Using Schedules for Operational Guidance€¦ · the methodology also provides a basis for incrementally revising the current schedule in response to unanticipated changes in Ihe

- - - - -- -- --- - :---- - - -

Using Schedules for Operational Guidance

Stephen F.Smlth.Pedro Muro.NlcolaMuscettola.peng SI OwThe Robotics Institute

Camegie Mellon UniversityPittsburgh, PA 15213

The broad goal o, production scheduling is to produce a factory behavior where products are produced in

a timely and cast-effective manner. In complex environments, advance development of. a scheduleappeélrs fundamental to this goal, as it is only through anticipation 01 resource contention (the prim~ryobstacle to efficient lactory behavior) that the deleterious effects of these conflicts can be minimized.

However, the extent to which the behavior impliedby a generated schedule is actually realized dependson the manner in which the schedule is "executed" on the factory floor. The unpredictability of factoryoperations (e.g. machine breakdowns, quality control failures) will inevitably lorce deviations from

prescribed behavior, and an advance schedule will be of little use unless the guidance it contains iscontinually adapted to the specifics of the current factory state. Research in production scheduling [2] has

trac:titionallyignored problems of schedule execution, evaluating its results as if the world were entirelypredictable and generated sch~dules could be executed exactly as planned. On the other hand, researchrelevant to factory floor oontrol"[6]has emphasized the development of local dispatch heuristics for

dynamic decision-making, and conceded any potential benefits 01advance scheduling (withthe exceptionof establishing release and due dates). In this talk we consider the scheduling problem from the

operational perspective of providingglobal guidance to the. actual decisions that must be made on thefactory floor.

Our approach to integrating predictive scheduling with reactive control builds on the schedulingmethodology implemented in the OPIS scheduling system [7]. This methodology is motivated by the viewthat effective compromise among conflictingobjectives requires an ability to reason from different local

scheduling perspectives (in particular from both order-centered and resource-centered viewpoints), and

. .'há1 de~i~ions a~ to w~ich local perspective to adopt at anypoint during-scheduling should be made""opportunistically on the basis of characteristics of current solution constraints. This point is argued in (8],and has been validated in the context of schedule generation in [5]. More relevant 10the discussion here,

the methodology also provides a basis for incrementally revising the current schedule in response to

unanticipated changes in Ihe production environment. In this case, detection and analysis of the actualconstraint conflicts that haye been introduced into the schedule (i.e. precisely those portions of the

schedule that haye become invalidated) are used to focus the reyision process. In [4], we present amodel for conflict analysis and selection of appropriate reactions, along with supporting experimental

justilication.

Assumingthis methodologyfor reactivelymaintainingschedules, the issue of defininga controlpolicytogovemits use in actual factoryfloordecision-makingremains. One obviouscandidateis a controlpolicy

(Draft) Proceedings of Workshop on Production Planning and Scheduling. 1988

----- - - - - - ---2

of "followthe schedule", which impliesthat decision-makingrequiresschedulerevision whenever any

discrepanG)'is detected between the scheduleand the actualfactorystate. However there are severalreasons whythis is toa extreme:.In most scheduling applications,the schedulermust necessarilyoperatewithapproximations.

of actual temporal constraints (e.g. operation durations, resource setup times) that willconsistently lead to minordiscrepanciesbetween the factorystate and the schedule. Suchdiscrepancies reflect the "normal"unpredictabilityof the productionsystem,and are unlikelyto undennine the sequencingand assignmentdecisionscontainedinthe current schedule.5imilarty,the scheduler maybe forcedto produce over-constraineddeeisions. In situationswhere detailed resource assignmentsare necessary to optimizesequencing decisions, itmay be that only the sequences are important (and not the particular asslgnments).Generally speaking, a choice to revise the schedule shoulclbe predicated on someexpectation that current discrepancies,have moreglobalirf1)lications..Even if it were possible to maintaina predictiveschedule at the Iowestlevel of detail inreal-time (whiCh is unlikelybut depends on the specificapplication),it makes little sense todo so. Many scheduling decisionsthat must be made.are Iocally~ntained (e.g. choicesbetween functionally.identical~aC~.p'rodu.~ k>.adingand unlo~dings~enCeS, ate.) andare of no consequence to the globalcoordinationproblem.The schedulershould opei'ate ata level of detail that is sufficientto imposeglobalguidanceyet retains as much execution-time flexibilityas possible(again,a functionof the specificapplication).

These arguments suggest the use of a control policy that provides some level of scheduleinterpretationlrefinementat executiontime. This,of course, requiresthat the control policyoperate withknowledgeof the schedule(s assumptionsand intenUonsin establishingthe constraints impliedby theschedule.'In this regard, we are investigatingthe use of preferenceconstraints(as defined and used inthe 1515scheduling system (1)).This approach is outlined in[3] in the context of a petri-net basedcoordinationsubsystem.

A second issue bearing on the use 01schedules as operational guidance concerns the extent to which the

unpredictability of factory operations is reflected in the maintained schedule, as this can have a significanteffect on both the frequency and the efficiencyof schedule revision. Here, concepts from hierarchical,

least-commitment planning are uselul in some respects (e.g. 'in varying the level 01 abstraction at whichthe scheduler makes decisions according to how close the decisions being contemplated are to being

executed). At the same time, the concept of least-commitment is somewhat at odds with the purpose ofscheduling (I.e. to make choices as to when and where activities should be executed so as to bestaccommodate overall objectives). To arrive at a reasonable compromise between the desire to makechaices and the desire to remain flexible, we are investigating techniques for taking into account

. knowledge about the specific sources of unpredictability in the production environment during scheduling.For example, in the production environment we are currently considering (a computer board assemblyand test line), the failure of tests at various points in the process constitutes the primary source 01

.unpredictability.In this case, historicaldata regardingsuccess and failurerates, along withknowledge01the necessary repair operationsfordiflerentfailures,is used as a basis for introducingslack at strategicpointsin the schedule.

We are investigatingthe issues discussed above via use of a simulationtestbed. The simulatoris uniquein its provisionto definecontrolpoliciesthat make use of an existingschedule and can involvethe OPISscheduler lor schedule revisionpurposes. Thus, policiessuch as "followthe schedule" as well as local

~ ~ - oo..' ._..

_. no.. . ...__...__


--- -- -- - - -----

3

dispatch heuristics presume noschedule canbesimulatedand evaluated.Byexarniningthe resultso. thesimulator (i.e. the "actual factory behavior") under different control policias and across differentenvironmental conditions (e.g. levels of unpredictability).we feelwe willbe able to both better quantify the

potential advantages of predictive guidance and establish the necessary coupling between schedulingand control.

We describe the above mentioned fechniques that we are currently investigating. the experimental

testbed. and the experimental results we have obtained to date.. .

. .

-.- ' . .. .-.


- - -- ---- -- - - - - - - --

4

References

1. Fox, M.S., and S.F. Smith. "ISIS:AKnowledge-Based System for Factory Scheduling"_ ExpertSyslems 1, 1 (July 1984), 25-49.

2. Graves, S.C. "A Review of ProductionScheduling". OperalionsResearch 29,4 (July-August 1981),646-675.

3. Martinez, J., P.R. Muro, M. Silva, S.F. Smith, and J.L. Villanoe!. MergingArtificiallntelligenceTechniques and Petñ Nets for Real-TimeScheduling and Controlo' ProcIuctionSystmes. 12th IMACSWorld Congress, Paris, July, 1988.

4. Ow, P.S., S.F. Smith, and A.Thiriez. Reactive Plan Revision. to appear in Proceedings AAAI-88,StoPaul, Minn., August, 1988.

5. Ow, P.S. and S.F. Smith. ViewingScheduling as an Opportunistic Problem Solving Process. InAnnals of Operations Research: Approaches lo Intel/igenlDecision Support, R.G. Jeroslow, Ed., BaltzerScientific Publishing Co., 1988, pp. 85-108.

6. Panwalker, S.S. and W. Iskander.."ASurvey01.S.cl:1edulingRules". Operations.F.!.~search25 (1977),45-61 .

7, Smith, S.F. A Constraint-Based Frameworkfor Reactive Management of Factory Schedules. Proc.1st Inter. Conf. on Expert Systems and the leading Edge in Production Management, Charleston, SC,May, 1987, pp. 349-366.

8. Smith, S.F. and P.S. Ow. The Use o, MultipleProblem Decompositions inTime-Constrained PlanningTasks. Probo IJCAI-85, los Angeles, CA,August, 1985, pp. 1013-1015.

-----

I

I.,


- - - --- - - --

---

!

WORKSHOP ON PRODUCTIONPLANNING ANDSCHEDULING

AAAI - ST. PAUL 25 AUGUST1988

PM

1.+ A Cooperative Approach to Large Scale Production Plan ning - P.Weisser*and R. Howie

2.+ Knowledge Based Planning and Scheduling - K.Fukunaga*and Y.Tozawa

3.+ Dynamic Scheduling in. a High-Volume Manufacturing Environment- V. Chiodini*

4. Preferential Relaxiation of Temporal Period Constraints - N.Sadeh* and M. S. Fox

5. Using Schedules for Operational Guidance - S. F. Smith*.P. Muro.N. Muscettola, and P. S. Ow

6. C.RYSTAL : Planning and Scheduling System - J. Ashfordand M.M .. *eler, ..

7.+ Scheduling Against Nonlinear Constraints - S. Becker*,E.Braude, P. Maglio,and L. Monk

8.+ Machine Learning Applications to Job Shop Scheduling - M.R.Hillard and G. E. Liepins*

9.+ A Fuzzy Job Shop Scheduling System - R. M. Kerr*,N. Walker. andC. Heine

10. Negotiation Based Scheduling in Software Projects - A. Safavi*


- --- - - --- - - - - -

-------------------------------------------

~~&~ D ~~@~~~-----------------------------------------

AMERICAN ASSOCIATIONFOR ARTIFICIAL INTE-lLIGENC-E

SPECIAL INTEREST GROUP IN MANUFACTURING

------- -----

WORKSHOP ON PRODUCTION PlANNING ANDSCHEDUllNG

AAAI - ST. PAUl 25 AUGUST 1988----- -

program Comm~.'....

Mark s. Fox, The Robotics Institute, Carnegie Mellon UniversityKarl G. Kempt, Knowledge Applications Laboratory, Intel Corporation

--------------

Riahts

Al! rights reserved by the individual author or as otherwise noted.----------------


- - -- - -- -

AAAIWorkshop on Manufacturing PI~nningand Scheduling

8:30am: Welcome

8:45am - 10:15: Process Plannlng.Soma Issiles In the Developmentof ExpertSystems For Process Planning of MachinedParts, Hummel and Brooks..Research Issues InProcess PlanningForNetShape Manufacturing,Mlllerand Waldron..Solid ModelingAndGeometricReasoningForDesignAndProcessPlannlng,Nau et al.

.Application 01 Automated Process Plannlng To Thin SheetParts Manufacturlng, Verrnaut andDetollenaere.

.Frame-based Stepwise Reasonlng Simulates Human Modelllng,Yu and EIam.

10:30am - 12:30pm: Modellng and Control.Logistics Management System (LMS): Continuous Flow Manufacturlng Uslng ArtificialIntelligence To Schedule AndControl Production In A Semiconductor Facllity, Fottlyce et al.

. DREAMA Framework For DistributedReactive Factory Managernent, Hynynen..Qualitative process Automation, Lagnese..Intentional-Qpportunlstic Planning And Scheduling For Robotlc Assembly, Xia.

2pm - 5:30pm: Scheduling.Knowledge Based Planning And Scheduling, Tozawa and Fukunaga.

.Crystall Planning And Scheduling System, Meler and Ashfottl.~..í.~!e.Prelerential Relaxation OfTemporal Period Constralnts, Sadeh and Fox.

. A Fuzzy Jobshop Scheduling System, Kerrand Walker..Genetic AlgorithmAnd Classifier System ApplicationsTo Scheduling, Uepins and Hllllatd..A Cooperative Approach To Large Scale Production Plannlng, Weisser and Howie..Negotiation Based Scheduling In Software Projects, Safavi..Scheduling Against Nonlinear Constraints, Becker et al.

'.~~''''..'_''--''-. .;' ~ '~'-..' :?3.'.":~illi~1f1.:~~~.'~~.~.'.'.' ..?:~~~~~~.~;~'€~t~.'~J ti'~'~~'t~0;;;~~~i_~~~~ul~,;.;¿z:t. ~:,~:}J'k:


Documents

Using Schedules for Operational Guidance€¦ · the methodology also provides a basis for incrementally revising the current schedule in response to unanticipated changes in Ihe