Manufacturing Systems By Djamila Ouelhadj and Sanja Petrovic Okan Dükkancı 02122013 Introduction Dynamic environments with inevitable unpredictable real time events Machine failures ID: 388041
Download Presentation The PPT/PDF document "A Survey of Dynamic Scheduling in" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
A Survey of Dynamic Scheduling in
Manufacturing SystemsBy Djamila Ouelhadj and Sanja Petrovic
Okan Dükkancı
02.12.2013Slide2
Introduction
Dynamic environments with inevitable unpredictable real time events;Machine failures Arrival of urgent jobsDue date changes Feasible schedules become infeasible Scheduling Theory vs. Scheduling PracticeVery little correspondence between these two (Shukla and Chen, 1996)Slide3
Introduction
Dynamic SchedulingThe problem of scheduling in the presence of real-time eventsImplementation to the real-world scheduling problemsDynamic Scheduling in manufacturing systems Handling the occurrence of real-time events Slide4
The Dynamic Scheduling Problem
Several manufacturing systems;Single and Parallel Machines, Flow and Jobs Shops, Flexible Manufacturing Systems Real time events;Resource-related; Machine breakdowns, operator illness, unavailability or tool failures, loading limits, defective materials, etc. Job-related;Rush jobs, job cancellation, due date changes, change in job priority and processing time, etc. Slide5
The Dynamic Scheduling ProblemSlide6
The Dynamic Scheduling Problem
Completely Reactive SchedulingNo firm scheduling in advanceScheduling decisions made locally in real-timePriority dispatching rulesQuick, intuitive and easy to implementLower shop performancesSlide7
The Dynamic Scheduling Problem
Predictive-Reactive SchedulingMost common dynamic scheduling approachSchedules are revised after real-time eventsDeviation from the original schedule affects other activitiesRobust predictive-reactive schedulingMinimize the effect of disruption on the performance measure valueConsider both shop efficiency and deviation from the original schedule (stability) at the same timeSlide8
The Dynamic Scheduling Problem
Robust Predictive-Reactive SchedulingA bi-criterion robustness measure for single machineMachine breakdowns Minimize of makespan and impact of the schedule change (stability)StabilityDeviation from the original job starting timeDeviation from the original sequenceStability can be increased with almost no effect on makespanSlide9
The Dynamic Scheduling Problem
Robust pro-active schedulingPredictive schedulesMain difficulty is the determination of the predictability measureMehta and Uzsoy (1999)Single machine, machine breakdowns, minimize the max. latenessThe effect of disruption measured by deviation of the job completion timeThe deviation is reduced by inserting idle time in the predictive scheduleSignificant improvement in predictability with very little effect on the max. latenessSlide10
Rescheduling in the Presence of Real Time EventsSlide11
Rescheduling in the Presence of Real Time EventsSlide12
Rescheduling in the Presence of Real Time Events
Scheduling StrategiesSchedule RepairLocal adjustment of the current schedulePotential savings in CPU time and stability of the system Complete Rescheduling New schedule from the scratch Optimal solution can be obtained But, rarely practical and very high CPU time Also, instability and shop floor nervousness Schedule Repair is most common strategySlide13
Rescheduling in the Presence of Real Time EventsSlide14
Rescheduling in the Presence of Real Time Events
Rescheduling TimePeriodic PolicySchedules made at regular intervalsSeries of static problemsMore schedule stability and less schedule nervousnessA real-time event just after rescheduling can create some problemsDetermining the rescheduling period is very important Muhlemann et al. (1982)Job shop environment
with
processing
time
variations
and
machine
breakdowns
At
each
rescheduling
period
, a
static
schedule
is
generated
by
using
dispatching
rules
Increasing
the
rescheduling
period
decreases
the
performanceSlide15
Rescheduling in the Presence of Real Time Events
Rescheduling TimeEvent driven PolicyRescheduling after the real-time eventsMost common policyVieria et al. (2000a, 2000b)Comparison between periodic and event driven policies on single and parallel machinesLower rescheduling frequency decreases the number of set-ups, but higher rescheduling frequency reacts more quickly to disruptionsSlide16
Rescheduling in the Presence of Real Time Events
Rescheduling TimeHybrid PolicyCombination of periodic and event driven policyRescheduling made periodically except the occurrence of real-time events Church and Uzsoy (1992) Rescheduling periodically Regular events are ignored
After an urgent events, complete rescheduling
When the length of rescheduling period increases, the performance of periodic scheduling decreases
.
Event driven method works well Slide17
Dynamic Scheduling TechniquesSlide18
Dynamic Scheduling Techniques
HeuristicsSchedule repair methods, not guarantee the optimal scheduleMost common; right-shift schedule repair, match-up schedule repair and partial schedule repairRight-shift (RS) schedule repair; the remaining operations are shifted forwards in time by the amount of disruption timeMatch-up (MU) schedule repair; rescheduling approach to match-up with the pre-schedule at some point in the futurePartial schedule repair; rescheduling only the operations in failure Dispatching rules are heuristics for completely reactive schedulingSlide19
Dynamic Scheduling Techniques
HeuristicsYamamoto and Nof (1985)RS heuristic outperforms dispatching rules with complete rescheduling Abumaizar and Svetska (1997) Partial Schedule Repair vs. Complete Rescheduling vs. RS Schedule Repair in terms of efficiency and stability Partial Schedule Repair decreases deviation and computational complexity compared to complete rescheduling and right shifting
Bean et al. (1991)
MU
Schedule Repair provides near optimal solutions and higher predictability than complete reschedulingSlide20
Dynamic Scheduling Techniques
HeuristicsNof and Grant (1991)Rerouting the jobs to alternative machines, job-splittingDispatching Rules No rule performs well for all criteria Ramasesh (1990) and Rajendran and Holthaus (1999) Classified these rules as; rules involving processing times, rules involving due dates, simple rules involving neither processing times nor due dates, rules involving shop floor conditions,rules involving two or more of the first four categoriesSlide21
Dynamic Scheduling Techniques
Meta-HeuristicsHigh level heuristics that guide the local search heuristic to escape from local optimaTabu search (TS), Simulated Annealing (SA) and Genetic Algorithms (GA)Dorn et al. (1995)Tabu search to repair a schedule Zweben et al. (1994) Simulated annealing to repair schedules Slide22
Dynamic Scheduling Techniques
Meta-HeuristicsChryssolouris and Subramaniam (2001) Genetic algorithms for dynamic scheduling of manufacturing job shops Two performance measures; mean job tardiness and mean job cost Performance of genetic algorithm is better than the common dispatching rules Wu et al. (1991, 1993) Genetic Algorithms vs. Local Search Heuristics to generate robust schedules Genetic algorithm outperforms local search heuristic in terms of makespan and stability. Slide23
Dynamic Scheduling Techniques
Multi-Agent Based Dynamic SchedulingCentralized Scheduling SystemHierarchical Scheduling System Scheduling decision made centrally at the supervisor level and executed at the resource level Central computer has responsibility for;
scheduling,
dispatching resources,
monitoring any deviation
dispatching corrective actions Slide24
Dynamic Scheduling Techniques
Drawbacks of Centralized and Hierarchical Scheduling SystemsExistence of one central computer; bottleneck of the systemModification of configuration is expensive and time consuming Latency time of decision-making; late response to the real-time events In highly dynamic environment, centralized and hierarchical scheduling systems are inefficient Decentralize the control of the manufacturing systemReducing complexity and costIncreasing FlexibilityEnhancing Fault Tolerance Slide25
Dynamic Scheduling Techniques
Multi-Agent Systems in Dynamic SchedulingLocal autonomous agents carry out local schedules that increases the robustness and flexibilityDynamic interaction and cooperation between agentsShorter and simpler software compared to centralized approach Slide26
Dynamic Scheduling TechniquesSlide27
Dynamic Scheduling Techniques
Autonomous ArchitecturesAgents representing manufacturing entities such as resource and jobsGenerating local schedules and react locally to local disruptionsCooperating with each other for global optimal and robust schedules Slide28
Dynamic Scheduling Techniques
Goldsmith and Interrante (1998), Oeulhadj et al. (1998, 1999, 2000)Simple multi-agent architecture with only resource agentsAgents are responsible for dynamic local scheduling of the resourcesThey negotiate with each other via “contract net protocol” to generate global schedule Each agent performs; SchedulingDetectionDiagnosisError Handling Slide29
Dynamic Scheduling Techniques
Sousa and Ramos (1999)Multi-agent architecture with job and resource agentsJob agents negotiate with resource agents for the operation of job via “contract net protocol”When a disruption occurs;Resource agent sends a machine fault message to job agentsJob agents renegotiate the other resource agents in order to process the operations in failure Sandholm (2000) Instead of “contract net protocol”, “levelled commitment contracts” are used Decommiting from the contract by paying the penalty Slide30
Dynamic Scheduling Techniques
Mediator ArchitecturesWith large number of agents, autonomous architectures have some difficulties;Providing globally optimal schedulesPredictability Mediator architecture combine;RobustnessOptimalityPredictability Mediator outperforms autonomous due to ability to plan further in the future
ability to react disturbancesSlide31
Dynamic Scheduling Techniques
Mediator ArchitecturesAdditional to local agents of autonomous architecture, mediator agentCoordinate the local agentsContribute to same decision making processOverview of the entire systemLocal agents deals with the reaction to disruptionMediator agents improve the global performanceSlide32
Dynamic Scheduling Techniques
Ramos (1994)Mediator architecture consists of;Task AgentsTask Manager Agents,Resource AgentsResource Mediator AgentsTask manager agent creates task agentsThe resource mediator agent negotiates with resource agents for execution of tasks via “contract net protocol”When a disruption occurs;Messages are sent to the resource mediator agentThe resource mediator agent renegotiates with other resource agents Slide33
Dynamic Scheduling Techniques
Sun and Xue (2001)Mediator reactive scheduling architectureTwo mediators;Facility MediatorPersonnel MediatorMatch-up rescheduling strategy and agent based mechanism are used to repair only part of the scheduleSlide34
Dynamic Scheduling Techniques
Other Artificial Intelligence TechniquesKnowledge-based systems, neural networks, case-based reasoning, fuzzy logic, Petri nets, etc. Knowledge-based systemsVariety of technical expertise on the corrective action to undertakeLa Pape (1994)SONIA; a knowledge-based job-shop predictive-reactive scheduling systemSchedule repair heuristics; Relaxing due datesExtending work shiftsOperation postponed until the next shiftReduction of idle times of resources by permuting operationsSlide35
Dynamic Scheduling Techniques
Hybrid Systems combines various artificial intelligence techniquesDorn (1995)Case-based reasoning and fuzzy logic for reactive scheduling Garetti and Taisch (1995) and Garner and Ridley (1994) Knowledge-based systems and neural networks in reactive schedulingSlide36
Comparison of Solution Techniques
Heuristics;Widely used due to their simplicityCan be stuck in poor local optima Meta-heuristics; SA and TS are more efficient to find a near-optimal solutions in a reasonable time compared to GA Knowledge-based systems are limited by the quality and integrity of the specific domain knowledgeSlide37
Comparison of Solution Techniques
Centralized and Hierarchical Manufacturing SystemsGlobally better schedulesProblems with the reactivity to disturbance Multi-agent Systems Decentralize the control of manufacturing system Localize the scheduling decisions Sandholm (2000): Agents can locally react to local changes faster than centralized system couldProviding an architecture that is reliable, maintainable, flexible, robust and stableSlide38
Comparison of Solution Techniques
Autonomous vs. Mediator ArchitecturesAutonomous; cost-efficient, flexible and robust against disturbancesSuitable for system with a small number of agentsBut, providing globally optimized performance is questionableThe behaviour of the system is unpredictable with a large number of agents Mediator; improve performance compared to autonomous in complex manufacturing systems Combining robustness against disturbances with global performance optimization and predictabilitySlide39
Conclusion
Most manufacturing systems operate in dynamic environmentDynamic scheduling;Predictive-reactive schedulingRobustnessSchedule Repair Local adjustments Savings in CPU time and the stability of the systemMulti-agent Systems Very promisingIntegrated Systems; OR and AI for robustness and flexibilitySlide40
Any Questions/Comments?