1 Scheduling and Control of Flexible Manufacturing Systems: A Critical Review Chuda Basnet Department of Management Systems University of Waikato, Private Bag 3105, Hamilton, New Zealand and Joe H. Mize School of Industrial Engineering and Management Oklahoma State University, Stillwater, OK 74078, U.S.A.Abstract Flexible manufacturing systems (FMS) are distinguished by the use ofcomputer control in place of the hard automation usually found in transfer lines.The high investment required for a FMS and the potential of FMS as a strategiccompetitive tool make it attractive to engage in research in this area. This paperpresents a review of literature concerning the operations aspect of FMS. Articlesemphasizing many methodological perspectives are critically reviewed. The reviewis done from multiple viewpoints. Future research directions are suggested.Key Words: Flexible manufacturing system, production planning, scheduling,production control.
11. Introduction Flexible manufacturing systems (FMS) are distinguished by the use ofcomputer control in place of the hard automation usually found in transfer lines.This enables FMSs to reconfigure very rapidly to produce multiple part types. Useof fixtures and tool magazines practically eliminates setup time. These featurespermit economic production of a large variety of parts in low volumes. FMSs areincreasingly being adopted in the manufacturing sector on account of theadditional advantages of rapid turnaround, high quality, low inventory costs, andlow labour costs. The high investment required for a FMS and the potential ofFMS as a strategic competitive tool make it attractive to engage in research in thisarea. The research problems raised by the industrial espousal of FMS could bebroadly classified into two problem areas: design problems and operation problems.At the design stage, one is interested in specifying the system so that the desiredperformance goals are achieved. The operation problems are aimed at makingdecisions related to the planning, scheduling, and control of a given FMS. Thispaper presents a review of the published literature on the operation problems ofFMS. We take stock of the progress in this area considering various aspects of theliterature. A considerable body of research literature has accumulated in this area sincethe late 1970s when the first papers were published. A few surveys of theliterature have also appeared (Buzacott and Yao 1986, Rachamadugu and Stecke1989, Gupta et al. 1989). However, these reviews focused on specific perspectivessuch as analytical models, or scheduling problems. In this paper we haveattempted to review articles having wider methodological perspectives whileconcentrating on the operations issues. We have also brought the review more up-to-date. We review the literature from multiple viewpoints: 1. Methodology used in resolving the problem
2 2. Applications viewpoint 3. Time horizon considered 4. FMS factors considered In the following sections we present the review from the above viewpoints. Inthe final section we will conclude with some directions for future research.2. Methodology Based on the methodology followed, FMS operations literature could beclassified in the following ways: 1. Mathematical programming approach 2. Multi-criteria decision making approach 3. Heuristics oriented approach 4. Control theoretic approach 5. Simulation based approach 6. Artificial intelligence (AI) based approach There is some cross fertilization among these approaches. For example,some AI based approaches use simulation to generate or evaluate schedules. Inthe following discussion, the approaches are classified on the basis of their mainemphasis.2.1. Mathematical programming approach In this approach, the researchers have cast the problem into an optimizationmodel. Buzacott and Yao (1986) present a comprehensive review of the analyticalmodels developed for the design and control of FMS up until 1984. They stronglyadvocate the analytical methods as giving better insight into the system perfor-mance than the simulation models.
3 To manage the complexity of the problem, Stecke (1983) and many otherauthors who have followed her divided the FMS operation problem into twosubproblems: preproduction setup and production operation. In this view, a FMSis prepared beforehand for the given part mix: loading the tools, allocating theoperation to the machines, allocating the pallets and fixtures to the different parttypes. After this preparatory planning phase, the remaining problems are calledoperational problems and solved later. Stecke (1983) places stress on pre-production setup of the FMS. This is to be carried out frequently, as the part mixchanges. To carry out a complete setup, a FMS manager would solve 5 problems:1) Part type selection problem. This problem determines the part types to be produced in the FMS out of the total production requirement of the company.2) Machine grouping problem. Stecke would partition the machines in the FMS so that machines in a group can all perform the same operations.3) Production ratio problem. This problem is related to problem 1 - determine the ratio of the parts selected to be manufactured in the FMS.4) Resource allocation problem. This problem determines the allocation of pallets and fixtures to the part types.5) Loading problem. The solution to the problem will simultaneously allocate operation of the part types and the corresponding tools to the machine groups. Stecke (1983) then goes on to describe models for the grouping and loadingproblems. For these problems, the major constraint is the capacity of toolmagazines of each machine tool. The minimum number of machines required tocover all operations is calculated using an optimization formulation to pack asmany tools as possible in few machine tools, at the same time making enough tool
4allocations to cover all the part types. This formulation gives the number of groupsneeded. If there are more machines than the number of groups, the additionalmachines are tooled identical to some of the ones that are grouped. This way, themachines are pooled to allow maximum flexibility. In Steckes methodology, theoperations and corresponding tools are then assigned (loaded) to the machinegroups. She suggests 6 different objectives to optimize during the loading phase: 1)Balance the assigned machine processing times. 2) Minimize the number ofmovements from machine to machine. 3) Balance the workload per machine for asystem of groups of pooled machines of equal sizes. 4) Unbalance the workload permachine for a system of groups of pooled machines of unequal sizes. This objectivestems from earlier results of Stecke and Solberg (1982) that recommendsunbalancing the workload for each machine when the pooled group sizes areunequal in order to obtain maximum production rate. 5) Fill the tool magazines asdensely as possible. 6) Maximize the sum of operation priorities. The formulations of Stecke (1983) lead to large nonlinear mixed integer prob-lems. She suggests various linearization schemes. Steckes planning problemsplace much of the scheduling problem in the setup stage. Once the setup is doneas per the five specific sub-problems, most of the resource allocation is alreadycomplete. The setup is carried out for a particular part mix. It is not clear whenone of the six loading objectives is to be favoured over the others. In some cases,where the machine tools are separated over a long distance, the choice is obvious.In other cases the answer is hard to discern. The grouping problem does notconsider the production ratio of parts. Thus, it could give an answer which is notdesirable from the view point of maintaining the production ratio. Another problemwith the formulation is the large number of variables and constraints that resultfrom the linearization of the problems. That makes the approach computationallyexpensive. Berrada and Stecke (1983) have proposed an efficient branch and
5bound procedure for solving the loading problem with the objective of workloadbalancing. Steckes approach is explained here at length because othermathematical modelling approaches build upon this foundational work. Lashkari et al. (1987) developed a formulation of the loading problem. Theirformulation considered refixturing and limited tool availability. Besides thisproblem, they place an upper bound on the number of tools that may be assigned.They consider two objectives: 1) Minimization of total transportation requirementsof the parts, and 2) Minimization of refixturing requirements. The formulationshave products of 0-1 integer variables. Lashkari et al. (1987) linearize theformulation to solve the problem using linear integer programming code. Theircomputational experience shows that even for small problems, the problem sizebecomes very large. In order to reduce the search, they suggested dividing theproblem into two sub-problems, the result of which could be used as an upperbound for the original problem. Unlike Stecke (1983), Lashkari et al. will permitonly one allocation of a machine to an operation. This would curtail some flexibilityat the operation control level. Their modelling is suitable only when the parts mustalways traverse to and from a central storage for every inter-machine transfer.Further, the objective function lacks the relative weighting for the different parttypes. Wilson (1989) used simpler and more straight forward formulation of theconstraints to solve the same problem as discussed by Lashkari et al. (1987). Hedemonstrated substantial savings in computational effort using his modelling ofthe constraints and the objective function. Shanker and Rajamarthandan (1989)present a similar model with the objective of part movement minimization. Incontrast to Lashkari et al. (1987), they do not require the parts to go to a centralstorage after every operation. Also, they are not interested in the distance travelled:only the number of movements is of concern. Like Wilson (1989), they exploit the
6particular structure of the problem to obtain linearization of the problem. Theyalso reported that high computational effort was required. Han et al. (1989) address the setup and scheduling problem in a special typeof FMS: where all the machines are of the same type, and tools are borrowedbetween machines and from the tool crib as needed. In their model, the number oftools is limited. The purpose of their model is to assign tools and jobs to machinesso that the borrowing of tools is minimized while maintaining a reasonableworkload balance. This is a nonlinear integer programming problem, and iscomputationally expensive. To solve the problem efficiently, the authors propose todecompose the problem. The two sub-problems each have the same objective asshown above. But the constraints are divided. The first problem finds an optimumtool allocation, given the job allocation. The second problem finds an optimal joballocation, given the tool allocation. Phrased in this way, both problems becomelinear. The first problem is a capacitated transportation problem, and the secondis a generalized assignment problem. It is suggested to solve the two problemsiteratively. The FMS investigated by Han et al., is special. All machine tools areassumed identical. Consequently, the jobs remain at one machine, and the toolsare moved to the machines as needed. Kimemia and Gershwin (1985) report on an optimization problem thatoptimizes the routing of the parts in a FMS with the objective of maximizing theflow while keeping the average in-process inventory below a fixed level. Themachines in the cell have different processing times for an operation. Network ofqueues approach is used. The technique showed good results in simulation. Chenand Chung (1991) evaluate loading formulations and routing policies in asimulated environment. Their main finding was that FMS is not superior tojobshop if the routing flexibility is not utilized. Avonts and Van Wassenhove (1988)present a unique procedure to select the part mix and the routing of parts in a
7FMS. A LP model is used to select the part mix using cost differential fromproducing the part outside the FMS. The selected loading is then checked by aqueuing model for utilization in an iterative fashion. Hutchison et al. (1989) provide a mathematical formulation of the randomFMS scheduling problem, where random (not preselected) jobs arrive at the FMS.Their formulation is a static one in which N jobs are to be scheduled on Mmachines. The objective is to minimize the makespan. They present a mixedinteger 0-1 programming formulation. They solve this problem by a branch andbound scheme. A single formulation solves the allocation of the operations to themachines and the timed sequence of the operations. However, their study assumesthat material handling devices, pallets, buffers, and tool magazines do notconstrain the system. Further, at most one alternative is allowed for any operation.An alternative approach to this problem is to decompose it into two subproblems.The first problem is the allocation of the jobs to the machines in the routings. Thesecond problem is the time bound sequencing of the jobs, the standard job shopproblem. Hutchison et al. (1989) report on a comparison of the performance of theabove two methodologies and another methodology which was based ondispatching rule (SPT). A novel feature of their simulation experiment is their useof a measure of flexibility: probability of an alternate machine option for anyoperation. They concluded that the programming formulations producedsubstantial improvement in makespan over the dispatching rules. However, ascompared to the decomposed problem, the unified formulation did not producesignificant improvement in makespan to justify the additional computational effortrequired. In the above approach, the tool magazines do not constrain the system.Hence the first subproblem of the decomposition can allocate all the jobs to theirmachines. However, when the tool magazine is considered restraining, it may not
8be possible to allocate all the jobs for one tooling setup. Then this subproblemresolves to a selection problem. Out of the pool of waiting jobs, jobs are selected tobe processed in the next planning period (part type selection problem). Theselected parts are then sequenced. The process is repeated period by period. In thisapproach, it is assumed that at the beginning of each planning period all the toolsare reassigned and replaced in the tool magazine. Shanker and Tzen (1985) propose a mathematical programming approach tosolve this part selection problem for random FMS. Their approach is similar to(Stecke, 1983). Stecke assumes the part ratio as given and the planning horizon asindefinite whereas Shanker and Tzen consider individual parts and a fixedplanning horizon. They have a constraint on the tool magazine capacity which isvery similar to Steckes. They constrain the model to find a unique routing for eachpart type (in contrast to Stecke). Two objectives are considered: 1) Balancing theworkload, and 2) Balancing the workload and minimizing the number of late jobs.The resulting problems are, again, non-linear integer problems. Even afterlinearization, the problems are computationally too expensive, and they furtherpropose two heuristics corresponding to the two objectives. For balancing theworkload, they propose essentially a greedy heuristic which attempts to allocate tothe most lightly loaded machine the longest operation first. For the secondobjective, the same heuristic is modified to include the overdue jobs with thehighest priority. Their computational experience showed that the analyticalformulations would be too formidable to be of practical use. Shanker andSrinivasulu (1989) modify the objective to consider the throughput also. Acomputationally expensive branch and backtrack algorithm is suggested as well asheuristics. In the above approaches for random FMS, the scheduling of the FMS isdecomposed into two problems: part type selection, and sequencing of jobs. The
9sequencing is done using one of the dispatching rules. Of course, some (e.g.branch and bound) search could be used to solve the sequencing problem too.Hwan and Shogun (1989) present the part selection problem for a random FMSwith machines of a single general purpose type capable of producing all part types.They include the due date and the quantity of parts needed to be produced in theirformulation. By ignoring the tool overlapping (cf. Stecke, 1983), they considerablysimplify the tool magazine constraint. Their objective is to maximize the number ofpart types selected over a planning horizon. They take care of due dates byweighting on the selected part types. By assuming a single machine type, theirproblem essentially boils down to maximizing the utilization of the tool slots in thetool magazines. They report computational experience on two Lagrangianrelaxation techniques they used to solve the problem. Their heuristics andLagrangian methods obtained solutions close to optimal solutions found by thebranch and bound method. The CPU times required by the three methods aresuccessively order of magnitudes higher. Sarin and Chen (1987) approach the loading problem from the viewpoint ofmachining cost. Computational methodologies to solve the integer programmingformulation are proposed. Ram et al. (1990) consider this problem as a discretegeneralized network and present a branch and bound procedure. Co et al. (1990)have suggested a four pass approach to solve the batching, loading and toolconfiguration problems of random FMS. In this approach, compatible jobs arebatched together using integer programming. The solution is then improved uponin three further stages. Jaikumar and Van Wassenhove (1989) propose a hierarchical planning andscheduling decomposition of FMS operation problems. In the first level, anaggregate production model is used. This is a linear programming model thatchooses parts to be produced in a FMS during the next planning period. The
10remaining parts are assumed to be produced elsewhere at a cost difference. Theobjective is to maximize the cost difference while allowing for the inventory cost forwork in process. The essential constraints are the demand for the parts and themachine capacity. Put simply, the objective of the second level is to minimize toolchangeover. The production requirements and the tool and machine allocation aredetermined in levels one and two. All that remains in the third level is to determinea feasible schedule that will fulfil the above requirements. Detailed requirementssuch as buffer requirements, and material handling constraints, are taken care ofat this level. Jaikumar and Wassenhove recommend simulation using somedispatching rule to carry out this level. If a feasible schedule cannot be obtained,the planning process is reiterated. They discuss the application of their frameworkin an existing FMS and point out that the primary problem is at the first level -selection of parts. Once this is decided upon, the other two problems can be solvedby simple heuristics. Mathematical models in the literature are not efficient for reasonably sizedproblems. Further, they make simplifying assumptions which are not always validin practice. The assumptions, of course, change with the models: some modelsassume automatic tool transport, some others will neglect delays caused byautomated guided vehicles (AGV), still others will assume that tool magazines,pallets and fixtures do not constrain the models in any way, and so on. Themodels also take a static view of the shop floor. It is assumed that all the plannedactivities will be carried out exactly, or the disruptions are infrequent enough thatperiodic solution of the problems will be practical.2.2. Multiple-criteria decision making approach Operating an FMS is an activity with multiple criteria. Some authors havebrought in these criteria in their modelling. Lee and Jung (1989) formulate a part
11selection and allocation problem using goal programming. Their model considersthe goals of 1) meeting production requirements, 2) balancing of machine uti-lization, and 3) minimization of throughput time of parts. Deviational variablesrepresenting the under- and over- achievement for each of the goals are used tomeasure the deviation from the goal. The model casts even the technologicalconstraints into goal constraints. The goal programming model of Lee and Jungcan provide the decision maker with a satisficing solution for given goals and theirprioritization. But even with restrictive assumptions, the model is computationallyexpensive for practical use.Ro and Kim (1990) discuss heuristics for solving six operational control sub-problems considering the criteria of makespan, mean flowtime, mean tardiness,maximum tardiness, and system utilization to solve sub-problems. OGrady and Menon (1987) present a case-study where multiple criteria wereused in making decisions about master scheduling a FMS. Conflicts are resolvedby using assigned weights for the criteria of tool magazine use, machine utilization,due-date performance, and choice of sold products. Integer programmingformulation is used. Kumar et al. (1990) present a multi-criteria approach to theloading and grouping problems in a FMS. Their approach aims to provide a largenumber of feasible solutions (and objectives) for the choice of the decision maker. Optimization of FMS operations is difficult. It is even more difficult to do itwith multiple criteria. But in view of the multi-objective nature of the operationproblems, much work needs to be done in this area, and we have just seen thebeginning of this approach.2.3. Heuristics oriented approach To counter the mathematical difficulties with optimization, use of heuristicshas been actively investigated. These heuristics may take the usual form of dis-
12patching rules or they may be more complicated. Extensive study of dispatchingrules have been carried out in the general job shop context (Conway 1965; Conway1965b; Gere 1966; Panwalker and Iskander 1982). In the same vein, numeroussimulation studies of dispatching rules have been carried out in the FMS area. Nof et al. (1979) carried out a study of different aspects of planning andscheduling of FMS. They explore the part mix problem, part ratio problem, andprocess selection problem. In the scheduling context, they report on three partsequencing situations: 1) Initial entry of parts into an empty system, 2) Generalentry of parts into a loaded system, 3) Allocation of parts to machines within thesystem (dispatching rules). They examined three initial entry control rules, twogeneral entry rules, and four dispatching rules. Their conclusion was that all theseissues were interrelated: performance of a policy in one problem is affected bychoices for other problems. Stecke and Solberg (1981) investigated the performance of dispatching rulesin a FMS context. They experimented with five loading policies in conjunction withsixteen dispatching rules in the simulated operation of an actual FMS. Underbroad criteria, the shortest processing time (SPT) rule has been found to performwell in a jobshop environment (Conway,1965 ; Conway, 1965b). Stecke andSolberg, however, found that another heuristic - SPT/TOT, in which the shortestprocessing time for the operation is divided by the total processing time for the job -gave a significantly higher production rate compared to all the other fifteen rulesevaluated. Another surprising result of their simulation study was that extremelyunbalanced loading of the machines caused by the part movement minimizationobjective gave consistently better performance than balanced loading. Iwata et al.(1982) report on a set of decision rules to control FMS. Their scheme selectsmachine tools, cutting tools, and transport devices in a hierarchical framework.These selections are based on three rules which specifically consider the alternate
13resources. Montazeri and Van Wassenhove (1990) have also reported onsimulation studies of dispatching rules. Buzacott and Shanthikumar (1980) consider the control of FMS as ahierarchical problem: a) Pre-release phase, where the parts which are to bemanufactured are decided, b) Input or release control, where the sequence andtiming of the release of jobs to the system is decided, and c) Operational controllevel, where the movement of parts between the machines is decided. Theirrelatively simple models stress the importance of balancing the machine loads, andthe advantage of diversity in job routing. Buzacott (1982 ) further stresses thepoint that operational sequence should not be determined at the pre-release level.His simulation results showed that best results are obtained when: 1) For inputcontrol, the least total processing time is used as soon as space is available, and,2) For operational control, the shortest operation times rule is used. In the study of Shanker and Tzen (1985), the formulation of the partselection problem is mathematical; but its evaluation was carried out inconjunction with dispatching rules for scheduling the parts in the FMS. Further,on account of the computational difficulty in the mathematical formulation, theysuggested heuristics to solve the part selection problems too. On the average, SPTperformed the best. Moreno and Ding (1989) take up further work on heuristics(for part selection) as mentioned above, and present two heuristics whichreportedly give better objective values than the heuristics in (Shanker and Tzen,1985). This, however, they are able to do by increasing the complexity of theheuristics. Their heuristic is goal oriented - in each iteration, they evaluate thealternate routes of the selected job to see which route will contribute most to theimprovement of the objective. Otherwise, their heuristic is the same as that ofShanker and Tzen.
14 Chang et al. (1989) report on a heuristics based beam search techniquedesigned to solve the random FMS scheduling problem. The root of their searchtree has no operation scheduled. They progressively go along the time line andschedule more and more operations until at the final leaf, all the operations arescheduled. At each node, to evaluate the schedule, they carry out a simulationusing the SPT rule. This SPT rule identifies the critical path in the schedule. Forthe first machine in the critical path, they evaluate all the possible alternate assign-ments. Only a certain number (beam width) of assignments is then selecteddepending on the makespan obtained. A contribution of Chang et al. is a measureof flexibility of the manufacturing system. This is called the flexibility index. Itdenotes the average number of workstations able to process an operation.Flexibility index is 1 for the conventional job shop. For various values of theflexibility indices, they compare their algorithm against several dispatching rules.As can be expected, their algorithm gives better results than the dispatchingresults at the cost of increased computational effort. It can also be seen that as theflexibility of the FMS increases, even a beam width of 1 gives very good results. Chang and Sullivan (1990) propose a reduced enumeration algorithm forgenerating sets of active schedules for FMS. Test problems showed this to be aneffective approach compared to complete enumeration. Donath (1988) developed a heuristic based hierarchical methodology toschedule a FMS in near real-time. In his approach, at every point of decision, e.g.completion of a job, a program called SCHEDULE is run. This makes decisions onthe next assignment of assignable operations. His decomposition has two mainsubproblems. In the first, a cost of assigning an operation to a machine iscalculated on the basis of process time, idle time, and the average time for thatoperation. Secondly, a generalized assignment problem is solved to assign the jobsto the machines. All the pending operations are assigned even if they were
15assigned already (but not carried out). The runtime of SCHEDULE is said to benear real time (about a minute). Slomp et al. (1988) consider three quasi on-line procedures for schedulingFMSs. These procedures are essentially heuristic rules for the selection of a work-station, a transport device, and an operator. The selections are madehierarchically, and the three procedures differ in the way these selections areplaced in the hierarchy. In the Function Sequential Scheduling (FSS) procedure,the selections of workstation, transport device, and the operator are made for eachoperation sequentially. The Function Integrated Scheduling (FIS) makes all thethree assignments simultaneously. In the Function Phased Scheduling (FPS)procedure, the workstation assignments are completed first, in phase one; then,the transport device and operator assignments are made in phase two. When themakespan is used as the criterion, the SPT/TOT rule performed the best. Thisresult is the same as that of Stecke and Solberg (1981), although their criterionwas the production rate. Slomp et al. concluded that FPS performed worse thanFIS and FSS, and that FIS is to be favoured when there is heavy workload ontransport devices and operators, otherwise FSS is recommended. Co et al. (1988) describe an investigation of scheduling rules for FMS wherethey found that performance (mean flow time) of jobs is insensitive to somecommon dispatching rules so long as the FMS is loaded lightly (less than 2jobs/machine). Choi and Malstrom (1988) used a physical simulator to assessseveral dispatching rules. Wilhelm and Shin (1985) tested the efficacy of threelevels of alternate operations in FMS. Adaptive, dynamic control of alternateoperations was found the most effective. Denzler and Boe (1987) investigatedheuristic loading rules to decide on the part to be introduced next into an FMS.Very simple rules were found quite effective. Sabuncuoglu and Hommertzheim(1992) investigated dispatching rules in the context of AGV scheduling rules using
16discrete event simulation. The effectiveness of scheduling rules was demonstratedparticularly for higher utilization levels. Among the rules for selecting jobs bymachines, SPT performed well, while for selecting jobs for AGVs the rule of shortesttravelling distance (STD) and largest queue size (LQS) performed well. Co et al.(1990) have compared the performance of two machine selection heuristicscombined with three grouping heuristics from multiobjective points of view. Mukhopadhyay et al. (1991) have developed an integrated heuristic approachto tool allocation, parts scheduling, pallets scheduling, machine scheduling, andAGV scheduling. Priority rules and the analytical hierarchy process (Saaty 1980)are used to make a series of operating decisions. Heuristic rules are excellent for dynamic problems. Some of them, forinstance, SPT, have very little computational overhead, and still give good results.As discussed above, extensive evaluations of conventional dispatching rules arenow available in the context of FMS. There is much scope for developing andevaluating heuristics for other operational problems specific to FMS.2.4. Control theoretic approach Gershwin et al. (1986) present a control theoretic perspective on the produc-tion control aspects of FMS. Kimemia and Gershwin (1983) presented a closed loophierarchical formulation of the FMS scheduling problem. Akella et al. (1984)describe the performance of a simulated model of an actual facility using thishierarchical policy. A FMS is considered where parts are manufactured to meet acertain demand which could be varying over time. There is a penalty for exceedingthe demand as well as not meeting it. Thus it would be best to produce exactly atthe same rate as the demand; but this cannot be done on account of the failure ofthe machines. Stochastic machine failures are considered, which are smoothed byproviding buffers of the parts. The heart of this control theoretic scheduling policy
17is to maintain a steady safety buffer of the parts produced in the FMS, as long as itis feasible to do so. A characteristic of the framework is that it is constrained tofind a solution within the production capacity of the FMS. For each machine state,a capacity state can be defined which is the set of possible production rate vectors.For each machine state, a safety buffer level is defined for each part type. At anypoint in time, the production rate vector is found by solving a linear program tominimize the production costs. Their hierarchy is based on the frequency of events.Decisions about events of higher frequency is made at a lower level of hierarchy.Three levels of hierarchy are suggested. The frequency of events at a particularlevel is an order of magnitude smaller than that at a lower level. The top level ofthe hierarchy calculates the safety buffer levels for each machine state. At themiddle level, calculations need to be done more frequently. From the parametersgiven by the top level, the vector of cost coefficients is calculated, and the linearprogram is solved. This is to be done on-line. This results in a vector of productionrates. The lowest level of the hierarchy dispatches parts in such a way that theflow rates established at the middle level are achieved. A rigorous formulation of the above hierarchical framework is provided byGershwin (1989). The simulation results of Akella et al. show that theirhierarchical scheduling methodology produces high output with low work inprocess. It is able to track the demand on the system very closely while copingwith disruptions due to machine failure. As can be seen, the closed loop controlpolicy is tailored for a dedicated FMS producing a particular part mix. The toolingof the FMS, buffer capacity and other constraints are not considered. It isassumed that the input of a part is a sufficient control decision, and the (alternate)routing, possible deadlocks, blocking, etc. need not be considered. Further, thepossible effect of long total processing times of parts in the FMS on the feedbackloop is ignored.
18 Han and McGinnis (1989) present a discrete time control method for a FMScell. Their objective is to minimize the stockout cost under time-varying demandfrom downstream cells. A single-stage cell with one or more workstations workingin parallel is considered. Machine failures, limited buffer capacities, and varyinginputs from upstream cells are considered. The control scheme periodically solvesan optimization model to determine the flow of parts.2.5. Simulation based approach Recently some authors have presented discrete event simulation as ascheduling tool. Basically, simulation is proposed as a tool to evaluate thedispatching rules. This is not an entirely new approach: the study by Conway(1965, 1965b) was based on simulation. What is new is that the authors suggestusing data from the actual FMS for simulation. Thus a simulation model of thereal production system is built. The simulation model is initialized to the exactcurrent state of the factory. The dispatching rules are then tested on this model. Davis and Jones (1989) propose concurrent simulation to carry outproduction scheduling. In their scheme, multiple simulators of a productionfacility are initialized to the latest state of a FMS. These simulators are stoppedafter some time. The simulations are then analyzed as terminating simulations todecide on the best rule to use. Synergism between expert systems and simulation is used in an on-linescheduling system called ESS (Expert System Scheduler). Jain et al. (1989)describe the development of a scheduling system which communicates on-line withthe factory control system, generating schedules in real-time. The schedulingdecisions are based on the expertise of an experienced scheduler. The system isbased on LISP, and uses object-oriented concepts for both the expert systems andsimulation. It is possible to run the simulation backward in time to obtain starting
19time-windows for jobs. The major reason for implementing backward simulationwas implementation of JIT concepts. With this concept the job can be started atthe latest possible time. Conflicts are resolved by shifting individual jobs in theschedule forward or backward. The system reacts interactively with the user, andpermits solicitation of more information by the user, or changing of the schedule.At the time this article was written, the system had been controlling production atan automated manufacturing facility for several months. Wu and Wysk (1989) report on a multi-pass expert control system (MPECS)which uses discrete-event simulation for on-line control and scheduling in flexiblemanufacturing systems. In their system, simulation is used to evaluatedispatching rules. An expert system is employed to compile the set of candidatedispatching rules (Wu and Wysk, 1988). This expert system has a learning moduleto learn from past decisions. The expert system generates the candidate set on thebasis of current system objectives, system status, and the characteristics of on-going operations. A Flexible Simulation Mechanism (FSM) collects all the data onthe current system status. A simulation model is then generated based on thisdata. A series of simulation runs is carried out starting from the current stateusing each of the candidate dispatching rules for the next short time period (dt),selected by the user. FSM provides performance measures for each of the runs.The rule that results in the best performance is used to generate a series ofcommands to the real-time control system of the FMS. The FMS is then run fortime dt under the best dispatching rule.Compared to single-pass heuristic scheduling, Wu and Wysk report an improve-ment of 2.3%-29.3% under different simulation windows (= dt) and measures ofperformance. Selection among waiting jobs for operation in a machine is, however,just one of the decisions that need to be made on the shop floor. Although Wuand Wysks control system addresses flexible manufacturing, it is not clear how or
20if other decisions in FMS, e.g. routing selection, tool change, AGV selection, etc. arehandled in this system. Ishi and Talavage (1991) propose a time-series based algorithm fordetermining the length of the simulation window. This is done on the basis of thesystem state which is evaluated by a measure similar to the utilization of the FMS.Strategies are proposed to select a dispatching rule avoiding the problem ofcensored data with arbitrary simulation windows. Improvements in performancemeasures of up to 16.5% are reported. Simulation is certainly more tractable than mathematical programmingformulations of FMS operating problems. With simulation, there is no concernabout feasibility, since there is no need to make any unnecessary simplifyingassumptions. The simulation model can be built as close to reality as one needs to.Simulation can work as a decision support tool when there is the possibility tosimulate under different decision alternatives. When considered as a candidatesystem for on-line control, response time of the scheduling system is a major con-cern. The response time would also depend on the number of candidate rulesevaluated. This issue can only be resolved by further investigations into this newapproach.2.6. Artificial intelligence based approach Artificial intelligence (AI) appears to be particularly suited to solvingoperation problems of FMS because AI was developed to solve similar problems -problems involving a large search space, and where human expertise can findreasonable solutions pretty fast. Many researchers have sought to utilize thissimilarity. So far, two techniques of AI have found use in the FMS literature:Expert Systems and Planning. Expert systems attempt to emulate a humanexpert. Planning, also called problem solving, concerns itself with situations where
21there is a goal, and different actions have to be planned to achieve the goal. Steffen(1986) has presented a survey of AI based scheduling systems. These systemswere developed to schedule production systems, not necessarily a FMS. Kusiakand Chen (1988) have also reviewed a number of AI-based scheduling approaches.Many authors have written on use of AI in manufacturing (Bullers et al. 1980, Foxet al. 1982, Bourne and Fox 1984, Bensana et al. 1988; Chiodini 1986). Althoughthese concern themselves with scheduling production in general, they are relevantto FMS operation. Hall (1984) proposes use of if-then rules for process determination,sequencing and scheduling. However, no description of the system or the resultsobtained are given. Sauve and Collinot (1987) describe an object-oriented systemto represent FMS which produces daily off-line schedules using knowledge aboutconstraints and flexibility factors. This system also provides for on-line controlwhich analyses effects of disturbances upon the daily schedule and responds witha local modification of the schedule. Bruno et al. (1986) present a rule-based system to schedule production in aFMS. They use expert systems to capture knowledge about the domain, andqueuing network analysis for performance evaluation. The expert system usesrules to select production lots to introduce into the FMS. Primarily, the lots areselected on the basis of the dispatching rule of critical ratio. A lot with highestpriority may not be scheduled if a constraint is violated. Production constraintssuch as release time, needed fixtures, maintenance, etc. are checked. Capacityconstraints such as system congestion and throughput are checked by a heuristicbased on the mean value analysis of closed queuing network. This heuristiccalculates the machine utilization, average queue lengths, and lot throughputs. Asimulation model is used to obtain the system state trajectory using the rule baseand the performance analyzer. This trajectory is the resulting schedule. It is well
22known that mean value analysis calculates steady state performance. However, aFMS is a dynamic entity where the operating conditions are continually changed bythe very actions of the scheduler and by the vagaries of nature. Thus a concern isthe validity of the results of mean value analysis for use in decisions aboutproduction lot introduction. A nonlinear planning algorithm for FMS scheduling is proposed by Shaw(1988). This approach is based on the A* search, where one starts from an initialstate and by applying successive operators (from a rule base), the goal state isfinally reached. In this methodology, the jobs are individually scheduled using thissearch procedure. These schedules are not feasible, due to the simultaneouscontentions on the resources. A plan-revision procedure is used to resolve thecontentions. Shaw found that a) good heuristic knowledge is important forimproving the computation efficiency of the scheduling algorithm; b) a globalheuristic is better than a local heuristic; and c) a domain specific heuristic is betterthan a general heuristic. Unlike many other FMS scheduling methodologies, thismethodology explicitly considers alternate job routing, and incorporates it in theoptimization. Although it will use AI heuristics to limit the search, the searchspace is still very large and may make it prohibitively expensive to use in practicalscheduling problems. Park et al. (1989) describe a Pattern Directed Scheduler (PDS) which learnsthe selection of best dispatching rule from simulation. Simulation was performedunder varying combinations of FMS attributes such as buffer size, relative machineworkload, and machine homogeneity. The resulting mean tardiness was used todevelop a decision tree for selection of a scheduling rule. The performance of thePDS was found almost identical to that of the best dispatching rule. OGrady et al. (1987) have described highly centralized and highlydecentralized modes of intelligent control of FMS cells. OGrady and Lee (1988)
23have proposed a multi-blackboard/actor framework (PLATO-Z) for the control of aFMS cell. This system would then be part of a hierarchical control scheme of theFMS. PLATO-Z has four blackboards whose functions are: scheduling, operationdispatching, monitoring, and error handling. The blackboard system wasoriginally proposed in the HEARSAY-I speech understanding project (Barr andFeigenbaum, 1981). It has multiple knowledge sources (KS) , which are expertsystems, each with their own field of expertise. KSs are activated under specifiedconditions. A scheduler, which is itself a specialized knowledge source, sequencesthe different knowledge sources. These KSs work cooperatively to solve theproblem at hand. KSs communicate with each other through generally accessiblemessages - hence the name blackboard. Blackboard architecture based plannersare particularly suitable (Young 1988) for factory scheduling: 1) they can be drivenby external events posted on the blackboard; 2) independent knowledge sourceslend themselves to ease of modifications. The knowledge sources called on by theblackboard in PLATO-Z are not just rule-based. They could be heuristic algorithmsand optimizing procedures. The FMS is monitored in detail: part status, machine-status, material handling, buffer capacity. This approach is particularly attractivesince it supports a distributed control scheme. Chryssolouris et al. (1988) report on the performance of a decision-makingframework (MADEMA) as compared to traditional dispatching rules in a simulatedenvironment. MADEMA uses decision analysis techniques of determining thefeasible alternatives, determining relevant criteria, and determining theconsequences of the alternatives. It then uses rules to select the best alternative.The alternate routing is determined within the framework through a processplanning interface. The simulation results showed that MADEMA performed betterthan the best dispatching rule.
24 Kusiak (1986) presents a FMS scheduling system which uses a rule-basedExpert System. This system follows priority rules to schedule jobs normally, butwhen a job cannot be scheduled because of resource conflicts, decision tables areused to select alternative machines, tools, fixtures, material handlers. In order toresolve resource conflicts, Kusiak (1989) proposes a knowledge and optimization-based scheduling system (KBSS). KBSS has an inference engine that can drawupon a knowledge base, an algorithm (optimization) base, and a database. Chandra and Talavage (1991) describe a FMS where a part goes to a generalqueue after finishing an operation. When a machine is idle, it picks up a part fromthis queue using an intelligent dispatcher. This scheme gave better performancethan common dispatching rules. Maley et al. (1988) report on an object-orientedplanning module which can capture dynamic data, simulation information, andpast history to learn. It can also use optimization or heuristics toschedule/control an FMS. Bu-Hulaiga and Chakravarty (1988) present anotherobject-oriented framework which collects data in real-time from the factory floor,checks for variance from production targets, and suggests feasibility of re-tooling ifthere is a variance. So far, use of AI approach to FMS operation problems has addressed generalproblems, but restricted in size. AI techniques have shown good results fordomain-specific problems. The need exists for applying these techniques toparticular case-studies of FMS operations to determine the desirability andfeasibility of this approach. The classification of the literature based on the methodology followed is donein Table 1.
25 Table 1. Classification from the Methodology ViewpointMethodology PublicationMathematical Stecke 1983; Shanker and Tzen 1985; Kimemia andprogramming Greshwin 1985; Berrada and Stecke 1986; Sarin and Chen 1987; Lashkari et al. 1987; Avonts and Van Wassenhove 1988; Hwan and Shogun 1989; Shanker and Srinivasulu 1989; Wilson 1989; Hutchison et al. 1989; Jaikumar and Van Wassenhove 1989; Han et al. 1989; Ram et al. 1990; Co et al. 1990; Chen and Chung 1991Multi-criteria OGrady and Menon 1987; Lee and Jung 1989; Ro anddecision Kim 1990; Kumar et al. 1990makingHeuristics Nof et al. 1979; Stecke and Solberg 1981; Buzacott 1982; Iwata et al. 1982; Wilhelm and Sarin 1985; Shanker and Tzen 1985; Denzler and Boe 1987; Co et al. 1988; Choi and Malstrom 1988; Donath and Graves 1988; Slomp et al. 1988; Jaikumar and Van Wassenhove 1989; Chang et al. 1989; Chang and Sullivan 1990; Mukhopadhyay et al. 1991; Sabuncuoglu and Hommertzheim 1992Control Kimemia and Greshwin 1983; Akella et al. 1984; Hantheoretic and McGinnis 1989Simulation Wu and Wysk 1989; Davis and Jones 1989; Jain et al.based 1989; Ishi and Talavage 1991Artificial Bruno et al. 1986; Kusiak 1986; Sauve and Collinotintelligence 1987; OGrady et al. 1987; Shaw 1988; Chryssolouris et al. 1988; Wu and Wysk 1988; Maley et al. 1988; Bu- Hulaiga and Chakravarty 1988; OGrady and Lee 1988; Kusiak 1989; Park et al. 1989; Chandra and Talavage 1991;3. Application area of the research In the previous section, we considered the literature from the viewpoint of themethodological approach employed. Another perspective is that of the type oftargeted FMS. FMSs may be classified on the basis of their complexity (Dupont1982) or on the basis of the diversity of the machined parts (Rachamadugu andStecke 1989). The dedicated FMS problem assumes a fixed part mix. The part mix
26is selected from the total production requirement of the company. When themachines in the FMS are grouped, and loaded with the parts, the operation of theparts is allocated to the machines. Then until the production allocation is changedagain, the FMS is operated in the same way as a job shop since the allocation ofoperation and tooling of the machines is taken care of. If the parts visiting themachine are not selected in advance, the operations need to be allocated as theparts arrive and the machines need to be tooled correspondingly. This type of FMSis called random FMS. From the viewpoint of variety of parts handled, the FMSliterature may be classified broadly as being applicable to: 1. Dedicated FMS 2. Random FMS 3. Flexible Assembly Systems A flexible assembly system is limited to the assembly of very few producttypes. A dedicated FMS is configured to machine few pre-selected parts, whereasthe random FMS handles a wider variety of parts, its configuration (tool-mounting)changing as needed. Most of the early literature was focused on the part selectionproblem of dedicated FMS. There has been a wide interest in the loading problemof random FMS. A classification of literature on this basis is given in Table 2.4. Planning horizon Researchers have looked at the scheduling and control problems fromdifferent temporal viewpoints. Some have looked at the long-term planning of FMS,while others have addressed real-time issues of controlling FMS. The following is aconvenient taxonomy to classify the literature from this viewpoint. 1. Planning problems 2. Scheduling problems 3. Realtime control problems
27 Table 2. Classification on the Basis of Application Area Application Publication area Dedicated Nof et al. 1979; Stecke and Solberg 1981; Buzacott FMS 1982; Stecke 1983; Kimemia and Gershwin 1983; Akella et al. 1984; Kimemia and Gershwin 1985; Wilhelm and Sarin 1985; Berrada and Stecke 1986; Sarin and Chen 1987; Denzler and Boe 1987; Lashkari et al. 1987; OGrady and Menon 1987; Slomp et al. 1988; Avonts and Van Wassenhove 1988; Choi and Malstrom 1988; Lee and Jung 1989; Wilson 1989; Kumar et al. 1990; Ro and Kim 1990; Ram et al. 1990; Ishi and Talavage 1991; Chen and Chung 1991 Random Iwata et al. 1982; Shanker and Tzen 1985; Bruno et al. FMS 1986; Kusiak 1986; Sauve and Collinot 1987; OGrady et al. 1987; Shaw 1988; OGrady and Lee 1988; Co et al. 1988; Chryssolouris et al. 1988; Park et al. 1989; Kusiak 1989; Hwan and Shogun 1989; Han et al. 1989; Davis and Jones 1989; Hutchison et al. 1989; Jaikumar and Wassenhove 1989; Shanker and Srinivasulu 1989; Wu and Wysk 1989; Chang et al. 1989; Jain et al. 1989; Chang and Sullivan 1990; Co et al. 1990; Mukhopadhyay et al. 1991; Chandra and Talavage 1991; Sabuncuoglu and Hommertzheim 1992 Flexible Donath and Graves 1988; Graves 1988 Assembly System Planning problems are long term problems including loading, grouping,selection of parts for manufacturing in a FMS, etc. Most of the literature ondedicated FMS is on planning problems. Resource allocation problems withsmaller time horizon are the scheduling problems. Except for the heuristicapproaches, few authors have worked in this area. Still fewer authors have writtenon the real-time problem of dynamically controlling an FMS. Table 3 presents aclassification of literature on this basis.
28 Table 3. Classification on the Basis of Planning Horizon Time Publication horizon Planning Stecke 1983; Shanker and Tzen 1985; Berrada and problems Stecke 1986; Lashkari et al. 1987; OGrady and Menon 1987; Sarin and Chen 1987; Avonts and Van Wassenhove 1988; Hwan and Shogun 1989; Wilson 1989; Jaikumar and Wassenhove 1989; Lee and Jung 1989; Ro and Kim 1990; Ram et al. 1990; Kumar et al. 1990; Chen and Chung 1991; Co et al. 1991 Scheduling Nof et al. 1979; Iwata et al. 1982; Shanker and Tzen problems 1985; Bruno et al. 1986; Sauve and Collinot 1987; Denzler and Boe 1987; Shaw 1988; Co et al. 1988; Choi and Malstrom 1988; Chryssolouris et al. 1988; Kusiak 1986 and 1989; Shanker and Srinivasulu 1989; Hutchison et al. 1989; Jaikumar and Wassenhove 1989; Chang et al. 1989; Jain et al. 1989; Chang and Sullivan 1990; Chandra and Talavage 1991; Mukhopadhyay et al. 1991; Sabuncuoglu and Hommertzheim 1992 Realtime Stecke and Solberg 1981; Buzacott 1982; Akella et al. control 1984; Kimemia and Gershwin 1985; Wilhelm and Sarin problems 1985; Sauve and Collinot 1987; OGrady et al. 1987; OGrady and Lee 1988; Slomp et al. 1988; Donath and Graves 1988; Bu-Hulaiga and Chakravarty 1988; Davis and Jones 1989; Park et al. 1989; Han et al. 1989; Wu and Wysk 1989; Ishi and Talavage 1991;5. FMS factors considered There is great divergence in the literature in the type of FMS considered. Formost of the writers, the flexibility in routing seems to be the main feature of FMS.Many other authors have included the tool-slots of the workstations in theirdiscussions. Some authors have ignored both of these flexibilities. Similardiversity exists in the consideration of pallets, material transporters etc. Very fewauthors have considered all the facets of FMS simultaneously. Based on thisconsideration, Table 4 depicts a classification of the available literature.
29 Table 4. Factors Considered in the Literature Reference Route Tool Part Machine Buffer Pallets flexi- slots tran- avail- spaces bility sport abillityKimemia and Gershwin 1985; Y N N N N NWilhelm and Sarin 1985; Shaw1988; Chryssolouris et al. 1988;Donath and Graves 1988; Chang etal. 1989; Avonts and VanWassenhove 1988; Chandra andTalavage 1991Nof et al. 1979; Stecke and Solberg Y Y N N N N1981; Stecke 1983; Shanker andTzen 1985; Berrada and Stecke1986; OGrady and Menon 1987;Sarin and Chen 1987; Bu-Hulaigaand Chakravarty 1988; Hwan andShogun 1989; Han et al. 1989;Hutchison et al. 1989; Jaikumarand Wassenhove 1989; Shankerand Srinivasulu 1989; Jain et al.1989; Kumar et al. 1990; Ram et al.1990; Co et al. 1990; Chen andChung 1991Lashkari et al. 1987; Wilson 1989 Y Y N N N YDavis and Jones 1989; Ishi and N N Y N N NTalavage 1991Sauve and Collinot 1987 Y Y N N Y NBruno et al. 1986; Choi and Y N N Y N YMalstrom 1988Park et al. 1989 Y N N N Y NKusiak 1986 and 1989; Y Y Y N N YMukhopadhyay et al. 1991;Iwata et al. 1982; OGrady and Lee Y Y Y N Y N1988; OGrady et al. 1987Chang and Sullivan 1990; Y N Y N N NBuzacott 1982; Ro and Kim 1990 Y N Y N Y NSlomp et al. 1988 Y N Y N Y N
30Akella et al. 1984 N N Y Y N NDenzler and Boe 1987; Lee and Y N N N N YJung 1989Co et al. 1988; Wu and Wysk 1989 N N N N N NSabuncuoglu and Hommertzheim N N Y N Y N1992Han and McGinnis 1989 N N N Y Y N6. Conclusions for future research Great strides have been made in the scheduling and control literature ofFMS. There is now a mature literature using different methodological approaches.Future work needs to be done in investigating the use of the methodologies in thepractical arena, in making the control systems more user-friendly, and indeveloping more comprehensive control systems. FMS control problems are very complex and difficult. Rather thanattempting to get the optimum solutions of the problem formulations, researchshould be done on interactive scheduling and control of FMS where there is humaninput in the loop. Godin (1978) presents a review of interactive scheduling.Adelsberger and Kanet (1989) provide a more recent review of the state of art ininteractive scheduling. A decision support system approach including interactivescheduling has a lot of promise for application in the operations of FMS. Samadi etal. (1990), describe one such management tool that provides information as well assuggestions to help in operating a manufacturing system. Modern workstationsprovide a splendid opportunity for the development of FMS control decision supportsystems using the graphics capabilities, and underlying heuristics or rule-basedsystems. FMS is different things to different researchers. Quite often only thealternate operations aspect is emphasised. It is time to move on to furtherdeveloping comprehensive control schemes which take care of the complex
31interaction of the multiple resources in an FMS: transporters, CNC machines,robots, tools, fixtures, pallets. This could be done using hierarchical orheterarchical schemes. Discrete-event simulation is another area which has the potential to makemajor contributions to FMS operation. Simulation can be used to model FMS quitecomprehensively, and may be used to evaluate control policies, heuristics, andrules. Distributed processing makes the use of simulation feasible. There aresome published papers using a simulation approach, but usually they do notprovide comprehensive modelling of FMS.ReferencesAdelsberger, H.H., and Kanet, J.J., 1989, The Leitstand - a new tool for computer integrated manufacturing. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems, 253-258.Avonts, L.H. and Wassenhove, L.N., 1988, The part mix and routing mix problem in FMS: a coupling between an LP model and a closed queuing network. International Journal of Production Research, 26, 1891-1902.Akella, R., Choong, Y., and Gershwin, S.B., 1984, Performance of hierarchical production scheduling policy. IEEE Transactions on Components, Hybrids, and Manufacturing Technology, CHMT-7, 215-217.Barr, A.B., and Feigenbaum, E.A., 1981, The Handbook of Artificial Intelligence, Vol. 1, 343-348, (Los Altos, California: William and Kaufmann Inc.).Basnet, C.B. and Mize, J.H., 1992, An object-oriented framework for operating flexible manufacturing systems. Proceedings: International Conference on Object-Oriented Manufacturing Systems, Calgary, Canada, 346-351.
32Bensana, E., Bel, G., and Dubois, D., 1988, OPAL: A multi-knowledge-based system for industrial job-shop scheduling. International Journal of Production Research, 26, 795-819.Berrada, M., and Stecke, K.E., 1986, A branch and bound approach for machine load balancing in flexible manufacturing systems. Management Science, 32, 1316-1335.Bourne, D.A., and Fox, M.S., 1984, Autonomous manufacturing: automating the job-shop. IEEE Computer, 17, 76-86.Browne, J., Dubois, D., Rathmill, K., Sethi, S.P., and Stecke, K.E., 1984, Classification of flexible manufacturing systems. The FMS Magazine, April 1984, 114-1117.Bruno, G., Elia, A., and Laface, P., 1986, A rule-based system to schedule production. IEEE Computer, 19, 32-40.Bu-Hulaiga, M.I., and Chakravarty, A.K., 1988, An object-oriented knowledge representation for hierarchical real-time control of flexible manufacturing. International Journal of Production Research, 26, 821-844.Bullers, W.I., Nof, S.Y., and Whinston, A.B., 1980, Artificial intelligence in manufacturing planning and control. AIIE Transactions, 12, 351-363.Buzacott, J.A., and Shanthikumar, J.G., 1980, Models for understanding flexible manufacturing systems. AIIE Transactions, 12, 339-349.Buzacott, J.A., 1982, Optimal operating rules for automated manufacturing sys- tems. IEEE Transactions on Automatic Control, AC-27, 80-86.Buzacott, J.A., and Yao, D.D., 1986, Flexible manufacturing systems: a review of analytical models. Management Science, 32, 890-905.Carrie, A.S., and Petsopoulos, A.C., 1985, Operation sequencing in a FMS. Robotica, 3, 259-264.
33Chandra, J., and Talavage, J., 1991, Intelligent dispatching for flexible manufacturing. International Journal of Production Research, 29, 2259-2278.Chang, Y., and Sullivan, R.S., 1990, Schedule generation in a dynamic job shop. International Journal of Production Research, 28, 65-74.Chang, Y., Matsuo, H., and Sullivan, R.S., 1989, A bottleneck-based beam search for job scheduling in a flexible manufacturing system. International Journal of Production Research, 27, 1949-1961.Chen, Y.J., and Askin, R.G., 1990, A multiobjective evaluation of flexible manufacturing system loading heuristics. International Journal of Production Research, 28, 895-911.Chen, I.J., and Chung, C.H., 1991, Effects of loading and routeing decisions on performance of flexible manufacturing systems. International Journal of Production Research, 29, 2209-2225.Chiodini, V., 1986, A knowledge based system for dynamic manufacturing replanning. Symposium on Real-Time Optimization in Automated Manufacturing Facilities, National Bureau of Standards, Gaithesburg, Maryland.Choi, R.H., and Malstrom, E.M., 1988, Evaluation of traditional work scheduling rules in a flexible manufacturing system with a physical simulator. Journal of Manufacturing Systems, 7, 33-45.Chryssolouris, G., Wright, K., Pierce, J., and Cobb, W., 1988, Manufacturing systems operation: dispatch rules versus intelligent control. Robotics and Computer-Integrated Manufacturing, 4, 531-544.Co, H.C., Biermann, J.S., and Chen, S.K., 1990, A methodical approach to the flexible manufacturing system batching, loading, and tool configuration problems. International Journal of Production Research, 28, 2171-2186.
34Co, H.C., Jaw, T.J., and Chen, S.K., 1988, Sequencing in flexible manufacturing systems and other short queue-length systems. Journal of Manufacturing Systems, 7, 1-7.Conway, R.W., Maxwell, W.L., and Miller, L.W., 1967, Theory of Scheduling (Reading, Mass.: Addison-Wesley Publishing Company).Conway, R.W., 1965, Priority dispatching and work in process inventory in a job shop. Journal of Industrial Engineering, 16, 123-130.Conway, R.W., 1965b, Priority dispatching and job lateness in a job shop. Journal of Industrial Engineering, 16, 228-237.Davis, W.J., and Jones, A.T., 1989, On-line concurrent simulation in production scheduling. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems, 253-258.Donath, M., Graves, R.J., and Carlson, D.A., 1989, Flexible assembly systems: the scheduling problem for multiple products. Journal of Manufacturing Systems, 8, 27-33.Donath, M.W., 1988, A scheduling methodology for flexible manufacturing sys- tems. Unpublished Ph.D. Thesis, University of Massachusetts.Dupont-Gatelmand, C., 1982, A survey of flexible manufacturing systems. Journal of Manufacturing Systems, 1, 1-16.Fox, M.S., Allen, B., and Strohm, G., 1982, Job-shop scheduling: an investigation in constraint-directed reasoning. Proceedings of the National Conference on Artificial Intelligence, 155-158.Gere, W.S., 1966, Heuristics in job shop scheduling. Management Science, 13, 167-190.Gershwin, S.B., 1989, Hierarchical flow control: a framework for scheduling and planning discrete events in manufacturing systems. Proceedings of the IEEE, 77, 195-209.
35Graves, R.J., 1988, A hierarchical system for scheduling and near real-time control of material flow in flexible assembly. Proceedings, 1988 International Industrial Engineering Conference, Institute of Industrial Engineers, Norcross, Georgia.Gupta, Y.P., Gupta, M.C., and Bector, C.R., 1989, A review of scheduling rules in flexible manufacturing systems. International Journal of Computer Integrated Manufacturing, 2, 356-377.Hall, M.D., and Putnam, G., 1984, An application of expert systems in FMS. Autofact 6, Society of Manufacturing Engineers.Han, M.H., and McGinnis, L.F., 1989, Flow control in flexible manufacturing: minimization of stockout cost. International Journal of Production Research, 27, 701-715.Han, M., Na, Y.K., and Hogg, G.L., 1989, Real-time tool control and job dispatch- ing in flexible manufacturing systems. International Journal of Production Research, 27, 1257-1267.Hutchison, J., Leong, K., Snyder, D., and Ward, F., 1989, Scheduling for random job shop flexible manufacturing systems. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems, 161-166.Hwan, S.S., and Shogun, A.W., 1989, Modelling and solving an FMS part selection problem. International Journal of Production Research, 27, 1349- 1366.Ishi, N., and Talavage, J.J., 1991, A transient-based real-Time scheduling algorithm in FMS. International Journal of Production Research, 29, 2501- 2520.Iwata, K., Murotsu, A., Oba, F., and Yasuda, K., 1982, Production scheduling of flexible manufacturing systems. Annals of the CIRP, 31, 319-322.
36Jaikumar, R., and Van Wassenhove, L.N., 1989, A production planning framework for flexible manufacturing systems. Journal of Manufacturing Operations Management, 2, 52-79.Jain, S., Barber, K., and Osterfeld, D., 1990, Expert simulation for on-line scheduling. Proceedings of the 1989 Winter Simulation Conference, 930-935.Kimemia, J., and Gershwin, S.B., 1985, Flow optimization in flexible manufacturing systems. International Journal of Production Research, 23, 81- 96.Kimemia, J.G., and Gershwin, S.B., 1983, An algorithm for the computer control of production in flexible manufacturing systems. IIE Transactions, 15, 353- 362.Kumar, P., Tewari, N.K., and Singh, N., 1990, Joint consideration of grouping and loading problems in a flexible manufacturing system. International Journal of Production Research, 28, 1345-1346.Kusiak, A., 1985, Flexible manufacturing systems: a structured approach. International Journal of Production Research, 23, 1057-1073.Kusiak, A., 1986, FMS scheduling: a crucial element in an expert system control architecture. IEEE 1986 International Conference on Robotics and Automation, 653-658.Kusiak, A., and Chen, M., 1988, Expert systems for planning and scheduling manufacturing systems. European Journal of Operational Research, 34, 113- 130.Kusiak, A., 1989, KBSS: A knowledge and optimization based system for manufacturing scheduling. International Industrial Engineering Conference and Societies Manufacturing and Productivity Symposium Proceedings, 694- 699.
37Lashkari, R.S., Dutta, S.P., and Padhye, A.M., 1987, A new formulation of opera- tion allocation problem in flexible manufacturing systems: mathematical modelling and computational experience. International Journal of Production Research, 25, 1267-1283.Lee, S.M., and Jung, H., 1989, A multi-objective production planning model in a flexible manufacturing environment. International Journal of Production Research, 27, 1981-1992.Maley, J.G., Ruiz-Meir, S., and Solberg, J.J., 1988, Dynamic control in automated manufacturing: a knowledge integrated approach. International Journal of Production Research, 26, 1739-1748.Montazeri, M., and Van Wassenhove, L.N., 1990, Analysis of scheduling rules for an FMS. International Journal of Production Research, 28, 785-802.Moreno, A.A., and Ding, F., 1989, Goal oriented heuristics for the FMS loading (and part type selection) problems. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems, 105-110.Mukhopadhyay, S.K., Maiti, B., and Garg, S., 1991, Heuristic solution to the scheduling problems in flexible manufacturing system. International Journal of Production Research, 29, 2003-2024.Nof, S.Y., Barash, M.M., and Solberg, J.J., 1979, Operational control of item flow in versatile manufacturing systems. International Journal of Production Research, 17, 479-489.OGrady, P.J., and Menon, U., 1987, Loading a flexible manufacturing system. International Journal of Production Research, 25, 1053-1068.OGrady, P.J., Bao, H., and Lee, K.H., 1987, Issues in intelligent cell control for flexible manufacturing systems. Computers in Industry, 9, 25-36.
38OGrady, P., and Lee, K.H., 1988, An intelligent cell control system for automated manufacturing. International Journal of Production Research, 26, 845-861.Park, S.C., Raman, N., and Shaw, M.J., 1989, Heuristic learning for pattern directed scheduling in a flexible manufacturing system. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems, 369-376.Panwalker, S.S., and Iskander, W., 1977, A survey of scheduling rules. Opera- tions Research, 25, 45-61.Rachamadugu, R., and Stecke, K.E., 1989, Classification and review of FMS scheduling procedures. Working Paper #481 C, The University of Michigan, Ann Arbor, Michigan.Ram, B., Sarin, S.C., and Chen, C.S., 1987, A model and a solution approach for the machine loading and tool allocation problem in a flexible manufacturing system. International Journal of Production Research, 28, 637-645.Ro, I.K., and Kim, J.I., 1990, Multi-criteria operational control rules in flexible manufacturing systems (FMSs). International Journal of Production Research, 28, 47-63.Saaty, T.L., 1980, The Analytic Hierarchy Process (New York: McGraw-Hill).Samadi, B., Morris, R.J.T., Rubin, L.D., Wong, W.S., and Ekroot, B.C., 1990, The operations assistant: a new manufacturing resource management tool. Journal of Manufacturing Systems, 9, 303-314.Sauve, B., and Collinot, A., 1987, An expert system for scheduling in a flexible manufacturing system. Robotics and Computer-Integrated Manufacturing, 3, 229-233.Sabuncuoglu, I., and Hommertzheim, D.L., 1989, Expert simulation systems - recent developments and applications in flexible manufacturing systems. Computers in Industrial Engineering, 16, 575-585.
39Sabuncuoglu, I., and Hommertzheim, D.L., 1992, Experimental investigation of FMS machine and AGV scheduling rules against the mean flow-time criterion. International Journal of Production Research, 30, 1617-1635.Sarin, S.C., and Chen, C.S., 1987, The machine loading and tool allocation problem in a flexible manufacturing system. International Journal of Production Research, 25, 1081-1094.Shanker, K., and Tzen, Y.J., 1985, A Loading and dispatching problem in a random flexible manufacturing system. International Journal of Production Research, 23, 579-595.Shanker, K., and Rajamarthandan, S., 1989, Loading problem in FMS: part movement minimization. Proceedings of the Third ORSA/TIMS Conference on Flexible Manufacturing Systems. 99-104.Shanker, K., and Srinivasulu, A., 1989, Some methodologies for loading problems in flexible manufacturing systems. International Journal of Production Research, 27, 1019-1034.Shaw, M.J., 1988, Knowledge-based scheduling in flexible manufacturing sys- tems: an integration of pattern-directed interference and heuristic search. International Journal of Production Research, 26, 821-844.Slomp, J., Gaalman, G.J.C., and Nawijin, W.M., 1988, Quasi on-line scheduling procedures for flexible manufacturing systems. International Journal of Production Research, 26, 585-598.Smith, M.L., Ramesh, R., Dudek, R.A., and Blair, E.L., 1986, Characteristics of U.S. flexible manufacturing systems - a survey. Proceedings of the Second ORSA/TIMS Conference on Flexible Manufacturing Systems , 477-486.Stecke, K.E., and Solberg, J.J., 1981, Loading and control policies for a flexible manufacturing system. International Journal of Production Research, 19, 481-490.
40Stecke, K.E., and Solberg, J.J., 1982, The optimality of unbalanced workloads and machine group sizes for flexible manufacturing system. Working Paper No. 290, Graduate School of Business Administration, The University of Michigan, Ann Arbor, Michigan.Stecke, K., 1983, Formulation and solution of nonlinear integer production plan- ning problems for flexible manufacturing systems. Management Science, 29, 273-288.Steffen, M.S., 1986, A survey of artificial intelligence-based scheduling systems. Proceedings, Fall Industrial Engineering Conference, Institute of Industrial Engineers, Norcross, Georgia.Villa, A., Mosca, R., and Murari, G., 1986, Expert control theory: a key for solving production planning and control problems in flexible manufacturing. IEEE 1986 International Conference on Robotics and Automation, 466-471.Wilhelm, W.E., and Shin, H.M., 1985, Effectiveness of alternate operations in a flexible manufacturing systems. International Journal of Production Research, 23, 65-79.Wilson, J.M., 1989, An alternative formulation of the operation-allocation problem in flexible manufacturing systems. International Journal of Produc- tion Research, 27, 1405-1412.Wu, S.D., and Wysk, R.A., 1988, Multi-pass expert control system - a con- trol/scheduling structure for flexible manufacturing cells. Journal of Manu- facturing Systems, 7, 107-120.Wu, S.D., and Wysk, R.A., 1989, An application of discrete-event simulation to on-line control and scheduling in flexible manufacturing. International Journal of Production Research, 27, 1603-1623.
41Wu, S.D., and Wysk, R.A., 1990, An inference structure for the control and scheduling of manufacturing systems. Computers in Industrial Engineering, 18, 247-262.Young, R.E., and Rossi, M.A., 1988, Toward knowledge-based control of flexible manufacturing systems. IIE Transactions, 20, 36-43.