Analysis and productivity improvement of an automated material handling system through simulation

Abstract

As Amazon.com has grown, the company has invested in warehouse systems similar to those traditionally found in manufacturing. These systems are a complex and interconnected set of manned work processing stations linked together by conveyance. This thesis examines one such system, arranged as a set of parallel single piece workflow lines joined by a central computerized sorter that route work between lines, and examines the effect of sorter algorithms and line capacity on production output. Work studies, interviews, and a practical experiment suggest a deterioration in the feedback provided to the central sorter for work routing. A Monte Carlo simulation model of the system further supports this hypothesis, suggesting that system throughput is 11.7% lower than a system that perfectly routes work to line in a pull fashion. While perfect routing is not practically feasible, the thesis then explores two routing heuristics designed around starvation response and dynamic capacity analysis, which simulation suggest may yield a 6.73% increase in throughput. In doing so, the thesis provides a case study on process improvement using simulation to characterize a complicated mechanical production system that is otherwise analytically intractable.