Operational Decisions in AGV-Served Flowshop Loops: Fleet Sizing and Decomposition

This paper considers design and operational issues that arise in repetitive manufacturing systems served by automated guided vehicles (AGVs) in loop layouts with unidirectional material flow. Such systems are in widespread industrial use, and play an important role in modern manufacturing environments. The objective considered is the minimization of AGV fleet size, given the minimum steady state cycle time required to produce a minimal job set (or equivalently, given the maximum throughput rate). We also study whether the decomposition of a large AGV-served flowshop loop into several smaller loops improves productivity. The original loop and the decomposed design are compared with respect to the minimum cycle time needed for the repetitive manufacture of a minimal job set. When there are three or more machines in the loop, finding the optimal cycle time is an intractable problem. We therefore use the genetic algorithm developed in the companion paper [17], to identify whether the original or the decomposed design is more efficient. Our study suggests that many systems perform more productively as a result of decomposition. Finally, we discuss a joint sequencing issue that arises in decomposed systems with limited buffers between the loops, and analyze the tractability of all the relevant joint sequencing problems.     

Operational Decisions in AGV-Served Flowshop Loops: Scheduling

This paper considers operational issues that arise in repetitive manufacturing systems served by automated guided vehicles (AGVs) in loops with unidirectional material flow. The objective considered is the minimization of the steady state cycle time required to produce a minimal job set (or equivalently, throughput rate maximization). Our models allow for delays caused by AGV conflicts. We define and analyze three nondominated and widely used AGV dispatching policies. For each policy, we describe algorithms and intractability results for combined job scheduling and material handling problems. We describe a genetic algorithm that estimates the cycle time within 5% on average for instances with up to 10 machines and four AGVs. Some related fleet sizing and loop decomposition issues are discussed in the companion paper [19].