the control system.

The worid modei generates expectations aa towhat the sensor data should look like. Thesepredictions may be based on prwious experiencewhen asimilar task was performed on asimilar pan,or may be generated from a Computer AidedDesign (CAD) data base which contains a geo-metrical representation of the part. The worldmodel hierarchy may contain information as to theshape, dimensions, and surface features of parts andtools and may even indicate their expected positionand orientation in the work environment. Thisinformation assists the sensory processing modulesin seiecting processing algorithms appropriate tothe expected incoming sensor data. and in comiat -ing observations against expectations. The sensoryprocessing system can thereby d n m the absence ofexpected events and measure deviations betweenwhat is observed and what is expccted.

Feedback can be used by the task dccomposi -tlon hierarchy erther to modify action so as to bringsensory observanons into correspondenct withworld model expectations. or to change the input tothe world model so as to pull the expectations intocomspondcnce with observations. In either case,m c e a match is achieved between the two, the taskdecomposition hierarchy can act on informationcontained in the model which cannot be obtainedfrom diren observation. For exampie. a robotcontrol system may use model data to reach behindan object and grasp another object which is hiddenfrom view.

If the symbolic commands generated at each

level of the task decomposition hierarchy arerepresented as points in the rnultidimenstonal"stateapace "consisting of the coordinates of al l thedegrees of fmdom of the machine or robot, andthesepoints areplotted against time, the behavioralhierarchy shown on theright of Figure5 results. Thelowest leveltrajcctories of the behavioral hierarchycampond to obscmble output behavior. All theother trajectories constitute the structure of condi-tiom deep within the control programs.

At each level in the behavioral hierarchy, thestring of commands makes up a program. Thisarchitmure implies that thm is a programminglanguage unique to each level of a hierarchialcontrol system, and that the procedures executed bythe computing modules at each level are written in alanguage unique to that level. This partitioning ofthecontrol problem into hierarchial levels limits thecomplexity of the progrnrnming language and theprograms at each level.It also generates a wholehiefarchy of languages for programming the robots.machine toob, and inspmion systems. and forperforming, planning and scheduling operations. I tis to be noted that sucha hierarchy lends itself to theutilization of IGES-Cype interface standards at eachlevel.

If the controi problem is further panitionedaiongthe time axis, an additonal degree of simplicity,can be achieved. If time is panitioned into a finitenumber of computational periods, each computa -tional module can be represented as a finite-statemachine. At every time interval, each computationalmodule samplesitsinputs (command and feedback)and computes an output. The programs resident ineach of the computational modules then becomesimple functions which can be represented byformulae of the form P=H(S), or by a set orproduction rules of the form IF<S>, THEY<P>.The control structure becomes a simple search. of astate transition table.

Each entry in the state-table represents anIF/THEN rule, sometimes called a production.This construction makes it possible to define *

behavior of high complexity. An ideal task per-formance can be defined in terms of the sequence o fstates and state transition conditions that take placeduring the ideal performance. Deviations from theideal can be incorporated by simply adding thedeviant conditions to the Itft-hand slde of the state-table and the appropriate action to be taken to the

26

Journal of Manufacturtng SystemsVolume I. \umber I

J

right-hand side. Any conditions not explicitlyc o v c d by the table results inan"1don't know whatto do" failure routine being executed. Wheneverthat occwf, the system simply stops and ask- forinstructions. If the condition can be corrected, ahuman programmer can enter a few more mles intothe state-table and the system can continue. By thismeans, the system gradually learnshow to handle alarger and larger range of problems. This cxtensi-bility of the system to new problems is essential inaresearch facility which. by its very ~ t m ,wi l lusually oprate at the very limits of the arrent stateof knowledge.

Sucha finite -state machine hierarchical controlsystem has been implemented on a microcomputernetwork. This network, shown in F igwe 6 has beenunder evaluation as a control system for the robotsin the AMRF".

b > b

The logical structure of Figure 5 is mappedinto the physical structure of Figure 6. Thecoordinate transformations of Figure 5 ar t impie-mented in one of the microcomputers of Figure 6.The elemental move trajectory planning is imple-mented inasecond microcomputer of Figure 6. Theprocessing of visual data is accomplished in a thirdmicrocomputer, and the processing for force andtouch data in a fourth microcomputer. A fifthmicrocomputer provides communication with aminicomputer wherein reside additional modules ofthe control hierarchy. It is anticipated that these willevennully be embedded in a sixth microcomputer.

Cornrnuniation from one module to anotheris accomplished through a common memory 'maildrop" system. No two microcomputers cornmuni -catt directly with each other. This means thatcommon memory contains a location assigned to

16 a B I T ~ Pr

16 B l l p P 16 a l r pMlNlCOMPUfER 16 COMMUNICATIONS ~ COORO

INTER FACETR AJ.

XFORM CALC.I

8 l i .

pP BUS

L

VISIONINTER F A C EVISION

h

16 611 pP16 6lTpP

A R MINTERFACE

COMMONMEMORY

C A M E R A

cROBOT

I I

77SWITCH PANEL

MISC. SENSORS

A

Journal of ManuPnunng SystemsVolume I. ?(umber I

every element in the input and output vectors ofevery module in the hierarchy. No location incommon memory is written into by more than onecomputing module, but any number of modulesmay read from any locatioa.

Time is sticed into 28 miibecond inmments.At the &ginning of each increment, uch l o g i dmodule reads its set of input vllucs from theappropriate locomputes its set of oback into the comillisccood i n t dends.Any of the1which do not complete their computations Morethe end of the 28 millisecond interval write extra-polatcd estimates of their output accompanied by aflag indicating that the data is extrapolated. Theprocess thcqrepeats,

Each logical module is thus a state-machinewhose outputs depend only onitspresent inputs anditspresent intrmal state. None of the l o g i d modulesadmit any intermpu. Each s t a r t s itsread cycle on aclock signal, computes and writes its output, andwaits for the next ciock signal.Thus,each logicaimodule is a finitc-state machine with the IFITHEN,or P=HfS) properties of an arithmetic function. .

The common memory 'maildrop"communi-cation system has a number of advantages anddisadvantages. One disadvantage is that it takes twodata transfers to get information from one moduleto another. However, this is offset by the simplicityof the communication protocol. No modules talk toeach other so there is no handshaking required. Ineach 28 millisecond time slice, all modules read fromcommon memory beforeany are allowed to wntetheir outputs back in.

The use of common memory data transfermeans that the addition o f each new state vanablerequires only a definltion o f where the newcomer isto he located In common memory. This informationis necded only by the module which generates it sothat It knows where to wnte it, and by the moduleswhich read it so that they know where to look. Noneof the other modules need know, or care, when sucha change ts impiemented. Thus. new microcomputerscan easily be added, logical modules can be shlftedfrom one microcomputer to another, new functionscan be added. and even new sensor systems can beintroduced with little or no effect on the rest o f thesystem. As long as the bus has surplus capacity, thephysical structure of the system can be reconfiguredwlth no changes requlred in the software resident in

the logical modules not directly involved in thechange.

Funhennore, the common memory alwayscontains areadily accessible map of thecurrent stateof the system. This makes it easy for a systemmonitor to trace the history of any or all of the statevariables, to set break points, and to reasonbackwards to the source of program erron or faultyl O g k

The rmkompute -write -wait cycle whereineach module is a statemachine makes it possible tostop theprocess at any point, to singie step througha task and to observe indetail the performance ofthe control system. This is extremely important forprogram development and verification ina sophisti -cated. real-time, sensory -interactive system in whichmany processes arc going on in parallei at manydifferent hierarchical leveh.

The hierarchical control stnffure just des-cribed is a generic concept which canbe extended toapply to a wide variety of automated manufacturingsystems. NBS plans to use this conceptual frame-work for the control system and data base archi-tixture for the AMRF. f igure 7 is a block diagramof the control system planned for the AMRF. Thesquare boxes arranged in the hierarchical structurein the center o f the figure represent the taskdecomposition modules at the various levels o fcontrol.

At the lowest levei in this hierarchy are theindividual robots, Ni C machining centers, smartsensors, robot carts, conveyors, and automaticstorage and retrieval systems, each o f which mayhave its own internal hierarchical control system.The bottom row of boxes represents the controlsystems for these individual machines. The smallsubboxes labeled s and C correspond to the sensoryand command interfaces to these control systems.The command input to the robot in Figure 7corresponds to the Y3 Elemental Move Moduleinput in Figure 5.

The bottom row of control modules in Figure 7is organized into work stations under the secondrow of work station control modules. A workstation may consist of a machine tool. a robot. and aset of smart sensors. It may also consist o f a set ofrobot carts, or an automatic storage and retrievaisystem with its associated robot. A machine workstation control module accepts input commands ofthe form <MMACHINE P.ART x>. A materialhandling work station may accept commands of the

Journal of Manufaccurlne S!sternsio lurne I. Vurnoer I

form <MOVE TRAY Y TO WORK STATION behavior ofthe work station to adapt to unexpected0.The machine work station controller decom- conditions such as broken tools or defective orposes itscommands into sequences of subcommands missing pans.to the machine controllen of the form <FETCHP A R T X>, <INSERT X IN FIXTURE Y>,<EXECUTIVE CUTTING PROGRAM t>.<CLEAR CHIPS>. etc. T h e material handlingwork station decomposes its commands intosequences of subcommands of the form <DIS-PATCH C A R i A TO PICKUP STATION B>,etc. in both cases, the decomposition is performedin the context of feedback information that is passedthrough the factory status data base shown on theleft of Figure 7.

The work station controllen may containprograms written in the form of state-tables, orproduction rules. Th is formulation wi l l allow the

~ e v e r a ~work station control units are organ-ized under and receive input commands from a cellcontrol unit. The cell controller schedules jobs.routes parts and tools to the proper machines. anabalances the workload among the work statlonsunder its control. The cell controller makes sure thateach machlne has the proper tools at the proper tlmeto perform the required work on each pan.

Programs in the cell controller are also wnt tenin state-cable form and can contain any number ofrules for adapting to error conditions such as toolfailures or changing pnoritles.

Several cells could be orpnltrd under a shopcontrol unit. However, the A,MRF lnltlall> at least.

29

Journal of Manufacturing SyrtmuVolume I. Number 1

will be considend as a single cell,and hence onlyonectilcontroller is planned. The possibility existsfor either further expansion of the AMRF oremulation of other cells if the research taskdemands it.

There are two data bases planned for theAMRF. On theright of figure 7is a Part Data Bascwhich contains design data such aspar?dimensions,d e s i d grip points for robot handling, group tech-nology codes, and material and tooling require-ments. A second scction of the right-handdata basecontains proccss plans for routing and schedulingand robot handling as well as cutter location datafila ncedcd for performing the various machiningoperations. These process piam arc, in fact, theprograms required at the various lcveir of thecontrol hierarchy in order to perfortu the neccssaqmanufacturing operations. Thus, the right-handdata base is, in parr, a program library whichcontains thecontrol programs needed by thecontrolmodules at the various levcis of the controlhierarchy. A third section of the righthand database contains data related to fc+dsand speeds whichmay be changed as a result of tensed conditions inthe factory environment.

When an order is entered into the shop controlmodule, the process plan to make that pan isdedin from the right-handdata base. The pr-plan ishierarchically structured so that at the top there isonly the name of the process pian. This name is sentto the cell control. The cell control computeraccesses the data base which calls in the sequence ofsteps (i.e., the program) that is the process plan atthe cell level. Each command ->inthis program ispassed in sequence to the next level down, which isthe work station. As each cell output commandenters the work station, it is the name of a processplan for the work station. The work station thengoes to the part data base as its level and calls up thesequence of instructions required to decompose thatproc.ss pian for the robot or for the machine tool.

The data that reside in the pan data base comefrom an interactive design graphics system and aninteractive process planning system shown at thetop right of Figure 7.

On the left o f Figure 7 is a second data basewhich contains dynamic Factory Status infonna-tion. This Factory Status data base is also dividedinto three parts. On the far lef t is a managementinformation and control data base. Entnes or

queries to and from this data base enable manage-ment to monitor and manage the whole factory bysetting priorities or entering control parameterswhich alter the mode of operation of the controlhierarchy.

The sccond Kction of the Factory Status database contains the status of tach machine tool androbot. in the plant as well as the sta tus of eachcornputex inthe control hierarchy: What program iseach machine running? What step in the program?How long in that step? What part is being operatedon?, etc.

The third section of the Factory Status basecontains the s t a t u of each parr in progress. In thisdata base, there exists a data fiIe corresponding toevery part that gives the part name, the tray that istransporting it, its position and orientation in thattray or in the work station, its state of completion,and a number of quality control parameters.

Al l thee data bases are sewed by severalInput/Output(I/0)controllers. The Factory Status

ais0has ahierarchy of feedback procrssonthat scan the various leireis of the data base andextract the information needed by .[he controlmodules at the next higher level. As in themicrocomputer robot control network, informationis passed from one level to.another, and from onecomputing module to another through the data basewhich serves as a common memory. This makes theSystem modular and defines the interface betweenmodules to be the data base. Thus, specification ofthe data base specified the principal interfaces of thecontrol system. This means that as long as a robot ormachine tool controller can read from and write tothe data base. it can be added to or deleted from asystem with a minimum o f impact on the othercomponents of the system. '

Because the status data base will be updated ateach time increment, it wil l always contain acomplete and current state description of the mirefactory. Th is will make it possible to restart thesystem easily in the event bf a computer systemcrash. I t wil l also be useful as a debugging tool.Activities of the various modules and of the systemvariables themselves can be traced and recorded fordebugging, analysis. or optimization.

The control architecture has been described inconsiderable detail since it is this feature that mostclearly distinguishes the A MRF from "just anotherFMS". T h i s system wilI provide the modularity

Journal of Manufacturing S w e r n rlolurne I . \umber I

needed to carry out the NBS research program ininterface standards and wil l eventually make FMStechnology practical for many smaller shops.

Although it will require a certain amount ofresearch to construct the AMRF and to test theconcepts on which it is based, the AMRFitself is notconsidered a research projcct. As various portionscome online, research projects, often withuniversityor private sector cooperation, wil l be started. Manyof these projccu will deal with new and improvedsensors to monitor machine performance. Otherswill deal with the problem of calibrating sensors sothat the product dimensions (not the sensorresponses) arc traceable to National Standards. Ifmore than one machine is involved in the manufac -ture of a part so that the refnturing effecu thecritical dimensions, this traceability becomes acomplex problem How both the mechanical opera-tions and their supporting software a n validatedopens v a t new areas for Measurement AssuranceProgram technology. Along with this metrologyresearch will. go research on the detailed nature of

. the data formats at each interface to determine howstandards can be designed so as to neither compro-mise proprietary interests, nor inhibit innovationsThe A.MRF, like other Bureau facilities, will bemade avallable to university and industrial groupsfor nonproprietary research in manufacturingengmetring which lies funher afield than themetrology and standards of NBS.

The AMR F is only one in a continuing series offaclllties that permit NBS to fulfill its designatedrole as the nation's measurement and standardslaboratory.

Concluding Remarks

AcknowledgementsThe Automated Manufacturing Research

Facility' owes its existence to the foresightedmanagement of the National Bureau of Standards

and the Depacment of Commerce, and the supportand encouragement of the machine tool community.especially those who have joined in the program asRcscarch Associates. The support o f the Depan -ment of Defense at various points in the programhas been most helpful as has been the universitycommunity working with us both informally andunder what is hoped to be an expanding grantsprogram. The AMRF is truly a National effort.

ReferencasI. Camema. J. M. and Huia. 0.. 'NBS Techntal Note 844'.

Nu~ondBureau of Sundudt , Wuhlngton. D.C.. 19742. Hacken. R.. Si-% J., Barchur. 8.. Lazar. J.. R m e . C.. and

Stein, P. A ~ u w l sof f h CIRP. V a l m 3. 1977.nu. L. F Q U ~ I ~ .E..Kelly. J C., Kenntcoct. P .~rld.M..Moorr.D..andWdlln~~onJ..(IGES YIJ.26M ResponseComnuua?. "DighlRepmmtatron for Communlacron of P TW UC~Definrtron Data'. Ropovd Amman Nauonal Standard. EngrnetnngDnwng a d Rebred Docruamutlon Pnctlcu. Approted 4\51sundud. kpnaber 1981 .

4. Ham. 1.. and Gonpwam. T.. 'Appliauon of Group TechnologConcept for H i g h ProduclrvttyoCNBS ShopOpmtroru'. RWK torNBS. The Pennsyi-Sum L'nivrmty. licuvenrty Park. Pennsi -van% 1981.

S. For eumpk. the yvubk Mruior Manufaaunng System o iCimnnau Mikron.

b. Antomorion of Smd Batrh Produrrron. Vatronai Eng!ncennglaboratory. East Kribndc. Ciugow. Scotland. 1978.7. Hmhnl. M. E.. 'Worfd Trends rn Advanced L1anu:actunng

Technology'. Pmrrrai~.0013Prh . I n 4 TrrJervrce WunularrurtnpTechnolo@ Confertnw. 0d.ndo. Ronda. 19ffLSutton. G.,C~rman.Cfshnolog). o i Macnlae Tools"-4 Surrey

of the Succ-ot.che-An by the Machtnc Tool Task Force. LawrenceL~vcrmoreLaboratory. LCRL-52960. Volume 1-5, 19809. Gtllap~e.L. K.. Advanres m Deewrrnr. Socrny o i Manuiacunng

En@-Dtubarn. Mlchtgra 1978.10. For example. the FW 5 System forengme blocks at Yawman Dteselhas duardcd opt ra l sensondue to cast Iron dust.11. D e V n a , U F. and Agaplon. J..'A Study of Hach~nabtltt\ DataBanks Adaptable to an 4utornattc Hanufactunng Test Facrl tn 'Report for SBS. Cinlveotty o( Wlsconsm, Hadisan. Wlrconrtn. 1982.1 2 Dct'ncr. M. f . and 4pp1on. J.. 'A Survey o[ ChIp BrmkrnjTooling Adaptable to an Automattc Hanufacturrng Test FmIIt\Report for Y 8s. L'nlvenlty of Wlrconsm. Hadaan. Isconrln. 198:.13. Tuffensamms. K:: ClRP Annu&. Volume M?). 1981. pp 55314. Cook. S H.. Subnnnmn. K.. and Lsllc. 5. A.. Turvey o i :neSute -of-the-An of Tool W a r Senrmg Tcchnqua ". prepared Insuppon wtn WSF Cant UCI 03861. Publ. by Yaswchuwu InstttuceoC Technology. 1975IS. Albus. J S .Barben. 4. J..anP Yagel. R. Y .'Theom and Pracrlcoof Hlcarchlul Control'. Ptorrrdrnqs. 3rd lfff Cornpurer Socu*tlnrrrnorlonal Conjrrmrr. Wuhmgron. D C.. 19%I16. Albus. J S.. Barbcra. 4 J . Fttrprrald. L1 L.. and hashman. J .'Sensorv Intenctne Robots'. Anna13 01ClRP Toronto. Canada.I98 I17. Albu. J S.. Barbm. 4 J .Fitrgcnld. J L.. ~H tc r~ r ch rca lCantto1for Senran* Intenctrve Robots'. Proczcdtngs. Iltn lnrrrrrartonolSImpcrtui on lndwrrral Robots and Robor Erhtbrt. Tok!o. Japan.1981.

Y

'Contnbutlon of the Yarional Bureau of Standards. Not sub~cc!IO copyrrght.

3 1

Jounul of Manufanunnu SystemsVolumc I. Number I

Author($) BiographyDr. John A. Simpson is presently Dimor of the Center for Manufacturing Engineering at the National

Bureau of Standards? (NBS), with background in the fields of dimensional metrology. electron optics.photwpticai instrument design, and mechanics: Prior to his appointment to this position, Dr. Simpsonscned as Chief of the Mechanics Division, Deputy Chief o f the Optical Physics Division. and Chief of theElectron Physiu Section, all at NBS, and was a Research Physicist at &high University.

Dr. Simpson is a FeiIow of the American Physical Society andhas been active initsDivision of ElectronPhysics. He has sendon the National Acldcmy of Scicnces/National Research Council Panel for NATOPostdoctoral Fellowship. In 1975, he received the Department of Commerce Gold Medal award inrecognition for his accomplishments inmodernizing the metrolow services of NBS. ln 1980, he received theNBS Applied Rcsearch Award (jointly) forhis part in thedevelopment andimplemcntation of theautomated,self-correcting threc-axk coordinate meamring machine which enables rnanufmuren to characterize andcorrect errors in machine tools during manufacturing processes.

Dr. Robcn J. Hocken is p m n t l y Chief of the Automated Production Technology Division at theNational Bureau of Standards, with background in the a m of critical phenomena. machine tool metrology.thm4irnensional metrology, lascr optics, manufacturing technology, and polarimetry. His divis~onat NBSdevelops and maintains competence inmachine tool dynamics, precisionengineering, robotics. and computeraided manufacturing, and is the incorporation of metrology into the ' * metal workingprocesses, including the stan for integration of equipment up to the uring cell level.Prior to appointment to this position, Dr. Hocken held a National Ruearch Council Postdoctoral position atNBS. was Leader of the Dimensional Metrology Group, and Chief of the Dimensional Technology Section at;YBS.. . Dr. Hocken is a member of the American Physical Sociefy, the American Society for-Testing and

Materials, and the International Institute for Production Enginecling Research. He is a widely recognizedexpert in production metrology and recipient of the Taylor Medal of CIRP for caritributions to metrology.the Department of Commerce Silver Mcdd for thmdimensional metrology, the IR-100 Award for thelarge-scale measuring machine. and the NBS Applied Research Award (jointly) for the development of thethree-axis measuring machine which enablqs manufacturers to characterize and corrcct errors in machinetools during manufacturing processes.

Dr. James S. Altrru is presently Acting Chief o f the Industrial Systems Division and Manager o i theProgrammable Automation Section. Center for Manufacturing Enginering, National Bureau of Standards.He has receaved the Department of Commerce Silver Medai for his work in control theory and manipulatordesign and the Industrial Research IR-100 award for his work in brain modeling and computer design. He isthe author of numerous aniclcs in technical journalsjncluding a survey article on robot systems for ScienlrJicAmerican (February 1976) and an entry to trcyclopedia Amerrconu on "Robors". He has wntten anlcles onrobotics for OMNI Magazine, lUeruf Working News. andBYTE Magazine. Dr. Albus has also been quoted inorher national magazines. such as TIME, Forrune. Reader's Digest, NEXT, and Discover, and has appearedin a number of TV interviews.

Before coming to the Bureau of Standards. he designed electr--optical systems for more than4 5 NASAspacecraft. seven of which are on permanent display in the Smithsonian Air and Space Museum. For a shontune he served as program manager of the NASA Anificial Intelligence Program.

His latest book Bruins. Behavior and Robotics was publlshed t y McGraw -Hill in Yovember 198I.Dr.Albus has also written a book entitled Peoples 'Capiralism: The Economm of rhe Robot Revoluiion in whichhe addresses some of the ctntral socral and econornlc issues raised by the advent oicomputer contrcllled robotindustries.

lure readmgs. and home positions. Arecord is kept of each failure.

The computer -aidedprocess planning(CAPP) module is a generative programbased on parr codes. I t lets engineers

Systmn Payoffs

rotational geometries to existing forgingsand geometries, sornetlmes avoiding theneed to retool, and cutting machiningtime if the process is near net shape.

Additional capabillties can be incor-porated into the CIM system, providingas-needcd flexibility. These capabilitlesmay include a robotic sermetal painting

ic application ofbrazing alloys,~ ~ ~ u t e ~ ~ ~ ~ t r o l ~ dprocess parametersand pan position in^ fur vacuum plasmadepmition, and DNC Iaserdrilling. Witha system that has already increasedproductivity and reduced costs by asmuch as 2.5%. addedcapabilitics can onlyenhance the AEBG CIM system.

Manufacturing at NBSA n update on the National Bureau of Standards'facility for automation research

DR. ROBERT J. HOCKENandDR. PHILIP NANZETTANBS Center forManufacturing Engineering

I N LATE 1980, the National Bureau ofStandards (NBS) made a decision todevelop and construct an AutomatedManufacturing Research Facility(AMRnby the midas to support research onmachine interfacing and inspection ofparts produced in small batches.

The facility is designed to serve t h mmajor sectors-industry, government.and academia-in developing, testing.funding. and implementing advances inautomated manufacturing. The AMRF isa direct resource for members of themanufacturing community who wish tosee demonstrations of existing and near-future technology as an aid in their ownautomation decisions. The project alsoplays a crucial role in meeting the NBS'legal commitment to leadership in stan-dards and measurement activity, espe-cially in a time of rapidly advancingtechnology. Finally, this new facilityserves as a "test bed" for industrialresearch associates, university workers,andscientistsat theNBS whoarcprepar -ing the way for the computer automationtechnology of the next decade.

testing and Calibration

The legislation which created and gov-erns the NBS specifically assigns to it the

08

function of "tesring and calibration ofstandard measuring apparatus [and] thesolution of probkms which arise in con-nection withstandards. "Two major prob-lem areas related to standards arc beingaddressed in the AMRF. These concerninterchangeability of parts and inter-changeability of mnnufacturing units.Work on the former involws dimensional

accuracy and falls under the heading ofdeterministic metrology. Work on thelatterdeals with interfacestandards whichallow easy transfer of part informationbetween different manufacturers' equip-ment and easy replacement or upgradingof design and manufacturing equipment.

Deterministic Metrology. Gage blocksand other artifact standards provide thebasis of a mature technology whichassures accuracy by direct comparison ofdimensions. For complex parts, thecomputer -controlled coordinate measur-ing machine has successfully automatedsuch comparisons, but it still dependsupon measuring the part itself.

In a welldesigned automated manu-facturing process. if one good part isproduced and if the process parameters(such as cutting force and temperaturedistribution in the machine tool) arecontrolled or corrected. then subsequentparts cut by that same program are alsolikely tobe good. Thisobservation under-lies the philosophy of deterministicmetrology which conantrates on under-standing. monitoring, and controllingthe manufacturing process itself ratherthan checking the part after cutting isfinished. Thus, the standards' responsi-bility whichcalled for careful custody ofmaster gage blocks and calibration ofother artifact blocks from thesc mastersisnow advancing into a technology whichrequires a fundamental understanding ofways to monitor' and conttol basiccutting processes.

Close attention to process control hasled NBS scientists to work in the area ofsoftware accuracy enhancement. Bydeveloping an "error map" for a CNCmachine or coordinate measuring ma-

A horizontalmachiningcenter work nation within theAutomatedManufacturingResearch Facilit -v ofthr NBSir tendedby a Cincinnati Milacron robot.

October 1983/Mmufacturing Enginafing

chine and incorporating a correctingalgorithm Into 11s control computer. theaccuracy of standard equipment can bemarkedly increased.

Interface Standards. In the area ofinterface standards. NBS scientists coor-dinate and support work on a commondomain graphics exchange standardknown as the Initial Graphics ExchangeStandard (IGEs). This standard, the workof an industry -wide committee, allowsthe transfer of CAD data between thesystems of various manufacturers. Pre-and postprocessors for ICES have beenannounced or promised for all of themajor CAD systems.

American industry has reached a con-sensus that it must increase its concentra -tion on quality and productivity in orderto develop and holdour country's leadingposition in international commerce.Thus. the NBS has found strongindustrysupport for development of the AMRFfrom industry, universities. and otheragencies of the government concernedwith manufacturing.

Under the NBS research program, atopic o f interest to an industrial sponsoris investigated by a research associateemployed by the sponsoring firm andassigned to work at NBS. Frequently. aprogram uses amachine toolwhichbelongsto the sponsor, but which is temporarilyincorporated into the AMRF. Such coop-erative research activities serve to assurethat theAMRFcombiM placticaieconomicsolutions to near-termproblems with thecapacity to advance the state of the art.

Developing the AMRF '

Developmental planning, procure -ment,and teotingofthiemanyelementsoft h e A M ~ ~ % ~ e ~ ~ t ~toextend wellinto1986. ~ ~ ~ h ~ ~ i ~ yhasalready begun

hr elcmmts of the laciiity azincorporated. The AMRF, &$

presently defined, consistsmachining centers (a verticalmachining center, a horizontalmachining center, a large tunringcenter,and a smaller turning center), B cand deburring station, an automaticinspection station, and a material hand-lingcomplex. The twomachiningccntm.one of the turning centers, and rworobots are present now on the shop floor,the automatic inspection station will bcinstalled by late 1983.

lnstrumt Shop. Anaddittons1IO,OOOft*(930 mz) off the main shop floor isutilized for a toolcrib area, computerspace, electronic and mechanical supportlaboratories, and a small conferencetraining facility.

MduLr Architecture. An early deci-slon in planning for the AMRF determinedthat the control architecture must be

modular and designed so that it can beimplemented in steps. Under this designapproach, a shop or factory can under-take automation in steps which are eco-nomically tolerable and still reasonablyexpect the various pieces to fit togetherinto an integrated production facility asmore cornponenu amadded to the system.

The AMRF itself i5 being constructedin a similar stepl ike fashion. &forecontrol system development advances tohigher levels of demonstration. a workingunit consisting of a single machiningcenter, its robot tender, fixturing. mater-ial handling interfaces, and varioussensory systems is fully developed andtested as a freestanding work stationdriven by manual or simulated hlgher-k v c l commands. -

Coardinated control of several works ~ t j ~ ~d ~ l ~ n g ,for the moment, with acornm~nfamily af parts wil l be main-

Interconnections. A majorobjmive of the AMRF is to study theinterface problems which arise when

into an integrated facility. In the processof solving such interface problems. NBSscientists can propose changes in equip-ment to improve interconnection andcontrol. Work with softwan ascuracy~ n h ~ n ~ ~ c ~ thas defined a whole rangeof new interface requirements for machinetool controllers. In a slmilar way,development of off-line programmingand ml-timecontrol for robots has givennew insight into the need for moresophisticated robot controller interfaces.

Various sensor types are being em-ployed in the design of the AMRF, mclud-ing force. proximity, vision, temperature,

and vibration. High-precision dimen-sional sensing is incorporated in tool-setting stations which use linear variabledifferential transformers (LVDTS) thathave been "hyperlinearized " by softwarecorrection techniques developed byNBS scientists.

Design of the control structure isb a d on distributed computing powerwhich takesadvantage of rmn tinmicroeltctronics. Because of thisdesigndecision, a high level of computationalcapacity can be economically incorpo -rated into sensory systems and other ele-ments low in the control hierarchy. Forexample. a dedicated safety system withits own controller is being incorporatedinto each work station inorder toprovideredundant safety checking for theprotec-tion of both humans and machina

Using Avaihbk Equipment. in con-trast with the approach adopted in theBritish ASP Planand the Japanese FMCProject, the NBS is using standard,modern, commercially available machinetools in the AMRF. The ASP and FMCapproaches call for the design of newmachine tools for automation. The NBS,based on reviews of published studies ofthe international state of the art inmachine tool design, has chosen insteadto rely upon the engineering experienceof a mature machine tool industry and toavoid radical design changes for theAMRF machining centers and otherfacility hardware.

The AMRF is meant to be a nationalproject which draws upon and contri -butes to work on automation of smallbatch manufacturing being conducted byIndustry, universities, and other labora-tories. While theeffortsofthe NBS relateto problems ofstandards, the bureaualsoencourages close cooperation wlth thosewho are addressing other components o fthe field.

Manufacturing Engineering/October 1983 69