REDECS ADAPTIVE SHIP MAINTENANCE RESCHEDULING October, 2001 RESIDENCE HOTEL UNITEN KAJANG PATHIAH ABDUL SAMAT (UPM) ALICIA TANG Y. C. (UNITEN) -- Presenter HAJAR MAT JANI (UNITEN) NOR’ASHIKIN ALI (UNITEN)
REDECS AGENDA (1) PROBLEM DEFINITION –WHAT IS THE PROBLEM? –OBJECTIVES BACKGROUND INFORMATION –WHAT HAD BEEN DONE? OUR APPROACH –CBR + GA –HOPFIELD Neural Network –Operational Research Framework
REDECS AGENDA (2) SOFTWARE CONCLUSION FUTURE WORKS
REDECS PROBLEM DEFINITION (1) Ships - assets in naval defence Ships - expensive They should be fully utilised High rate of availability is anticipated AVAILABILITY –depends on effectiveness of Preventive Maintenance Schedule (PMS) Unable to avoid rescheduling!!
REDECS PROBLEM DEFINITION (2) If (uncertainty) breakdowns occur –availability of ship is Low availability and high maintenance costs are problems in ship maintenance management This problem can jeopardise the defence system of the country
REDECS PROBLEM DEFINITION (3) SHIP MAINTENANCE (RE)SCHEDULING –is a process of deciding start-times of maintenance activities that satisfy all precedence and resource constraints & optimize the ship availability. variables domains constraints result
REDECS Objectives: Proposals –to develop Adaptive Algorithms to decide (select) which activity to reschedule –to develop Hopfield Neural N. to reschedule PROBLEM DEFINITION (4) Go There Click Me
REDECS MAINTENANCE SCHEDULE FOR A SHIP Factors –Running hours of the ships –Operational requirement –Status of parts availability –Status of operational defects –Dockyard availability
REDECS BACKGROUND INFORMATION (1) Scheduling / time-tabling problem –Neural Network –Constraints Logic Programming –Graph Coloring –Heuristics, etc E.g. ILOG, CHIP
REDECS BACKGROUND INFORMATION (2) CONSTRAINT SOLVING –Reduce search domain/space –therefore faster & save storage –how? It minimizes backtracking Solve problems: ‘design’, ‘diagnosis’ & ‘planning’ Build schedule that satisfies ‘temporal’ and ‘resource’ constraints
REDECS BACKGROUND INFORMATION(4) Improve G.A. by improving chromosome representation (increase ship availability) Achieved by search space (such as minimising overlapping of maintenance activity) WHAT HAD BEEN DONE? Table 1 overlapping Refer to articles 1 & 3, references section.
REDECS OUR APPROACH (1) USE GA – To “optimise” USE CBR – To find near optimum schedule that maximises availability Hybrid Vs just CBR
REDECS OUR APPROACH (2) TO RE-SCHEDULE: –USE HOPFIELD NN CONSTRAINTS NEURON –BASED ON CBR-GA DERIVED DATA 2 LAYERS Soumen and Badrul (1996) - rescheduling of power system Item#7
REDECS THE HYBRID G.A. ALGORITHM Step 1: code the start times and pattern of activitystart times and pattern of activity Step 2: create initial population Step 3: determine start times and pattern of activity by the GA Step 4: build feasible schedule using CBR Step 5: evaluate the schedule.
REDECS R. O. F R A M E W O R K N N N
REDECS SOFTWARE PLATFORM –Unix, Windows NT/ME/2000/9x PROPOSED LANGUAGE –C++ Used in previous works
REDECS Proposed Software Components Scheduling program Ship program (Solver) Constraints program G.A Maintenance program Many header files Adaptive scheduler Rescheduling using Hopfield Neural Net Keeps repeating until “fit” enough
REDECS G.A CBR G.A
REDECS Constraints Also constraintsNew Schedule
REDECS CONCLUSION (1) Re-design of existing algorithms is necessary. Therefore, new algorithms need to be developed. Reschedule of activities based on the temporal and resource constraints is required so as to adapt to the changes that may occur. Rescheduling Algorithms
REDECS CONCLUSION (2) CBR + G.A - to produce near optimum solution. Enhancement to be made to CBR. Hopfield Neural Network - to reschedule selected activities. Our solutions:
REDECS FUTURE WORKS Fuzzy Logic - to address “over constraints” of the selection of activities and the rescheduling process. Application in other areas: School time-tabling, Financial control and planning, Classification & Prediction.
REDECS THE END Thank You. Questions?
REDECS Improve Chromosome Representation less higher
REDECS Schedule Overlapping Overlapping!!
REDECS CBR Vs. Hybrid Comparison between the CBR and the hybrid approaches: ApproachesObjective function (minimising no. of overlapping activities) CBR alone CBR+GA0.98 CBR alone CBR+GA0.82 Class A Class B
REDECS Pattern activities and start-time An allele Combination of no. of activities + duration of operation Refer to figure 2, full paper
REDECS Values of GA parameters for Ship Class A No. of population = 45 No. of generation = 60 Probability of mutation = 0.01 Type of crossover = single-point Type of GA = steady state Size of chromosome = 4 Size of allele = 96 Fitness function = maximise availability Scaling = Linear scaling