A Study of Sequencing Heuristics for PeriodicProduction Environments

This paper considers the identical jobs, single end-item cyclic scheduling problem (CSP) of determining good schedules for repetitive production systems with a exible manufacturing capability. The goal is to optimize system throughput and work in process inventory by determining the eecient trade-oo frontier between cycle length and ow times. We present a new, tighter integer programming formulation of CSP corresponding to a network ow problem with some non-network capacity constraints. Since CSP is computationally intractable, we consider several heuristic approaches some of which use subproblems which we have considered in previous research. The ve heuristics we examine fall into two generic categories: Myopic methods for schedule generation / improvement (the Graves et al method and a new push / pull heuristic) and truncated exponential search strategies (beam search / truncated branch and bound, simulated annealing algorithms and a new gambling procedure). Our computational tests comparing the heuristics indicate that the quick, myopic push / pull approach and the relatively slow, enumerative beam search generate reasonably good (near optimal) solutions which dominate the other methods.