Scheduling and layout in flexible manufacturing systems
This thesis covers a variety of inter-related scheduling and layout issues encountered in flexible manufacturing systems. The principal focus is upon systems which adopt the commonly implemented loop layout configuration. A pivotal idea behind the work is that products must revisit machines during their manufacture. The work encompasses both computational and theoretical results. The computational work consists of testing both new and standard heuristic and local search techniques on two strongly NP-hard combinatorial optimisation problems, one related to layout and the other to single machine scheduling. In the layout problem, machines must be sequenced around a loop of conveyor belt with the objective of minimising the amount of movement carried out by the worst affected product type within the manufacturing system. In the single machine scheduling problem, coupledoperation jobs must be scheduled so that the maximum completion time on the machine, the makespan, is minimised; each coupled-operation job consists of two arbitrary processing time operations separated by a time lag that is bounded both below and above. Our results suggest that local search techniques, while well suited to standard search spaces, do not perform well when the search space contains infeasible neighbours and the cost of evaluating candidate solutions is high. The theoretical work stems from the repetitive manufacture of single product types in loop layout flexible manufacturing systems. We demonstrate that in such a system, the efficiency is strongly governed by the balance of workloads on the machines. We develop mixed integer programming models for tool allocation and machine sequencing with the objective of balancing workloads, and as a secondary criterion, minimising product movement. Focusing on a single machine, we study the computational complexity of a class of coupled-operation scheduling problems; each job consists of two unit processing time operations separated by a fixed delay of two units of time, and their order of production is restricted by precedence constraints. For several standard scheduling objective functions, we provide either polynomial algorithms or proof of NP-hardness for parallel chain and tree precedence constraints.