Data Flow Control of an Automated Job Shop: Are Dynamic Allocation and Scheduling Comparably Effective?

We investigated management of an automated job shop with an unpredictable job stream and randomly failing machines. The shop used automated materials handling to move jobs between automated machines, so that entering and removing jobs were the only manual operations, excluding maintenance and repair. The stochastic nature of the shop precluded a traditional scheduling approach. We devised a system based on dynamic allocation of machines to jobs under a distributed data flow control architecture. The data flow architecture reduced sensitivity to machine failures, and simplified dynamic allocation. It also simplified shop expansion. Unexpectedly, machine utilizations exceeded 93% for machine failure rates below 16% in a simulated shop with machine and job stream characteristics typical of contemporary Computer-integrated Manufacturing Systems (CMS). For this typical job shop, dynamic allocation produced utilizations comparable to those expected from scheduling. Machine utilization in contemporary job shops seldom exceeds 70%. Our paper describes the architecture and the experiments, and speculates on reasons for the high utilizations.