Keynote Paper
Computer-Aided Process Planning: The Present and the Future lnyong Ham ( 1 ) . The Pennsylvania State University/USA; Stephen C.-Y. Lu, University of Illinois a t Urbana-Champaign/USA Abstract: This paper examines the current status of. and suggests some future directions for. research efforts in an area important for computerintegrated manufacturing: computer-aided process planning (CAPP). Rather than discuss a specific aspect of the subject or present details of a particular prototype system. the major emphasis i s on the global perspectives of fundamental issues involved in developing computcr-based planning systems for various manufacturing tasks. In reviewing the current rcsearch, references are made to technical papers presented a t the 19th ClRP (Inrernationol Institution for Producfion Engineering Research) International Seminar on Manufacturing Systems held at the Pennsylvania State University, June 1.2, 1987. with the major theme of Computer Aided Process Planning. In suggesting future directions. an integrated planning framework as a logical exlension of current CAPP activities i s proposed. The need for. and challenges of, such an integrated phnning approach to manufacturing problems are summarized. and. specifically the potential role of artificial intelligence (AI) based techniques within this framework are explained. The objective of this paper is to promote a better understanding of the nature and potential of manufacturing-related planning tasks by critiquing the present efforts with respect to the ultimate goals of unmanned production ~n the future.
Key
Words: Computer-aided process planning (CAPP). Artificial intelligence (AI) in engineering. Knowledge-based expert systems, Manufacturing automation. Computer-integrated manufacturing, Simultaneous engineering. Production scheduling. Factories of thc future.
With Contributions From: W. Evershem ( I ) , Technischcn Hochschulc Aachen. FRG (with M. Cobanoglu and T. Luszek) H. J. I . Kals (1). University of Twente. The Netherlands F. Kumura ( 2 ) . Tokyo University. Japan V. R. Milacic ( I ) , Univcrsity of Beograd. Yugoslavia F. 0. Rasch (1). NTH-SINTEH. Norway G. Spur (I). Tcchnische Univcrsitat Berlin. FRG H. K. Tonshoff (1). Universitat Hannover. FRG H. J. Wamecke ( I ) , Universitat Stuttgart, FRG R. Weill ( I ) , Institute of Technology, Israel
1
Introduction
Since the beginning of the last decade, it has been recognized by borh academic and industrial communities that planning in a manufacturing environment is vital l o arhieving the ulliniate goal of unmanned and integrated faclories in the future. Planning is not oiily a key link between design and tnaiitrfacturing within a production enviroirmvnt, but is also a decisive activity which distinguishes a smooth-running farlor) from a chaotic one. To date, many research and development efforts h a w h*en devoted t o analyzing, modeling, and automating such planning activitir.9. Many computer-based planning systems [ l , l l ]have been developed which siiccessfully demonstrate the great potential of thpse efforts As the need to intrgrate design and manufacturing increases, the need for more robust planning systems will also increase accordingly. Given all Ihr research and development efforts expended thiis far, it is discouraging til realize that only a few romputer-based planning systems have actually been usvd by industry, and that even kwer hare reached the stage where they could make a significant impact on manufacturing practice. Part of this slow propless is due t o the complex and dynamic nature of the planning domain which uill pose great challenges t o the research community for matiy years to come. A more important reason is, however, the lack of a theoretically sound foundation and a scientifically rigorous base for current planning approaches. As we n w r the second decade of manufacturing planning research, i t is important to c i ili, ally reexamine the methodologies and approaches currently being pursuf.il in our research community t o determine whether they will lead t o the goal of practiral computer-based planning systems. If not, what factors prohibit us from success, and what strategies are needed t o address those factors? This paper tries t o address thPse questions in order to suggest and define future research directions for this important area of computer integrated manufacturing. We examine critically some basic, but often neglected, concepts relating t o planning in general, and to computer-aided process planning in particular. In reviewing current research efforts in this area, references are made mainly to papers presented a t the 19th CIRP International Seminar on Manufacturing Systems held a t the Pennsylvania State University, June 1 and 2 , 1987 [ I ] . The main theme of this seminar was "Computer-Aided Process Planning," and 50 technical papers, contributed by researchers working in this area world-wide, were presented. Due t o the broadness of participation and the quality of the work presented, this meeting can be regarded as a comprehensive world-wide status report of the current approaches t o this important subject. In pointing out future directions, we will not merely formulate a list of suggested research items for the community, Rather, we will present our view of the comprehensive goal toward whirh this technology should logically evolve This goal subsumes any specific subjects for individual research, and should be cooperatively pursued by researchers in the community as a whole. We observe t h a t manufacturing planning activities are undergoing a long development from manual planning in the past, through computer-assisted planning and interfaced planning a t present, to integraled planning and intelligent planning in the future. Even as we are developing mnre computer-assisted planning systems and t.rying t o interface them, we should look ahead to the challenge of developing integrated planning systems and adding true intelligence to them. Integrated planning is different from interfaced planning, and intelligent planning requires more robust Artificial Intelligence (Al) techniques which go beyond these simple rule-based approaches. To illustrate this point, we will present a framework
Annals of the ClRP Vol. 37/2/1988
for integrated planning and describe the potent,ial role of AI-based techniques in achieving intelligent, planning It should he pointed out that a pappr of this kind normally tends t o be somewhat, subjective and will certainly raisp more questions than answcrs. It should also be noted that,, in order t,o achieve oiir purpose of promoting a better iinderstanding of manufacturing planning, this article will take a somewhat crit,ical vicw in examining current research activities. In doing so: we d o not intend t o criticize individual research. Rather. we hope t,o define challenges t h a t must be addressed by the entire research community, including ourselves, in the future. In other words, our focus will he more on finding out what is lacking in current approaches rather than on applauding what has been done so far. But this is, by no means, meant lo undervalue either the significant efforts present,ly pursued by t h P research community or t,he achievements ahat have been obtained t o date
2
The Fundamentals of Planning
Since we believe t h a t a theoretically sound fniindatton 1s vital t o the SUCCPES of computer-aided process planning research; w e will first lay t h a t foundation by discussing some basic concepts involved i n planning in general, and in coniputerhawd process planning in particular T h r basic questions discussed will be: what is planning, what is manufacturing planning, and what is computer-aided procrss such as how to d o planning and or how to drvelop romputerplanning? ISSIIPS
based process planning systems will be addressed later Thus, we will first try to rlrarly define thc problem and its srope. and our goal before searching for soliit.ions.
2.1
What is Planning?
Planning is an intrinsic part of intelligent beharior, often performed uticonsriously by human beings in everyday life. In a generic sense, planning can he viewed as the activity ofdevising means to achirve desired goals under given ronstraints and with limited resources. Thus, three basic components of any planning artivity are. goals, constraints, and resources. An intelligent planner, whether a human being or a computer program, should have the ability to understand, represent, and manage these three components. Sometimes, the term planning is Pxt,ended t o include activities relaled to plan monitoring, which insures t h a t plans generated can hr executed properly In this regard, an intelligent planner should also have the ability to takP feedback from the plan execution phase for future improvements, and to adapt t o its changing environment for different needs. In sunimary, it is the closed-loop activity of planning, monitoring, and re-planning which gives a planner the ability to deal with real-world situations full of surprises Planning is a multi-perspective problem. Its constituent activities of problemsolving, constraint-reasoning, goal-achieving, resource uliliration, and ronflictresolution. These activities are also part of intelligent human behaviors. For this reason, planning has been the focus of many studies within different lheoretical and practical disciplines. While some have tried to understand behavioral models ofplanning, others have tried t o model these behaviors on computers. In general, thpsc studies can be classified into two rategorips: The theory ofplanning The st,udy of planning t,heory is aimed a t improving our understanding of planning activities, and hence a t enabling US to d o better and more effective planning The mechanics of planning The study of the mechanics of planning focuses on developing algorithms which allow non-intelligent entities (e.g., digital computers) to perform planning functions in an intelligent manner. Simulating planning fiinct,ions on computrrs has posed a great challenge to researchers since the invention of digital computers. Since modeling hwnan intelligence on computers is the goal of artificial intelligence (AI) research, planning lim been an active subject of study in the A1 community since the field first evolved in the mid 19.50's 121. Many Al-based approaches to general planning problems have been developed whirh have showti great, potential for application Lo certain practiral domains i31. These studies on planning, whether traditional
591
or :\I-hasrd, h a w taught us Important lrssons which arp relevant to the rescarch in cornpiiter-aidrd inaniifarliiring planning. hniong ot hers, s p w r a l lessons that, are of partiriilar roncprn t o our siihjrrt are
INPUT
T h e thenry of planning is an rwr-poolring siibjprt
SPECIFICATIONS T
l ' h ~I l i ~ o r yand the nierhanirs of planning are inter-related and miit,iially suppnrt,ive in their evolution
-
I
I
l'he merhanirs of planning ran he iinderslood only after the planning theory I- mature enough rn
A soiind theor) and a rohust rnechanirs are hoth needed for a planning to he truly siircessful and practically iisefnl
. -
T h e points listpd above rorirtitiite the basic guidelines that have influenced and (IuggestiuiiS for. rompotcr-aided process planning respnrrh presented i n this papcr
PRECLDEh'CE RELATIONSHPS
1 . 1 1 ~ rritir-ism of,
2.2
REQUIREmNTS
What is Manufartiiring Planning?
In daily human art,ivities. planning is the coordinating link between intpriial intentions and external actions. Similarly, manufacturing planning can he seen a5 t h r r o o r d i n a t i q link brtivren the Intention of design engineers and those of manufacturing engineers in a prndurtion environment Designers express their int.rnl.ion ronrerning the required specifications 111 terms of the functionality and reliability of an engineering part Manufacturing engineers must then take proper actions to rpalize these intentions based on Ihr additional constraints impowd from shop floors. 'l'he goal of manufartiiring planning i s t o coordinate these t,wn kinds of artivities and ronstraint,s (ie. deqign intentions and manufartiiring actions) so t h a t a n overall effective produrtion unit ran be creat.ed In any production environmrnt,. sprcifir p l a n n i n g functions take place at diffrrent at,ages. and the global sum of these artir,ities constitutes L complete nianufarturing planning system T h ~ r rare many technical and non-technical(e g econornic and sorial) factors th at iriiist IIQ I a k w into consideration in arhieving an effective nianufactiiring plan werall On the technical side, manrifact,uringplanning functions ran he characterized a- :I!
Production Planning Hriefly, the goal of production planning is to decide the questions what. whpn. who. a n d where in n production rnvironmmt based osi a giver) duetime and available r ~ s o i i r ~ e This s is t h r highest level on the technical side of a manufacturing planning hirrarrhy. and it focuses mainlj- on the global prrsprrtives of product,ian bithout, louking intu tlie details of how t,he part is to be produred I'rodiict,ion planning has strong links to cooperation policies outside the terhnical domain It i s also referred to as ~chedulingin mine researrh effort.; 151 Prore-5 Planning T h e purpose of process plnnnin6 is to splprl, a n d drfine, in detail. the processes t,liai have t,n be performed in order bo transform raw material into a g i r r n shape Thr primar) rnhlerti\e I S l o ddinr feasible proreases C o s t and Lhroiighpiil are serondary objectives. and available rrsources (inarhine I n n l s . r u t t i n g tools and l a b o r ) art ac r-onyrraints Process planning includes 16' ~
arlrction uf machinr t n n l ~
selrction of tools s r t s
~
.-
selection of set-lips
srlwtion of machining operations a n d their sequence selection of cutting tools design of jigs and fixtures calculation of cutting ronditions delerrninatioii of
tool
paths
NC p a rt program generat,ion Figiirp 1 shows an t.xnmplp of the p r o r r v planning steps for convpntional machining operations i i '
Unlike prodiictinn planning artivit,ieswhirh are concerned with the production environrnent as a whole. all the decisions made at the process plaiiiiing stage are liiriitrd to a specilir part o n l y Like the production planning which serves as the link b d w - r m tkrhnical and non-technical planning, the process planning serves as the critical link between design and manufacturing. It is sornet,imes r a l l d macro-planning IS1
SELECTION OFHOLDMG DEVICES AND DATUMS
DETERMINATION OF PRODUCTION TOLERANC5S
MACHWING DATA
/EDll"G
OF PROCESS SHEET
/
Figure 1: Process Planning Steps for Conventional Machining Operations 17; a r r separate, thry should not be studied or modeled in isolation. Thereforr, w h i l e forusing on computer-aided process planning in the rest of this article, we will also stress irs intimate relationship with production planning and operat.ion planning. O u r proposed integrated planning framework covers all three t j p r s of p l a n n i i i ~in a logical fashion As we progress toward intelligent and inlegiated planning, the boundaries bPt,wt=rn production, process, and operation pimining w i l l gradually disappear
2.3
What, is Ckmiuutw-Aitfed Process Plannina?
Cnmputer aiitomat~ionof any rpal-world tank requires a detailed ~lnrlcrstanding of the task to be automat,ed. Not only do the individual compimc,nts of t hat t.ask (tlie micros, or components) need to he explicitly rrpresrntecl, but , more import,antly, the fundamental st,ructrirr of the task (t,hr macro, nr the system) nwst h e studied carefully so that comput,er programs can he drvel<,ped t o model its local and global logics. In that sense: before pursuing any romputer-based Pffort toward proces.s planning, it, is necessary for us to move one st ep back and ask ourselves: d o WP rpally understand the process planning tasks well enough fnr romput.er autoniation? In other words, the fundamental queations such as what is process planning. how is it being done presently iu w t u a l practice, are there niore logical ways to doing it miist he answered hvfore any programming efforts should take place. Given even the current staLr !>f computer technology, including art,ificial intelligencP, it is still not possible to rlevelop computer prograins t o intelligently carry out tasks unless we. as intclligvnt human beings, fully understand those tasks. In studying computer-aided process planning. it. is important t o differentiate efforts to ailtomate process planning activities froiii efforts to program these activities Even though computrra can assist in htrtli automating and programming, the degree of task understanding required by lhese two kinds of efforts is very different 1Vit.h some limited understanding of the micros of a task, it is powiblp to program a romputer to repeat those niicros This local understanding i s , however, not suficient for automating t,hat task, which additionally needs iindersianding of both niirros and the macro. This difference has a major bearing on the fundamental difficulties we are currrntly facing in automating process planning t,asks. Rerause our present. irnderstaiding of process planning activities i s limited to local micros. we can attain sonie success in programming specific kinds of process planning activities (e g., l i o w to machine a cylindrical p a r t ) , but a r r unable to aiiloniat,e the wholr process which would require a much broader iiiiderstanding of the necessary tasks.
Operation Planning Operatlon planning has an even narrower focus th an process planning, being conctmed w i t h the sppcific manufacturing operation rather than wit,h the parts whirh may reqiiiw sen=ral nperations. The goal of this type of planning is t o determine the details of the parameters t h a t will ensure the smooth rompletion of planned manitfar-luring opwations Once operation planning has been done. enough non-ambiguous delails should be generated for a part t o he dirrrtly m a n u f a r t n r ~ dnn a specific production machine From lhe planning hwrarchy point of view. operation planning is the lowest I e v r l a n d s rrv w as thr, link hrtuswn th r planning decision and an artlinl opt-ration It 1- s~iiiietiniescalled rnir.ro-~~lanning IS]. It, is Clear t h a t t h e w th rre types of planning activity. even though each has its own dist,inrtive goals and sropr. a rr highly interrelated in pract.ice They arp i n fact built iipon earh o th rr. aiiri shoiilrl h r considered as integral roinpoiients rather than separatrtl pirrrs of a r.ompletP ~i~ariirlartiiring planning syst,rrn. Alt,hniigh l he above definit.ions describe t,hesr planning functions as thniigh Lliry
592
3 3.1
Some Basics of Computer-Aided Process Planning The Import,ance of C h P P Research
Long hefnre computer-based autoniation was int,roduced onto the fact,ory floor in t,he a0tempt t,o increase prodiicl quality and decrease production cost, process planning was recognized as one of 1 hr most important activities in helping t o achieve highly competitive production units. Several previocs articles [1,9,10] h a w discussed the central role played by process planning in aut,omated factories by siirreying rpsearrh activitirs dirertly related to this subject, and this informa(,ion will not be repeated here It is however important to note t h a t there are many researrh t,a.sks which may seem to he irrelevant, but which in fact, contribute greatly to thr iiltimate goals of manufacturing planning. Good examp1.s of this are eKnrts to build logical links bPtween design and manufactiiring 1111, l o iniplmx-nl, thr ronrept of drsign for rnanufacturahility /121. to address the need of autimintkin iiitegratirin. and trr synthesize prodiict and process information
iniposslhhlr. Hithout, this frrtlhark f r o m the shop floor. i t hecoinrs dilfiriilt to i i w a q i i r p t h e , quality or gondncss of a p l m for fiitiirr enhanrrmmt
: I ? / , PIC I'lanning niay not br the r ~ p l i r i -tihjc-ct t of thpsp ~ t u d i hut ~ ~ (bey , all share ttir same goal with planning research: integratinK design with niaiiiifactiiriiig Generally speaking. prucprs plaiining i s rlo-ely rrlatpd to the iurgrnt new1 for factory integration arid is t h e fundamental striictore upon which a tightly integrated manufacturing organization ran hr built. Process planning is thr rritiral link hetween computer-aided rlt,sign (CAD) and computer-aided manufartoring (CA\4), hobh of whirh nred this indispensahle interface Thus, any pffnrts IPsiilt,ing in a higher level of integration among factory artivitirs ran hp V I P W F ~ as a rontrihiition to planning It IP for this reason th a t proress planning is often referred to as a critical s t p p in achicvlng c om put ~r - int r gr at e ~l inaniifarturing (CIM). In pxamining the iiiiportanrr of proreas planning. onr can rnrrelatp t h r dpvelopment of planning systems and coinputer aiitornation with the evolutioii of thP manufartiiring industry. Thr folloa.ing tablr summarizes this rorrelat,ion
6 In an aiitomated environment. process and operations planning n i i i s t rrsult in d a t a which h a w to deal with the utmost drtails and all I his dat a has t o he available hrfnre Ihe artiial maniifartiiring starts. Intrrpretation. adaptation a n d rmnplr,tion of #data in a later phaw is not applirahlp
The- abnve- difficulties arP real rhallrnges faced by resrarrhers attenipting to develop computer-aided process planning sy stema. Sump research efforts already rlwotrd t o addrrssing these iliffiriilties tlowevrr. we feel that they will not resiilt in signifiraiit impact unlrss means are devisrd to deal with these difficulties in a mhrsive and integrateri fashinn Fiirtherniorr*.t hrre are three interrelated aspects of (he subject that must he addressed cnopwatirrly by researchers working in t h e prorpss planning area in order to arhirw the goal of integrated planning I
enmve
.Automating eai-tinfi planning a r f i v i t i e ~ ( h m p i i t r r systems inlist hP dewloped ti, as+t and or aiitomatP some portinn of planning activit,ies according to the way in which they are ciirrently being perforiiwd hy hnimnn hcings T he majority of !.he present rewarrh i n niaiiufartiirirtg planning falls into tIIiis rategory ! l o /
Separated Automa
I
2 .4nlicipating future jifanninc rhallc-ngPs
The= requirements for maiiiifart.iuring planning i n frillire fartories must he antiripated, and planning techniques developed to meet, t,hese future needs Although this has hem rerngniwd = a n important artivity, t o dat.e there are only few research rfforts aimwf in this direction. The progrpssive introductioii of iirn automated mauuiacturuig systems. such as flexible manufartiiring rells or sysleins. leads 10 a task expansion i r i process planning as shc,wri i n Fig. 2 A functional c o r d a t i o n between (,he degree of aulomat.ion and t,he necessary planninq effort is obvious 1141.
our present production indiistry IS information-intensive Th e next challenge will be t o convert massive amounts of information into iisefiil knowledg? arid to utilize this knowledgp effectively to make better decisions. In ot,her words. new computer-based methodologies must be developed to help us go through the transition from the current information-int,rnsive industry to the prospective knowledge-intensire industry. A similar correlation ran be observed between diffrrrnt Lypes of planning s ) stems and st,ages of industry evoliition Whenever t,he manufacturing industry is a b o u t t o evolve into a new era, differrnt maniifacturing planning systems must be developed t o ensure the si~ccessof th a t rvoliition In the early labor-intensive industry, planning did not play an important rnle. It, wasn't until t,he matrrialintensive st.age t h a t manufacturing planning started Lo play an important role in industry Various c o m pu ter-b as d data-rrtrieval types of planning systems were drveloped for betler management of materials and other resources. They made significant contrtbiitiona to t,hr maniifartiiring indiistry a t t h a t time, hut w'ew insiiffirirnt onre the indiistry changed and hecaine information-intensive. A s a way t o rpspanrl thr tlw new challenges posed by massive anmontr of inforrna1,ioii in manufart,iiring, semi-generativp planning systems s tart to evolving as a major planning approach at. present,. I n the future, as we evolve into a knowledgeintensivr manufacturing indost ry. W'P uill be challenged by the need t u develop truly iiitegrakd and more int,elligrnt planning systems It is clear that the.successhil deueloprnrnt, of such a new planning paradigm is critical to the furthpr evoliit.ion of our manufarturing indost ry
3.2
3 Suggesting a more logical planning striirtiire-
mad^. hnsed on t,he rhararteris1,ics of variniis romput,er aiitoniatinri t e c h i i o l n g i ~and ~ planning approaches. on the most logiral ways of rondiirting process planning in practice so t h a t it becomes more autonialahle hy roinpntrra This rpqiiires a flexible iise of different, planning vrnarin*S's and the genpratinn'9 of alternat,iw solution's on a rost-selective hasi- T h e s e siiggcst.iiins ran he viewed as the feedhack f r o m conipiitrrha.spd ant.ornatiun ttd,ntilogy t,n process planning This rhallrnge. perhaps t hr mnst significant in i.wiiis nf i t s m s t - h e i i d i t r d u r n , I S . Iunfmrtonatkly, t hr nirnt r i p g l p r l r d onc. l3ut i t i 4 c,hviou.i tliat Iirnress planning races a iiuniher <,r i i i f l ~ e i i c cwhit ~ h will d i a r i g e the r n u t r t t t 9 d thr planning p r o c ~ d r ~ r r S i i g g ~ sions t should hr
3.3
The Challenges of C A P P Resenrrli
T h r iswes involved in planning in the manrtfact,lirlnqpnvironment require r.nordination hrtween people, organizat ions. prrtiaps iivcr t i m e and across distanr r , anrl thus are often much inore r o m p l i r a l ~ dthan those involved In individual human planning. T h e difficiillies are i i i a i n l y d w to t,hp following factors: 1. The designer's intention may not a l w a y s h r clear to the manufart,riring I pnginrrr who will act on that intent.ion. Th e respertive langiiages I
in their profession, the ways in whirh t.hry exprrss t,heir intentions, t.heir rritical concerns, and their perspectives may all diffpr 2 . Aut,omation of prorrsa planning reqiiires th a t part features r an be automatically extracted from the prodiict model without hiiman interaction. b u t existing interfares of (:All nystFms do not sufficiently consider this requirement of automated process planning Needed informat~ionmay he inaccessible or in an inappropriate forin Engineering drawings are t , h r rnpdiiirn currently used by designers to communicate with nianiifactiiritig engineers, but t,hese drawings may sometimes contain insiiffirient d a t a . or t h e d a t a may be hidden in fnrrns which cannot be directly ~ised.or rxtraneoiis d a t a may be irirluded which ohscnrps the relex'ant informat.ion.
3 . T h e designer generating the drawiiigs is
T h e aniniint of t i m r ht-twrm tli? planning generat,ion p h we (at, the (design department) and the planning execution phase (on the shop floor) is normally much longer than th a t involved in individual daily planning. Due t o the dynaniir nature of a production environment, it is very likely that by the time a design is ready to be manufactured the constraints that were used in generating this plan have already changed grurtly, and thus t h a t plan has berome less-optimal or even totally invalid.
s, The
generation exer,,tion of a rornplete prodllction plan normally involve many different organization units, and may often span a long period of t i m P and diffrrpnt gPogrilphir loril,inna Thesp conditions m a k e t,hr plan-monitoring prur.ess, c r i t , i r a l fr,r plan ilnprovement,v e r y difficiilti~ not
~
The IJltimate Goals of CAPP R.csearrh
Pr or w s planning is a vrr) I>roatls i i l ) j r r t \+hirli i n i n l v p s a w i r h rariPty of activities w i t h i n a prrduction environment ( : r r i i s ~ ~ ~ ~ i i ~ ~ 11 r i t has . l y . mul1,iple goals that, must lit, addwsswi s i m u l t a n r m i s l y 111 r ~ w n r c h A l l these goals, hoth Fhort,-tprin and long-term, shoiild rrlatr to ttie kr)- ronrept and fiiiidamental need r i f .syrtein ,ntegration I n t h e short-term, cornpiiter-based planning sysbems must hv dewl nprd t o ailtonrate tndivirlrial planning f u n c t i o n s and t,o integrate those functions into a unified environment In nthrr words. rompotrr-hased planning systems shoirld, at least. s r r w two rrlarcd goals: funrtinn arironiation and .iyFtem integration ( o r cnordinatmn) T he integration m i - t n r r ~ i rnot nnly among varioiis planning fiinrtions. hiit alsn arross design and mariufact~rririg activities Here. w e must point mlt. an n f t m orwlookrd a.;prrt, of ihnrt-trrni planning research. which i s to a i i t o r n a t ~planning fiinrt,iotin .wlPrtiwlp r a t h w than uniformly across t he uhole sprctriirn of activitirs. This is a matter of rnnrern hecaiise. based on rrirrent practire, there ar r still many p l a n n i n g liinr tims that are impossible to autoniafe given t,lie current s t a t ? of the art of computer terhnology. Many resoiirces anrl niiirh pffort u-ill he- wasted tr)in[; t o automate such fiinctions unless t.lieir feasil,rlity is first st,udied Tlierefore. a rrit.ical first step is to exaniiiie the curr m t p l a n n i n g functions with rrspert to t heir atitrm~atahility.focussing elforts on t,hose porbions anienahle to compiitrr aiitoina1,ion: and leaving those t hat are hrst, performed by homan brings to hliiiian planners Another related concern is the degree of autonomy of rornpiitc+hasrd planning systems. It is iinrPalistic t o rxpect, t h a t all romputrr-Iiasrd planning systems will he fiilly autonomous. Cnmprehmsive stiidies s l i i n i l d he rondiicted to examine present planning functions i n order to indicate which function? are autoinatable and to what degree. .\ realistic picturr of the romp~:tcr-hasrd planning environineiit P h o i i l d consist of planning fiinctinns thnt are r a m p l ~ t e l yperinrnied hy cnmputers. t h a t are parI ially assisthl hy ronipiiters. and I h a t are inaniially r o n d u r t d hy hiiinaii beings H u t st,i!l. the integration part of t h r Pffort should hp explicitly supported in these partially automated systems t o intdliRenlly coordinate these three categories of planning. All thrse short-tprm efforts in planning research shoiild serve to support t he iulliinate. long-term goal of this venture. which is t o do away with the need for planning This may sound ironic a t first, but it is a real challenge t,hat mugt be addressed. As mentioned above, one of thP main goals of planning is coordination and integration; we shoiild not only coordinate the current planning functions across design and manufacturing, but also suggest ways that will make t hew funrtions easier t o coordiiiatr A s we gain more siircess in the second aspect (IF, siiggrst npw plnnning strurtures that are naturally coordinated), t,he need fclr coordination. and hence the nerd for planning, will diminish gradually. A preliminary example of this t r m d can be sw n in research elforts in the areas of design for manufartiiral~ility.a nirjor corirrrii of manufartailring indiistry today 'rhrse niaiiufarttirability concerns have tn br checked inaniially to insure that parts d~aigneriran t i i d P i 4 he riianufart tired rost-rffertivrly With romputers
593
t c process planning
Ciirrrnt Emphasis
4.1.4
Flerlblo M
a A
/
Figure 2. Increase of Planning Complexity by New Manufacturing 1141 helping t o bring these maiiiifacturing ronsiderations into the design phase, the coordination betwren design and manufacturing is partially built-in, and thus the nerd for this part of the planning function is partially eliminated. In this respect, one of the first challenges we will meet is the definition of features (geometric shapes) in a way t h a t they can serve both the design and the manufacturing functions. A problem is t h a t in many cases both functions cannot be covered by one definition. The design of a global relational d a t a structure as a framework for identification of geometric shapes in relation to design and manufacturing funrt,ion.s is therefore of extreme importance. As we progress in this direct,ion. ultimately integration and coordinaticc will be addressed as parts are being des i g n d The need for having a separate activity, called manufacturing planning, will he rrduced. It, is, thrrrfore, fair to say that a truly intelligent planning system will d o planning implicitly. and thus greatly reduce the need to perform planning functions expliritly
4 4.1 4.1.1
Critique of Present Approaches The E v o l n t i o n of C A P P Approarhes Variant Approach
Sonic of the earliest work in applying computers to aid the process planning task has been in the area of variant process planning syst,ems In this type of C A P P system, parts are grouped into part families, a unique code is generated for each family, and a standard process plan is developed beforehand for each family. Most, systems use a well-defined Croup Technology ( G T ) based coding system t r j develop the unique codes for t h r various part fanlilies The standard plans :are s t o r d in a computer. coiweniently keyed under the unique code generated for each family. This type of process planning systFm is used by first developing t,he code for a new part t o show which part family it belongs to. and then retrieving and filling up the standard process plan to reflect the characteristics of the new part. CAM-I C A P P [15! IS one or the well known examples of such computeraided process p l a n i n g systems
4.1.2
The next, major development in generative process planning systems is the use of A1 t,erhniques to model thP activities pertaining to the manufacturing logic rompoiirnt of these systems Concepts from knowlpdgr-bawd systems and AI-based planning techniques are found to be particularly suitable to generating process plans G A H I 1231, TOM [24i, XCUT j251 and XPLAN j261 are examples of the systems incorporating some Al techniques t o develop generative process planning systems There exist, many potentials for applying these techniques t o different aspects of the proress planninp, prohlwn. such as process sequence determindtion, fixture selection. cutt.ing tool selrrtion. cutting parameters selection, etc. Recently, work in the area of C A P P research has focused more on the issue of int.egration of design. planning and manufacturing tasks in a manufacturing organization T h r desrription of the geometrical and topological features of a part has h e n considered to be a central link bet.ween the tasks a t all these levels. With this in mind. work has concentrated on developing part description schemes that can support tasks a t all levels. Part description schemes under consideration range from geometric mndelling schemes, such as CSG or R-rep 127,281, t o feature-based schemes 129,30.311, and even part-description schemes combining more than one representation methodology 1321. Currently, the variant and semi-generativr approaches seem to be the most practical ones, and these syatrnis are readily amenable to certain real-world applications. However, generative planning systems will best serve the long-term needs of thr process planning arena, and current research has focussed on detrrmining the best way to combin? artificial intelligence techniques with vario w part-description schemes to develop a t,rvily generative process planning system. To support the devdopinmt of w c h generative systems, various conceptual frameworks for Lhe task h a w been prnposed i33,34,8] which elaborate the relevant research issues that need to br addressed before these systems can be fully rralised For rxaniple, a g r o u p t,erhnology technique as a primary empirical approach could be rnhanred by introdiicing theory for geometrical and manufacturing pattern recognibion as well as a basis lor manufacturing logic. By starting with group of parts, it is possible to generate a coniplete set of parts which belong to the same group S A P T system is developed on this concept [35,36,37] An important t.ask in process planning IS the generation of the process plan for asmnhly. For bhis task, there exist so far only a few systems introduced in prnrticr. The concept of a C A P P system for assembly is shown in Figure 4 . The influencing variahles in the planning of assembly systems, e.g. proliferation of produrt typea and models, shorter development time of products, the changing needs of thr production personnel, are changing and create increasing pressure for the productions Planner in deadlines and costs. Because the technical aspect and rrquircment of t h r assembly process h a w t,n influence the product design, trrhniral prohlcms of aasernhly h a w to he integrated in the computer aided design proress as well as CAPP. Also, evaluation procedurrs have to be applied to the preliminary design t o optimize it in terms of not onlv manufacture but also assembly. 'rhesr hasir procedures consist of Flcxihilit,y Analysis
.
.
Functional Analysis Feeding Analysis
Gripping Analysiq Insertion Analysis
Semi-Generative Approach
The semi-generative systems were the next generation of process planning systems. These systems are advanced variant systems, incorporating quasi-generative features. After the part family has been identified, as in a basic variant system, thrse systems offer the user srveral options. One opbion is t o make suitable changes t o the standard process plan for each part family. The second option is to begin with an incomplete process plan and complete it for a specific part. A third option is to start from the beginning and completely create a new plan by using various standard process descriptions stored in the computer. Preliminary versions of GENPLAN [161 and CORE>-CAPP 1171 are some examples of such systems 4.1.3
Generative Approach
The generative approach marks a notable change in the evolution of compoteraided process planning systems. These systems are designed to automatically synthesize proress information to develop the process plan for a part. These systems contain the logic to use manufacturing d a t a bases and a suitable part description scheme t o generate the process plan for a particular part. Early versions of generative process planning systems used decision tables and decision trees to capture t,he manufacturing logic, and G T code or specially developed languages to provide a precise description of the part. These systems, being more complex than Lheir variant counterpart, were also more restrictive in the breadth of their application UCLASS IlB], APPAS [19] and C P P P [20] are some of the systems of this kind. With the advent of C A D databases and wire-frame modrlling, grnerative process planning s y s t e m are being developed using this type of part description scheme. TlPPS 1211 and RPO 1221 are some of the early aystpms Lo explore the link brtwem CAD databases and bhe generative approach
594
I
I
1
I
Figure 3: Use of C A P P System for Assembly 1141 Manufacturing Analysis T h e tasks of ;Lssembly planning range from the selrction of suitable system principles, the development of the preliminary concept, the development of layout variants, the deterniination of performance d a t a parameters such as costs and output t o t,he simulation of system behavior and the evaluation of system alter-
natives [38,39,40] 4.2 4.2.1
Some Critiques to Current Rpwarch L a c k of F u n d a m e n t a l S t r u c t u r e in The P l a n n i n g D o m a i n
Planning in a manufacturing setting has historically continuously been viewed as a hierarchically structured activity. This hierarchical structure may vary in detail from organization to organization; but certain generic characteristics exist which can serve as a foundation for developing some general planning strategies To date, there have been only very few research efforta [B] t h a t have focused nn defining overall logical structures for planning wilh respect to the organizational structure of a production site. These efforts naturally involve many coitsiderations t h a t g o beyond the pure technical donlaln: but without the existence of such a clear planning structure, the systems developed will not have a major impact on manufacturing organization in practice. Furthermore, without a clear structure, i t is difficult to develop planning systems that are modular enough t o incorporate local constraints, T h e ability to include local information effectively is a key t o the robustness of planning systems in real-life applications, and is why some systems developed to d a t e can not go beyond the stage of the academic prototype. 4.2.2
The O p e n - l o o p Planning A r c h i t e c t u r e
T h e complexity of manufacturing activities makes it impossible to generate an ideal plan the first time. I n practice, it usually takes many iterations based on feedback from the plan execution stage t o gradually refine a plan into its optimal f o r m A planner, whether a human being or a computer system, should possess this kind of self-adaptive capability t o rcach intelligent decisions. Most current planning activities are mainly concerned with the plan generation phase; no major effort has been made t o link plan generation with its execution and monitoring activities 1411. This opm-loop planning approach falls t o automatically incorporate the valuable lessons which might he learned from execution feedback. Human beings could serve to monitor plan execution and hopefully t o improve the performance of futiire plans But since the execution of a manufacturing plan normally spans a period of time with the lnvolvement of multiple departments, this manual monitoring and feedback task is difficult t o accomplish effectively in practice. The lack of this self-improving ability is cne of the reasons why most C A P P systems developed in a laboratory environment fail t o make an impact on planning practice. 4.2.3
Isolation A m o n g D i f f e r e n t P l a n n i n g S y a t e m s
T h e importance of manufacturing planning is largely rooted in the urgent need to integrate design and manufacturing activities. Planning is considered to be logical means t o achieve factory integration by creating a bridge that connects the islands of design and manufacturing in present production environments. Ironically, most current planning systems are themselves not integrated, even though they were motivated by the need for integration in the first place. Current research efforts tend t o focus on a specific kind of planning function for a specific type of part. These micro-viewpoint planning approaches produce computer systems that perform individual tasks in isolation from other planning activities. They result in many snail, and still isolated, islands between the islands of design and manufacturing. Only a few efforts are aimed a t the macro-viewpoint (or system-viewpoint) approach which addresses overall planning, including production planning, process planning, and operation planning, from a n integrated system point of view. Because manufacturing planning ie a system problem, planning systems based on isolated micro-approaches will unlikely result in any real impact in practice Approaching planning problems from a system point of view is much more difficult, but greater efforts in this direction are needed because lhe present isolated approaches will not solve the need of manufacturing planning. A new planning approach is needed in the future to meet the goal of integration. First, production planning, process planning, and operation planning must be addressed together in an integrated fashion, rather than by continuing to build separate systems t o automate them individually. In other words, decisions at the production planning level must be supported by those from the process planning level which, in turn, should be supported by detailed specifications made a t the operation planning level. Furthermore, a t its higher level, production planning must be tied together with long-term strategic factors to arrive a t decisions which are optimal from the engineering, economic, and social points of view. At its lower level, decisions made a t the operation-planning level should be based on physical models of the process being planned. The phenomenological studies, process modeling, and mechanlstic simulation techniques developed thus far should be linked together with various planning functions to support planning decisions in a more realistic fashion. All these critical links must be addressed when planning systems are being developed rather than after they are built 4.2.4
N a r r o w - F o c u s e d Planning Systems
As a result of these microviewpoints planning activities and the lack of an overall planning structure, most planning systems developed to date focus on a narrow range of activities which severely limits their applicabii!ty in practice. For example, most efforts in current generative C A P P systems ere aimed a t generating machining sequences of certain limited part shapes Since they are based o n machining handbooks and databases, decisions made a t the operation-planning level are treated as a black-box, without detailing how each sequeiice will be carried out in actual operation. Another example of this narrow focus is that few planning systems are being developed for manufacturing operations other than machining. Other areas such as bulk and sheet materials forming.operations, elcctronir packaging, etc , are all rhallrnging domains for pianning syslems, but have received only minor attention so f a r Even in t,he machining domain, most
systems developed can only address Iini-operational planning tasks and are unaiiitnble for mulliplr proressrs Some jobs could be processed simultaneously whirh. together with the future introduction of multi-processor terhnology. may result in significant time savings. Furthermore. economic considerations. an iniportant factor in real-life planning activities, are not incorporated in current planning systems which ronsist mainly of technical knowledge 1421 4.2.5
The U n i - a p p r o a c h to I n t e r - d i s c i p l i n a r y Planning Problems
Due to the complexity and breadth nf manufac1,iiring planning probl?ms and t,he dynamic nat.iire of bhe modern nianiifart.oring environment, a competent pianner needs t o have a variety of capabilities and knowledge t o accomplish hisj'her tasks. Similarly, a robust planning system should possess divrrae techniques to address different planning functions effertively. Ilnfortunately, with only a few exceptions 1431, the majority of current planning systems are based mainly on a single approach to the problem There is no single hest approarh whirh can handle all bhe functions required in an integrated planning system Uybrid approaches, which logically combine the strengths from several different, techniques, should be devised and used for practical planning problcnts For example, various techniques from optimization and operation research should he incorporated into plan sequence generation, mathematical models of processes should be part of the operation planning systems, and AI-based techniques should be intrgrated with proper traditional methods to enhance their efficiency. Similarly: researchers in process planning should have an Inter-diariplinary view of their a r m s of interest, and be willing to exchange and learn valuable lessons frorn other domains. A good example of this can be found i i i those current research efforts to find ways of extracting interesting features from CAD databases. Feat,ure extraction has been a suhject of study for years in the area of computer vision, and researrhers from both disciplines should consult with each other about their common and respective goals.
Suggestions for Future Directions
5
A Framework for Integrated Process Planning
5.1
It is only logical to expect t h a t manufacturing planning, which is aimed a t integrating design and manufacturing, should be itself integrated. Intelligent and integrat.ed planning will be the paradigm for csmputer-integraced manufacturing in the future, and a critical driving force in the evolution of industry from the current information-intensive stage to a knowledge-intensive stage in the future. Whrn discussing integrated design and manufacturing or integrated planning, it i s important t o understand the true meaning and implications of the tern1 "integration". Currently, there exists some confusion between integration and interfacing. There are many proposed approaches to integrating design with manufacturing or various planning functions, and vice versa. But careful examination of these approaches show that they are mainly trying to interface various separated activities a t the design, manufacturing, and planning phases. Since these activities are currently approached separately in research and practice, these interfacing efforts are indeed needed to tie together their isolated results. lntcrfacing should, however, he differentiated fram integration which, while much more difficult to achieve, could result in a much greater increase in engineering competitiveness. One difference hetween interfacing and int.egration is t h a t interfacing can he achieved nt the result level while intpgration must be addressed a t the task levet. In other words, i t would he too late t o integrat.e a task when its sub-results (such as design and manufacturing specifications or decisions for process and operation planning) are already decided separately. To achieve truly integrated deaign and manufacturing, the interactions between them should be addressed a t a much earlier stage than that of our current focus. If these activities can be integrated a t the task level, their differences will diminish gradually as we approach the result stage and their results will be naturally integrated. Similarly, various interrelated subtasks in manufacturing planning must be integrated at the task level, rather than be interfaced at the result level after they are already carried out separately. The reasons for the integration of process planning are (61:
-
Improved efficiency in the information
-
Row
Improved quality of the process planning Reduction of human errors Functional integration of process planning and scheduling, enabling a quick search for alternative solutions for optimization in the use of equipment and production control Flexible use of the different functions
6
Also, the implementation of the integrated system must be based on: A uniform product description based on proper features
-
The use of difierant modcles for difterent functions
.
The use of a uniform data base interface for every module
rn
The use of a uniform user interface for every module
The possibility of facilitating user interaction a t the request of the operator
Proress planning systems, in order t,o provide the integration, automation and flrxibilit,y needed in futiire process planning, should:
.
be generativp
595
. .
be technology based use features as a technological and cominunicational interface between design and process planning
. .
be able to automatically extract all product d a t a use a supervisory control system t o ensure user-friendliness and flexibility in use integrally support all planning tasks, including capacity planning and scheduling
take decisions based on optimization techniques (also in the edge base systems)
case
of knowl-
. .
be fit for closeloop planning.
A framework for integrated planning should have a multi-dimensional perspective. In other words, integration is needed in the following areas'
Planning knowledge: Sclence-based principles must be integrated with experienced-based knowledge. Physics must be integrated w t h heuristics.
.
Planning activities: Process planning must be integrated downwards with operation planning and upwards with production planning. Operation planning must be integrated with physics (or models) of the manufacturing process being planned Planning techniques: Techniques such as group technology, modeling and simulations, optimizetion, and knowledge-based approaches must all be integrated into a truly robust planning system. The approach of building separate planning systems based on a single technique (eg. rule-based systems) and then interfacing them with other systems which were built upon different techniques is not sufficient for integrated planning. Planning constraints: Various planning constraints, local or global, techniral or non-technical, user-provided or expert-advised, should he integrated during the planning stage, rather than being added t o the system afterwards. This requires viewing manufacturing planning as a cooperative problem solving activity which incorporates various concerns simultaneously.
Planning feedback: Mechanisms must be provided to automatically incorporate feedback from the planning results t o improve future planning decisions. This closed-loop planning is fundamentally different from the current, open-loop planning architecture. It requires the extension of present planning activities t o the plan monitoring and execution stages. Figure 4 shows a proposed concepbiial structure for such a n integrated planning environment. This structure can be better understood by compariiig it with current planning activities. Decisions on operation planning, process planning, and production planning are presently reached in isolation. At the operation planning stage, the use of machining handbooks and databases is still the most cvinmon way of deciding operation parameters. At the procesa planning stage, the main concern is to decide on operation sequences without information about operation parameters and production schedules. Similarly, details of process and operation planning are treated as a black-box when decisions for production planning are being made Human planners take the responsibility of interfacing and coordinating these separated planning decisions after they are generated. This hierarchical structure suggests t h a t decisions for production should be based on those of process planning which, in term, are rooted upon those of operation planning. T h e three main factors of any manufacturing operations, namely physics of processes for operation planning, geometry of products for process planning, and factory resources for production planning, are all dealt with in an integrated fashion. T h e following two sections describe two critical integration efforts: one is between C A D and CAPP, and the other is between J o b Shop Control (JSC) and CAPP.
5.2
Integrating CAD and CAPP: Modeling the Product and the Process
Today, many software systems are in use to support the functions of different departments within a company. Several problems must be solved before these systems are really integrated. In general, connection oi computer systems leads to four basic software problems which indicate the need for a common factory database: the combinatorIal problem (the number of interfaces explodes with the increasing number of systems) the problem of redundancy and inconsistency (multiple storage of data; different update states) the problem of closed software packages (no accew t o d a t a structures; algorithms are not available) the model problem (different models in different software systems)
596
F~~~~~ 4. A Conceptual Structure for Integrated Manufacturing Planning Let us consider a C A D and a CAM subsystem One can see that a specific d a t a nlodel can not be established in one system if information is missing in the otner system. If all drawing information is available in the CAU system, this system then covers all necessary inforniation about the workpiece which we need in the CAM system. Although there is a n optimal affiliation of surface parameters t o geometric d a t a of surfaces, the relation is missing. It is impossible t o transfer the surface parameters from CAD to CAM because, in the C A M system, storage for these parameters is related t o the surfaces of the workpiece. Dicerent d a t a models can not be converted into each other without human input being necessary. This problem can only be solved by the accurate planning of model structures. If the structures of system-overlapping models are congruent, i t is not necessary to store them in diffprent computer systems. This leads to the need for a common product model for the whole company. Seen from the perspective of manufarturing, the product model should cover the geometric and topological information of the workpiece. In a solid model, the shape of a part is usually rrpresented by lines, faces, and volume primitives. Dimension and positional tolerances should be modeled as attributes of relations between geometric elemenh, and form and surface tolerances as attributes of geometric elements Recent concepts of technological product modeling cover these topics. Designers must be able t o define the relations of functional adjacent faces of a workpiece. This can be implemented in CAD systems if internal d a t a structures and input techniques qualify for these purposes However, for design within sectional views and interference checking, it is necessary t o define the relations of functional adjacent faces of different workpieces (fits) (441. For manufacturing, assembly, quality control, and other purposes, it is important t o represent the organizational product topology in computer-aided systems From this follows the deconiposition of the product into its structural components. The organizational topology supports the disposal of parts, the pooling of batches with similar machining technology, or the joint manufacturing of assembly parts. No specific capabilities t o represent functional or manufacturing elements (form features) are available yet. Two existing systems can deal with form features: one by bounding a region on the part, and one by selecting a set of faces on the part to be a feature. This has t o be done after a specific design task is complete. It is not possible t o generate the model by specifying features. No system is able t o retain tolerance data as anything other than textual information. It is obvious that commercially available systems are still geometry processors and d o not sufficiently support the emerging needs of users [45j Especially for manufacturing applications, solid modeling systems do not provide enough facilities for the user. Some systems can demonstrate the ability to generate N C tool paths in 2 and 1/2-D mode. Problems arise if the geometric elenients of a contour line are mixed, contour lines are not closed, or ditferences in scale exist. No system is known which is able to automatically generate NC tool paths for complex surfaces. Besides, companies would still have to buy new post-processors, as existent ones are usually not compatible. Sheet metal nesting is supported automatically or interactively by a few systems, and bills of materials for assemblies can only be generated in one system 1461. T h e complete modeling of a product, including all necessary information for manufacturing, is a basic requirement for the integration of C A D and CAPP. This means t h a t C A D must provide an element-oriented user interface which allows the input of technical elements such as holes, pockets, chambers and grooves. Technical elements can be used by a designer in describing the part, or by a process phriner in planning the operations t o manufacture the part. T h e user interface slioiild consist of input macros which define such technical elements as the relatim of lines, face, or volume primitives. They should access standards as well as d a t a for tools, fixtures, and machines. For example, if the designer wants t o place a groove in a shaft, he should be able t o choose a specific macro for t h a t groove. He should be able t o specify the standard sheet to refer to and the face of the shaft where he wants t o position
the element. Depending on the diameter and length of the cylindrical face, all par;tni~tersof the groove should be automatically determined and shown on the screen by graphir and textual output The designer should be able to change every single parameter of the groow i 4 i i T h e abovr mentioned tasks of design and process planning indicate the direction in which computer aids has to go in the future:
whirh ronlaiiis all rekaved shop nrders The job-shnp contrnl system manages job-shop oprratioiis. It accesser tlrr y h o p orders database and updates the jobshnp statiis database There is nrrasional feedback from the joh-shop control syste-rri t,o the P P S system Systrms a w gPnPrally implemented in procediiral programming languages This architc-rturr h a y Some severe flaws. h n w w w
.
.]oh-shop stat,iis is not rnnsiderpd diiring process plan generalinn. Thus plans are generated which cannot be processed given the actual State of the job shop.
open, extendable solid modeling systems
.
.
a common product model for all applicat,ions a common factory d a t a base for all computer systems within a company
Nearly all ronipany parameters ran become important for the planning and execution of manufacturing processrs Ever) workpiPce parameter should be available on the computer. Due t o its object-orientation it should be part of the product model. This leads to the demand for arbitrary extendibility o f t h e model Product models should he manipiilatable with commercially a\-ailahle program modules as well as with user written programs. Algorithms for solid model manipulation milst come as well-documented subroutine programs Softwarevendors must not be reluctant about disclosing the internal d a t a structures of their products [48.49] 5.3
Integrat,ing Job Shop Control ( J S C ) and CAPP: Modelling the Production and the Process [ S O ]
Due t o its complexity, process planning is often carried nut without consideratinn o f j o b shop status information, such as resource availability. Resides, the t,hroughput of orders in a j o b shop often siiffers from disrupt,ions caused by sthchwtic bottlenecks, non-availability of tools, or breakdown of equipment. Replanning has to be done by improvisation and can result in long through-put times. A large number of process plans perhaps cannot he executed and have t o he altered. The process plan containing a linear sequence of operations is not flexible enough. A process plan that. provides several possiblr ways t o manufacture a given workpiere would be of great help. This shows the necessity of breaking away from the process plan as a st,atic and linear operat,ion sequence. Plan rigidity has to br overcome by enriching the plan representation Assuming that process plans for mechanical parts can be automatically derived from design d a t a bases, traditional process planning seems to be too highly abstract. Too meny constraints referring to the workpiece and the j o b shop are ignored. A new process plan representation has t o be developed which covers the causal structure of the manufacturing process. Instead of pruning all alternative sequences of manufacturing operations considered during planning except the "opt,imal" one, it is proposed t o raise the pruning process t o a higher level T h e plan representation should he able to express parallelism and alternative operations. It should support decision making on the shop floor hy providing dependency information referring to the job shop st,atus. Job shop control can benefit from such "intelligent" plans, which can he adapted t o the actual resource constellation. Noii-linear, directed graphs are suitable to hold workpiere states as preconditions and sperific manufacturing operations as events Ttir similarity t o rulebased knowledge reprepentation enables the methodiral integration into artificial intelligence systems A process plan can for instance be represented by a Petri Net which providps the power t o model logical and t,emporal rPlat,ionships between manufacturing operations. Petri Nets easily represent, concurrency and can contain loops; therefore, the modeling of interaction is possible. We can represent the actual state of the j o b shop as part of the planning knowledge represeiit,ation by Petri Nets, but since the number of objects relevant to the j o b s h o p state is huge, a hierarchy of Petri Nets must he created which, takeii Logether, will provide enough modeling power. Different objects (e g resources such m machines and tools) in the job shop can be modeled in separate Petri Nets. There must he some commiinicat,ion facility for the Petri Nets which make u p the hierarchy. This can be achieved via token passing through global elements of the hierarchy Since it is possible to represent job-shop status and process plans (which are part. of t.he current set of shop orders) by Petri Nets. both representations can he merged lor job-shop control purposes Assuming that a valid job-shop schedule exists, control of operations in the job-shop according t o schedule is possible utilizing the task-bidding principle The basic idea is t h a t , in order for a machining operation t,o take plare. t,he required machine and tools must be available and in the proper status This t,ranslat.es directly into the requirement t h a t the corresponding Petri Nets modehng the resources must have a distinct token distribution. Simple task dispatchers hased on this principle can be built t o dispatch shop orders, tool preparation orders, transport orders, etc. Job-shop control algorithms and strategies should be developed to prove t h a t shop control benefits from non-linear, directed graph process plans. T h e requirements of new algorithms for controi purposes have t o he defined Such algorithms have to he develnped and tested so that non-linear process plans can optimally be utilized Strategies, or job scheduling and order dispatching, have to be evaluated. ~ is fuzzy. PlanThe borderline hetwem planning and control ( i .scheduling) ning and control depend on each other and must ultimately use the sanie d a t a . However, today's systems for production planning, scheduling, and job-shop control d o not take these dependencies into acrount In a factory environment, a CCP system typically depending on heavy interaction with a process planner is utilized for process plan generation. T h e order-independent process plan is stored in a database of active process plans. T h e process plan itself consists of a linear sequence of operations Once an order is initialized by the PPS (Production Planning and Scheduling System), the corresponding process plan is retrieved from the database, associated with the order d a t a and, thus, a shop order is generated. At t,he schediiled releasr time. the shop order is stored in the datahme
CThen a process plan is retriewd from the database of active plans, it is often months. if not, yenrs. old and therefore possibly o i l d a t e d Sincc there is no support for re-planning, the job-shop control system caiinot adequatrly react to di.irupt,ions. In the most primitive case, such a reartion could be the retrieval of n contingency process plan from the process plan dat,abase and its iitilization inst,ead of the original plan.
.
Knowledge, which plays an important role in the realm of production planning, is very dificult to represent I[ procedural programming languages are employed.
Effectively, there is a complexity barrier which hinders the automation of more complex Lmks like generation of operation sequences or schedules. An advanced system which integrates process planning, scheduling, and the job-shop control system should manage all functions Such a system should h p knowledge-based and possibly implemented utilizing A1 software technology This niakes the integration of knowledge and, therefore, the higher automation of functions like process-plan generation c a w r . An integrated system is monolit,hic oidy from a logical point of view, its implementation may very well have a dist,rihuted natuw It could work like this: nnre an order IS received, the correspondlng process plan is automatically generated. t,aking into account the act,rial joh-shop status. For this, we propose the term just-in-lime process planning becaiise the plan is generated on a when-needed basis The resulting shop order is stored in the datahaw of released shop orders. Job shop control is exerted hy the intrgrat.ed systrm already m ~ n i i o n e d Replanning In case of diarllptinns is farilit,at,rtl hecause process plans can be altered with respect t o the changed Jnh-shop status. Such an architrrt,iire can b r realized if several precondit,ions arp fullilled The int,egratpd knowledge-based system i s powerful enough to allow aot,omated grnerabion of process plans Inrreasrd computing power is available, There will still ?xist, a database ronthining all active process plans. This is for pragmatir reasons Process plans are needed not nnly for scheduling arid job-shop control tasks, biit also for a variety of other tasks. For example, i n the area of invrstnient planning, the t,ask of marhinp procurenwnt r q u i r e s produrtion times, set-up time*, and desrriptions of processes per operaLion from a large number of proress plans. Typically, t,hesr tasks involve many hundreds or thousands of active process plans from the databas? Of rourse, if automatic proress plan genrrat,ion is available as assumed here, I t would also be possible to generate all necessary process plans o n a whcn-ne&d basis, but this would consume excessive computing power So. t h r strict correctness of all process plans in the database with respect. to act,ual job-shop staLus is sacrificed here for the sake of prudent iise of computrr resources If, in relation t o today's computing resources, unlimited resources are available, this restriction is no longer necessary. Ilere, the database of act,ual process plans is substituted for a temporal database whirh is generated following the j u s t - i n - t i m e philosophy. This ensures the strict accuracy of all process plans with respect to actual job-shop status. Scheduling and replanning is further facilitated by the integration of the shop-order database and the job-shop status database into a unified factory modrl i51/. We believe that futiire systems for pror.e.ss planning, scheduling, and Job shnp rontrol will evolve along the paths outlined above if batch size one and just-in-tirne concepts are to be realized in ~iianufacturing.If these concepts are employed, there will no longer he a rhancp to dproiiple process planning, scheduling and j o b shop rontrol i n time by material buffers. The strong interconnertion of all factory subsystems. which i s needed for the realization of this approach, is regarded as essential in this context [ 5 l ] .
6
The Potential Role of A1 Techniques in Integrated Planning
Integrated planning is knowledge-int,ensive in nature. Traditional compiiterbased methods are unable to deal with the challenges of integrated planning because they are good a t processing d a t a for information-intensiv domains, but not well suited for automatic inferencing for knowledge-intensive tasks. Rather than simply processing information and data. Al-based techniques are designed for capturing, representing, organizing. and utilizing knowledge on computers, and hence will be the key technology for intelligent and integrated planning in the future A1 will also play a central role in the evolution of current semigenerative planning approarhes t o full-generative systems in the future. This is 4niilar t o th? evolution from manual planning t,o variant planning due t o the introduction of group technology. and from variant planning t o semi-generative planning due to thP impact of CAD representations. These key technologies and t,he assoriated evoliition of romput,f~-nidrdplanning approaches are summarized in t,hr follnwing table
597
I
..
/I
Evolution
',
Character of Industry -.
__.p..t--L ---- Labor Intensly
Past 1 '~ / Eouioment ~ Intensive _
~~~~
,,
-
Key Technology
Type of Planning htanual Planning ~~
-
C- ! o m w t ? r - aGroup i ~ ~Technology ~ p ~ Retrieval Planning
CAD Representation Present
1 Informstion IntenwvP
Smu-generative Automatic Planning
Intelligent planning
1
I,
fivrn though t,heir potential role in CAPI' euolntion has been envisioned, Al-based techniques have not yet reached the st.age where they can demonstrate their value in actual practice Moat of the current activities in Al-based planning approaches are focused on building knowledge bases to capture manufacturing logics. This branch of A l , knowledge-based expert systems, has received the niost attention and stirred controversial debates in the engineering conimunity. and aniong researchers in the planning area. hlany prototype expert systems have been developrd for planning functions, but none of t h r m have made as significant an impact as has been promised. Therefore, in discussing the true potential of A1 in CAPP, it is important to examin? current limitations before making future predictions. In the following two srrtions, wr will first explain why current Albased approaches can deliver only limited success in the planning domain. We will bhen suggest the potential of A1 techniques, beyond knowledge-based expert systems, in integrated and intelligent planning, if such techniques can mature successfully in the future through joint basic research between the engineering and computer science communities
6.1
Current Lirnitat,ions
To date, most AI-based approaches used in C A P P are some variation of knowledge based expert systems. These expert systems in the planning domain are being developed based mainly upon the wisdom and lessons learned from the medical domain Manufacturing planning is, howevrr, very different from medicine, and consequently the traditional wisdom is either inapplicable or insufficient for engineering tasks. This is a serious problem for the engineers who are eager to adopt this new technology. Before applird use is possible, many fundamental at.ildiea are needed to develop new wisdom for the engineering applications of knowledge-based expert. systems. Based on the current state of the art of the technology and due t o the lack of knowledgespecifically applicable t o the domain, building knowledge-based expert systems for engineering tasks is still a research effort with high cost and risk, no6 a routine development. Naturally, engineers who understand t.he engineering domain best must. be involved jointly in these research efforts with coniputrr scientists. The following is a list of difficulties engrndrred by expert system applications to the engineering domain, and thus it is also a list of research needs in this area. Other diffirulties, such as knowledge representation and uncertainty reanoning. which are of generic concern to AI problem-solving or knowledge-based expert system technology, are not included here Readers who are interested in the general limitations of current expert system technology should refer to other more AI-oriented articles
6.1.1
T h e Knowledge A c q u i s i t i o n M e t h o d .
The process of getting domain knowledge from human experts and programming it into computers is called knowledge engineering Current practices in knowledge engineering rely heavily on pprsonal interviews with domain experts. This knowledge acquisition method, which wolved from earlier expert system projects, is very time consuming and difficult t o manage. Nevertheless, it has been proven t o be useful for those projects. and is s l i l l the most popular method in developing expert systems for various domains including engineering. One of the reasons why these interview techniques work so well in the medical domain is t h a t human experts in t h a t domain (doctors. physicians, well-trained scientists, & . ) a r e normally articulate and able tn express themselves well They usually can explain their reasoning steps and the knowledge they use to engineers during interviews without great difficulty. Unlike physicians in the medical domain, human experts available in the engineering domain can be very diffrrent. For example, LO develop an expert system for machining processes, the best candidates for domain experts are those who artually work with the machines on the factory floor, but they, however, normally d o not receive formal scientific training in their profession as doctors do. Most of llieir knowledge is accumulated through years of hands-on experience, and is often biased toward their own heuristic, and may not be verifiable by others. They may be less articulate than are doctors, and may have difficulty explaining why and how they make a particular decision. Thus the traditional expert-system wisdom of conducting interviews t o acquire knowledge may not be as successful when dealing with this kind of domain expert. Engineer's knowledge and logic for problem solving, such w process planning, are apart from being empirical also a deep cognitive process. It seems t h a t it is possible t o develop paradigm for technological features pattern recognition and manufacturing logics T h e knowledge t h a t is acquired depends upon knowledge representation. The formal grammars which are represented as regular exprvssions and finite-state autoniat,a could be new tool for knowledge acquisition and even for generating new knowledge.
6.1.2
The T y p e of D o m a i n K n o w l e d g e .
Due t o t h r interview terhniques used in t,he knowledge-acquisition process, the domain knowledge gathered is largely of the heuristic type. Heuristiw are im-
598
portant to intelligent problern-solving and. if used properly, can result in a highperformance knowledge-based system n u t . it is important t o note that heuristic knowledge plays very different roles in different domains. In the domain of mrdirine. heuristics are important, and this is often the only t t p e of knowledge that is needed and/or available The nature and the coiiiplexity of the medical profession prevents doctors from using purely deterministic approaches or cornpuler-basrd simulations. for example. Consequently, hriiristic knowledge hecomrs the main focus of knowledge acquisition in building medical expert syst,ems. Heiiristirs a r r also important, to engineering, but play a relatively less predominant role than they do in medicine. Engineering knowledge, particularly in the manufacturing domain, is a combination of scienrr-based principles and experience-based heuristics. Traditional computer-based methods used by engineers, such as numerical siinulations, are mainly for the science-based deterministic knowledge. They are useful, but are limitvd due to a n inability to include valuable domain heuristics. Following the Ieswns learned from the mediral domain, the present expert-system technology t,mphasizes mainly heuristic knowledge. SystPms built in the engineering domain using personal interviews exclude murh useful deterministic knowledge, and are difficult t o integrate with traditional software packages. As with conventional software, these heuristicoriented systems are useful to engineering, but are also very limited because the deterministic techniques are excluded. After all, i l is certainly not intelligent for engineers t o build intelligent, systems based on heuristics only with no relation to engineering models, physical principles, and governing equations t h a t have been known to us for years.
6.1.3
The R e a s o n i n g A p p r o a c h .
W-hen faced with difficult problems, hutiian beings reason both inductively and deductively 111 inductive reasoning, niany specific cases and task examples are given, and generalized principles or rules of the domain are sought. In contrast, deductive reasoning takes known general principles and domain rules to produce sperilic cases and recommendations. It is important to realize t h a t inductive reasoning i s for acquiring knowledge and deductive reasoning is for using knowledge. Both are essential for triily intelligent knowledge-based expert system. They should not however, be conlused with the forward and backward reasoning approaches that are known t o tiiany expert systems builders, since both induction and deduction can go either forward or backward in their execution stages. T h e present knowledge-based expert system paradigm is based mainly on the deductive reasoning approach. These systems must possess domain knowledge and rules in their knowledge base, then reason deductively to output specific suggestions to users. The major requirement is chat domain knowledge and rules must be known beforehand and must be strurt,ured in such a way that deductive reasoning can take place. In other words: these systems address only the issue of iising knowledge, leaving the task of acquiring knowledge t o knowledge engineers with the interview process. This paradigm works, but mainly for those domains where t h a t knowledge is readily available or t,asks are well known. Eiigineering certainly does not fit into t h a t category. Not only are many engineering tasks only part,ially known, but also, even when tbe task is known, knowledge may not be readily available. Popular engineering methods, such as numerical analysis and computer simulations, provide engineers with only d a t a and information, not direct knowledge, lor their decision-making. Based on the current deduction-only paradigm, this d a t a and information is not useful for building knowledge-based expert systems, for they require domain knowledge as t,heir inputs. This is one of the main reasons why engineers currently have great difficulties linking expert system efforts with traditional software. If the induct,ive reasoning approach can be added t o the current paradigm, a logic link can be easily built betwwn conventional software approaches and knowIedge-based expert systems for the engineering domain. It will also become clear later t h a t induction is a powerful tool to connect deterministic knowledge with heuristic knowledge. 6.1.4
The Problem Sizes a n d T h e i r A s p e c t s .
Current expert-system technology ran only be applied t o problems t h a t are narrowly defined. There is a trade-off between the size of the problem being considered and the quality of system performance In order t o achieve high performance, which is necessary for a system to be called an expert system, the problem size must be limited If the selected problem is too complex, the present wisdom says that you should only address t h a t problem partially in order t o keep its knowledge base within a manageable size. For the same reason, the aspects of the problem t h a t can be considered by the system are also limited. Most systems provide recommendations based only on considerations of a single aspect. Unfortunately, real-world engineering problems such as manufacturing planning do not fit well wihh this narrowly defined model. They tend t o span broad activities and require consideration of multiple aspects. As a result, most existing expert systems in the engineering domain are very limited in terms of their prartiral values. 6.1.5
The C o n s u l t a t i o n F r a m e w o r k .
Most expert systems developed to date are used mainly for,consultation due t o their origin in medical consultation. In this consultation framework, the problem domain must be well defined, the solution space must be known beforehand, and the internal structures among domain knowledge must be well specified. The system does no more than select a suitable pattern from many pre-stored ones t o satisfy the given constraints. This selection or classification process is the most popular way of building reasoning mechanisms for expert systems today. As a result, most systems developed, even if labeled as design or planning systems, are actually of the diagnostic type Due mainly to this consultation framework,
rurrent, expert systeiiis adopt a studwit-advisor model in their usage. T h e systrm is a s s u m d t,o possess domain knowledge much superior to that of the users, and serves as a domain advisor to users who are acting as students in t,his case. The advisor dominates conversations with its .itudents, h a s them wait for qiieslions froin the system, and does not allow them to actively participate in the problemsolving processes. Furthermore. t,hr advisor does not, in fact, can not, consult with other Pxperts (rithcr human or machine-based) in relevant domains in case it, does not possess enough knowledge for a required task Thus, current systems u-ork in an environment isolated from the world. with only one communication channel 1.0 their users Not surprisingly, this paradigm does not work well for the engineering domain. First, consultation is only one of the many types of aids that engineers may want. T h e selection and rlassificat,ion model drarribed ahovr is not sufficient for many important engineering tasks, such as design, planning. and on-line control. These tasks require systems t o prrform not only selection hut also more importantly, some generative activities i n solving problems Furthermore, the student-advisor model is not suitable for engineering applications whether, in most cases, the engineers who use knowledge-based systems possess a certain amount of knowledge themselves about the domain. They should be allowed to participate in problem-solving niore actively than it is possible with the current paradigm. A more serious limitation, from the viewpoint of engineering application, is that only one-way comniuniratiun is allowed by current expert systems. In engineering, many on-line applirations are needed which require systems t o take raw d a t a devices as inputs rather than answPrs in predetermined form prompted from human users. Other applications. like planning, design, and scheduling, require a strong data-base support for intelligent systems. Surprisingly, current expert systems designed as decision-making aids d o not have strong supports from, or links t,o, existing d a t a base technology This problem is perhaps more serious for engineering than for others since much existing engineering information is already included in many cnmprrhmsive data haws
fnr manufacturing purposes. ‘These manufacturing-based representations should fall a representational continuum with the present CAD represenItitions a t the most detailed elid and with such schemes as G T codings at the most general end. The exact location of manufacturing planning represrntations on this continuum can only he determined through experiment Most Al-bmed techniques provide hierarchical, symbolic representations whirh should be very useful in these needed experiments
.
Capture Planning Knowledge Flexibly T h e rver-rhanging nat,urr of the manulactiiring industry demands computer-
based methodologies that are flexible enough to adapt to the different structures and concerns of mariufarturing planning. It is very likely that portions of the planning knowledge representkd in present systems will become totally inappropriate once the manufacturing environment, which the plans are designed for, changes. This will entail major re-development efforts if planning kiiowledge was not represented in a flexible way. In comparison w t h traditional computer approarhes, AI-based methods are more flexible and adaptive t o modifications and updates and, thus, will be more applicable to future manufacturing-planning needs. 6.2.3
Integrating Planning Constraints.
Planning can he viewed as a constraint-driven reasoning process, and manufacturing constraints in real-life settings always come from very diverse sources. Fiitiire Al-based methods should not only continue to capture these constraints, a 5 they ciirrently do. but also help in tracing and integrating them to create more romprehensivr planning drrisinns Some resrarch in progress is experimentally with the use of A1 techniques as a planning knowledge synthesizer 157) and constraints integrator 158,591 These efforts are critical t o the evolution of integrated planning in the future Synthesize Planning Physirn, Grornrtries, and Time
6.2
Potential Roles
I n general, operation planning drcisinns are rooted in the physics of manufacturing operations; the process planning decisions are based on the geometries of products and tools; the decisions for production planning are related to the time arid resources of a shop. Since we are unable to synt.hesize among physics. geometry, and time simultaneously in the planning stagr, operation, process. and production are dealt with separately, In the future. soine AI-based methods such as induction, concept generalization and spcrialization should be used as knowledge synthesis tools to integrat,? physirs, geometry, and time. Some initial research efforts have shown promise in synthesizing planning knowledge from physical models for operation and prowss planning 1601
Much research is still needed before Al-based techniques are used to their full pot,ential in integrakd and intrlligent planning for manufacturing tasks in the future To better direct research efforts, this section suggests some specific t,asks in manufacturiiig process plannlng where fully matured A l techniques can be useful. Wheii its t,echniques are fully matured and its usages in engineering are fully explored, A1 will have a miirh broader impact on manufacturing planning than do currpnt efforts in building expert systems which only focus on the uailization of existing intelligenre. In the future, AI will become the key technology in generating, representing, integrating, and utilizing the planning intelligence t h a t is essential t o the future of coniputer-integrated manufacturing, 6.2.1
IntPgratr Multiple Planning Constraints and Viewpoints
G e n e r a t i n g P l a n n i n g Intelligence.
A sound manufacturing plan should inrorporat,c multiple viewpoints gathered from a team of manufacturing experts who posses overlapping and competing knowledge ahout, the task being planned. Different suggestions from expert.s will enhance. t.he quality and completeness of the plan These diffwrnt opinions should not be judged on their logical correctness since each of them might be valid given its individual discipline’s point of view Current centralized problem-solving methods in AI, which require logical consistency before implementation. are evolving into distributed and cooperative problem-solving paradigms. ThPse new paradigms allow the modeling of a team expertise, which will be more appropriate for integrated nianufacturing planning in the future, and they can address multiple constraints in parallel, rather than resolving them sequentially before problem solving. This ability t o model team expertise is very important since design and manufacturing are expected t o be conducted in parallel in integrated manufacturing. In that environment, planning becomes the simultaneous cooperation between designers and inaniifacturing engineers. Thus a n AIbased cooperative problem-solving paradigm will play a central role in integrated planning.
A key to the siicress of inlelligenl planning is I,he availahility of planning knowlMost c u r r m t AI-basrd approachrs require t h a t human experts provide this knowledge through interviews. In the fulure, A I methods should be used not. only to organize this human-expert knowledge, but also to automatically generate planning knowledgP from both manufacturing data and feedback from previous plans edgr
Extract Planning Knonlcdge from Maniifarturing Data and Shop-Hoor Informat,ion Maniifaciiiring is inherrni.ly an illformation-rich h u t knowledge-poor domain Much useful planning knowledge cannot. be provided directly by human planners. Knowledge is often hidden w,ithin massive amounts of d a t a collerted from the shop-floor and/or gathered from computer simulations of specific Operations 1521 Therefore. a major need in this doinain is the development of computer methods for automatic information-teknowledge conversion. Some AI techniqries. such as inductive inference [53] and conwptual clustering 1541, are useful in data compression and abstraction. This compressed d a t a can he viewed as domain knowledge for enhancing our planning intelligcnrt- in other AI-based systems 1551. Learn Planning Knowledge from Ferdbark of Previous Plans As mentioned before, a truly intelligent planning system must he selfadaptive and must improve itself in a rlosed-loop fashion. Human beings are often incapable of closing this feedback loop due to the dispersion of manufacturing activities during the execution stage Research in AI machine learning has produrrd methods which can aiitomatically learn domain concepk from given examples and background knowledge 156). In the luture, it is possible that he feedback from executions of previous plans could he collected as training examples for an AI-based learning program to generate new planning knowledge which improves the system’s performance gradually
6.2.4
7 6.2.2
Representing P l a n n i n g Knowledge.
One of she fundamental cont,rihiit.ions which A1 provides to other disciplines is the emphasis on systemic representations of domain features, concepts and logiw. In the future, as more comprehensive representatrim schemes are developed from A1 research and are exercised by engineers in different domains, more diverse types of planning knowledge can he added t o the systems. The flexibility of these AI-based represent,ations will alqo allow us to keep pace with t,he dynamically rhanging manufacturing environment RPprrsrnt hfanufncturing Ffatiires Intelligently I n IhP 6 i I . i i r r . Al-hased methods will lirlp current C A P P research efforts i n fpatiire-based rrpresen1ation.s to yield ncliernes which are well-suited
Utilizing P l a n n i n g Expertise.
Ctilizabion of domain expertise through building knowledge-based expert systems has been one of the most successful applications of A1 to date. In the future, it will st,ill be one of the most useful contributions of A l to manufacturing planning, Its full potential will, however, not be realized unless research efforts can overcome its limitations as discussed above. If these limitations were remedied, Al-based approaches could he used widely to capture, represent, organize, and apply planning expertise t o or various manufacturing settings. They will help in the formalization, preservation, dissemination, and communication of valuable planning expertise to greatly improve productivity in a the manufacturing rnvironmmt
Conclusion
The current status of C A W has been examined and suggestions made as to future direci,ions for research and development in the area of computer-aided process planning Instead of rrviewing specific systems or studying individual efforts, the discussion w-as focused on the fundamental issues which challenge our whole research ronirniinity. T h r development of different planning approaches in terms of the wolution of manufacturing industry was reviewed. Key technologies rrit,iral t o this evolution, siirh as groiip t,echnology in the past, CAD representation in the present, and artificial intelligence in the future, were identified. Sonie basics of computer-aided process planning, which are often overlooked by the rmearch commrrnit.y, were prrsented and criliques of the present research efforts were provided An integrated planning framework has been identified
as t h r next stpp in process plannlng research, and artificial intelllgence techniques were proposed as t h r key t o arhiwing this step Details of this integrated planning framework were presented T h e current limitations and future roles of Al-hased methods in manufactiiring planning have been explained. Computer aided process planning is certainly one of the key areas for successful implementation of computer integrated manufacturing and requires the rooperation not only of academia and industry, but also t h a t of internat.ional research workrrs 11 is hoped t h a t world-wide mutual effort and cooperation will result in lurther improvements and developments to meFt o u r ever-inrreasing challenge for the future, and t h a t Lhis paprr provides enroiiragement for such efforts
8
Ackiiowledgements
References 1. J u n e 1-2, 1987. Proceedings of t,he 19th C l R P International Seminar o n
Manufacturing Systems (Majnr Theme: CAI'P). T h e Pennsylvania State University, U.S A . 2 , Feigenbaurn, E. A,, and Feldrnan. J . 1 x 3 , "Compoters and Thought," McGraw-Hill Piiblishing Company, New York 3. Fpigenhaum, E. A , 1977. "Artifirial Intelligence- Themes and Case Studies of Knowledge Engineering," IJCAI-5, pp. 1014-1029. 4. Chang, T C . , and Wysk, R. A , , 1985, "An Introduction to Computrr-Aided
Process Planning Systems," Preiitice Hall Publishing Company.
5. Chryssolouris, G., and Chan, S., 1985," An Integrated Approach t o Process Planning and Scheduling," C l R P Annals 6. Kals, H. J .I., 1986, special contributions, Universiteit Twente
7. Weill. R. Il
, 1988, special contributions, Israel
IR3-2li
24 Matsiishima. K , Okada. N , and T. Fata. 19R2, " T h e Integration of CAD and CAhl by Applira(ion of :\rtificial n t ~ l l i g e n c eTechniquvs," Annals of the (:IRt', Vol 11 25 Brooks. S I,., May 1986. "Applying Artificial Intelligence Techniques tv Gpnt-ralive Prorra3 Planning Systems." iinpublished M.S thesis. 26 Van't E w e , A. H.. Kals, H. . I . .I.. 1986."XPLANE, A Generative Computer Aided Proress Planning System for Part Slanrifacturing." Annals of the CIRP, Vol 3 5 , No. 2
T h e arithors would like to express sincere appreriation to those who made rontributions in preparing this paper They are especially grateful to Professor \+'. Eversheim, Professor II. J . J Kals. Profrssor F . Kimura. Professor V . R. Milacic, Professor F.O. Rasch, Professor G . Spur, Professor H K Tonshoff, Professor H Warnecke and Professor R Weill who provided very extensive and vahiabk contributions.
9
23. I)t.srottr. Y , and Latomh?. J - C , 1085, "Slaking Compromises Anlong Anlagonist Constraint- in a I'lai~ner." hrt,ifrial Intelligenre, Vnl 27. pp
lnatitutr of Technology.
27 Joshi. S.. and Chang, T., June 1-2. 1987. " C A I) Interface for Automated Proccss Planning," Procerdings of th? 19111C l R P International Conference on Manilfact tiring SystPms. P~nnsylvaniaSt,ale University.
28 W a i i ~ ,W.,.June 1-2. 1 9 8 i . "Application
of Solid Modelling t o Automate .Machining Parameters for Complex Parts," Proceedings of the 19th C l R P Inl.ernational Confermre on Manufartiiring Systpma, Prnnaylvanim St,atP University.
29 Eversheim. W , and I h l s . June 1-2. 1987. "Changing Requirements for C A P - Systems Lead to a New CAP-Data hiodel," Proceedings of t h 19th C I R P lntprnational Conference on Manufacturing Systems. Pennsylvania S t a t e University. 30. Tqang, J . , June I-?. 1987, " T h e Propel Process Planning," Proceedings of thP 19t,h C l R P International Conference on Manufacturing Systems, Pennsylvania State Univrrsity. 31 Iniii. M . . Suzuki. 11.. K i r n i i r a . F , and Sat.a. 1' , June 1-2. 1987. " E x t m d ing Proress Planning Chpabilities with Dynamic Manipulation of Product Models," Proceedings of the 19th C I R P International Conferencr on Maniifacturing Systems, Pennsylvania State University 32. Phillips. R , and Arunthavanathan, ' 1 , J u n e 1-2. 1987, "An Intelligent Design and Process Planning Sysbem," Proceedings of t h e 19th C l R P ln-
ternatinnal Conference on Manufacturing Systems. Pennsylvania %ate IJniversity
zation of the Process Planning Task from an Art.ificial Intelligence Perspective," Proceedings of the 19th C l R P International Seminar on Manufacturing Systems, pp 197-206
33. Woodhrad, R , Dobolyi, 2.. and Pennington, A D , 1986, "Process Planning as a n Application for Expert Systenis Tecl~nology,"Proceedings of Production Engineering Confrrrnre. ASME Winter Annual Meeting, Anaheim. California, pp 143.155
9. Wysk, R. A , , Chang, T C , a n d IIam, I . , 1985, "Automat,ed Process Planning Systems- An Overview of Ten Years of Activities," 1st C l K P Working Seminar on Computer-Aided Process Planning, pp. 13-18.
34. Joshi, S , Chang, T. C , and Liu, R C , 1986,"Process Planning Formalization in an AI Framework." lnternational Journal of Artificial lribelligence Applications in Engineering, Vol. I . Xo. I , pp. 45-52.
10. Eversheim, W , 1985, "Survey of Computer Aided Process Planning Systems," C I R P Technical RPports. ClRl' Annals. pp. 607-611
35 Milacic, R. V , Kalajdsic. hl., I98,I. "Imgical Structure of Manufactiiring I'rocess-I)esign-FundampntaIs of an E x p d System for Manufacturing Process Planned," Proceedings of t,he 16th C I R P Intkrnational Seminar on
8. Subramanyam, S., LII,
S. C-Y., and Zdeblick. W. J , 1987, " A Characteri-
11. Boothroyd, G . , April 1986, "Design for Machining," Report #S, Department of Indust,rial and Manufactunng Engineering, University of Rhode
Island.
- T h e Key t o Design for Manufacture," Report #9, Department of Industrial and Manufacturing Engineering, University of Rhode Island
12. Root,hroydBd, G , J a n u a r y 1987,"Design for Assembly
13. Dewhurst, P., and Boothroyd, G , March 1987, "Early Cost Estimating in
Product Design," Report # I I , Department of Industrial and Manufacturing Engineering, University of Rhode Island. 14. Eversheim, W., 1988, spPrial cont,rihutions, Technischen Hochschule Aachen. 15. Link, C . H., March 1976, " C A P P - CAM-I Automated Process Planning
System," Proceedings of the 13th Numerical Control Society Annual Meeting and Technical Conference, Cincinnati, Ohio. 16. Tulkoff, J., May 1981, "Lockhwd's GENPLAN," 18th Numerical Control Society Technical Conference," Dallas, Texas, pp. 417-421.
17. Li, J . , Han, Chingping, and Ham, I.,June 1-2, 1987, " C O R E - C A P P - A Company- oriented Semi-generative Computer Automated Process Planning System," 19th C l R P International Conference on Manufacturing Systems, Pennsylvania S t a t e Uniwrsity. 18. Allen, D. K., and Smil.h, P. R., October 15, 1980,"Computer Aided Process Planning," Report of Computer-Alded hlanufacturing Laboratory, Bringh a m Young University, Provo, Utah. 19 W y s k , R. A , , 1977, "An Aut,omated Process Planning and Selection Prcgram: APPAS," Ph.D thesis, Purdue University, West Lafayette, Indiana 20 Kotler, R. A,, 1980, "Compulerized Process Planning - P a r t 1 and 2," Army ManTech Journal, Val. 4, Nos. 4 and 5 2 1 . Chang, T C , 1980, "Intrrfacing C A D and C A M A Study of Hole Design," unpublished M S thpsis, Virginia Polytechnic Institute and State University, Blacksburg, Virginia. 22. Vogel, S. A,, and Alard. J . E , 1981. " T h e Autoplan Process Planning System," Proceedings of the 18th International Technical Conference, NC Control Society ~
Manii fact urin g Systems, Tok yu 36. Milarir, R V , 1985, "S..\PT-Expprt System for Manufacturing Process Planning," L'ed , Vol 19. ASME. Florida
, "Conreptual Design Rased on the Linguistic Approach and the Aulomata Theory, Annals CIRP, Vol. 35/1, 1986.
37. Milacir, R. V., Pilipovic, h.1
38 Warnecke, H J , Seidel, U A , 1985, "Computer-Aided Planning of Assembly Processes," Proceedings of C A P P Workshop, Paris. 39 Warnecke, H. .I., Schraft. R. I)., Spingler, J. C.. Schoninger, J , 1987, "Computer-Aided Planning nf Assembly Syst,ems," Proceedings of the 8th ICAA. Kopenhagen. 40 Warnecke, H . J., Schweizer, hZ..Schoninger, J , 1987,"Methodenent-Wicklung zur Taktzeitprognose bei Robotern," C A E Journal, No. 2, pp. 5 6 6 0
H K., Reckendorff, U . , and Srhaele, M , 1987, "Same Approaches t.o Represent the Interdependenre of Process Planning and Process Cont.rol," Proceedings of the 19th CIKP l n t e r t i a h n a l Semin-r on Manufacturing Systems, Pennsylvania State University, pp 257-271.
41 Tnnshoff,
4 2 . Hatvany, J . , 1987," W h a t is Economic, and How Am I t o Know?" Proceedings of t h e 19th C l R P International Seminar on Manufacturing Systems," pp. 5-8.
13. Subramanyam, S., and Lu, S. C-S , 1988, "An Integrated A!/OR Approach to Operation 'Planning Based on Process Perforn;ance Models," Proceedings of the 16th NAMRC Conference 44. Grahowski. E l . , 1985, "Neue Arheitstrrhniken heim Entwerfen mit 'Intel-
Iigenten' 3D-CD-Systrmen, VDI-Rerichtr 5fi5,"Voraussetzunger Erfolgreiclien CAD/CAM Einsatzes, llusseldorf 45 Tonshoff, II K , and Beckendorff. U., 1986, "Forderring der Fert,igung an
die Herhnerintegrierte Konstruklionstechnik," Produktionstechiiisclies Kolloqiuni, Berlin 46. Cotter, S. L., 1985, "Siimrnary of I3enrlrmarking Results," Proceedings of CAM-1's Third Geometrir Modeling Seminar 1985, Computer Aided Maniifartiiring- Intkrnational, h r . , i\rlington.
4 7 . Tonshoff, H . K.,Shunke. A., and Rprkendorff, l J , 1987."Fertigungs-und Normgercchte Konstruktion rind Ahnlirhtcilsuche diirch elementorientierte CAD- Benutzersrhale." VDI-2. 129, N o 5 . pp. 52-57
48. TonshoR, H K , Deckendorff, II , and Scharle, M , 1986,"Integration Aspects of Automated Process Planning iii CIM," Proceedings of the 18th ClRP MFS Seminar, Stuttgart. 49 Tonshoff, H K., and Beckendorff, U . , 1986, "Forderung d P r Fert.igung a n
die Rechnerintegrierte Konstruktioostechnik, Produktionstechnisches Kolloqium, Berlin. 50. Tonshoff, 11. K
,
1988. special contributions, Universitat Hannover.
H. K , Horns, A , and Schade, M , 1987,"Integrated Model Hierarchy for Factory Automation." Prowedings of the IFIP WG 5 3 Working Conference on Software for Factory Automation, Tokyo.
51. Tonshoff,
52 Ian, S C-Y., 1986. "Knowlerlgc-Rased Expert Systems - A New Horizon for Manufacturing Automation," Proceedings of the ASME Symposium on Knowledge- Based Expert, Systems for Manufacturing, pp. 11-23. 53 Michalski, R. S., 1983, " A Theory and hlethodology of Inductive Learning," Artificial Intelligenre, Val 20, No 2, pp. 111-161.
54, Michalski, R. S., and Stepp, R.. 1983,"Automated Construction of Classification: Conceptual Clustering Versus Numerical Taxonomy," IEEE Transaction of Pattern Analysis and Machin? Intelligence, Vol. 5, No. 4, pp. 396410.
55. Rendell, L. A,, 1986,"A General Framework for Induction and a Study of Selective Induction," Machine Learning, Vol. I , No. 2 , pp. 199-226.
56. Rendell, L. A., Benedict, P., arid Cho, FI , 1987, "Concept Acquisition from Examples: Measurenlent of Systein Performance and Suggestions for Iinproved Design." LIIUC Rrporl, N o DCS-R-87-1315. 5 i . Chen, I<.. and Lu, S. C-Y . 1988, " A 54achine Learnkg Approach to bhe Aut,oniat,ic Synthesis of Mechanistic Knowledge for Engineering Decision Making,'' 4th IEEE Conference on Artificial Intelligence Applications. S8. M a l - ~ rA, K , and Lu. S C-Y , I>pcpmher 1987, " A n AI-Baed Approach for the Integration of Mi11t.iple Sources of Knowledge to Aid Engineering De-
sign," Journal of Mechanisms, Transmissions, and Automation in Design, ASME Transaction, to appear. 59. 1.11, S. C-Y., and Thompson, J . R , Jim? 1988, "Multiple, Cooperating Knowledge Sources for Integrated Engineering Design," The First loternational Conference on Industrial and Engineering Applications of Artificial IiiLelligence and Expert Systems, Tullahonia, Tennessee.
60. Lu, S. C-Y., Decerriber 1987, "Autornal,icAcquisition of Domain Knowledge from Mechanistic Simulations for Building Engineering Expert Systems," Journal of Engineering fur Industry, ASME llunsactions, to appear
60 1