Manufacturing control with a market-driven contract net

This dissertation presents four requirements for a next-generation manufacturing computer control architecture: (1) to reduce lead-times and work-in-progress inventories; (2) to interactively link the customer to the current and future capabilities of the shop-floor; (3) to make manufacturing decisions automatically using detailed accounting-system-provided costs; and (4) to implement a fully distributed, parallel, or 'heterarchical' architecture. This dissertation describes an architecture which was designed to meet these requirements, and provides initial case-study-based performance results which indicate that these requirements can indeed be met. This 'Market-Driven Contract Net' manufacturing computer-control architecture is described, primarily by the messages it passes over a communications network. The case study presents a detailed model of a GE job-shop--the modeled shop produces thousands of different parts a year. This case study demonstrates the feasibility and potential performance of fully distributed, production-reservation, and dollar-cost-based scheduling in a job-shop production planning and control system. Feasibility and potential performance are also shown through complexity, dynamic, and optimality analyses. For these analyses an automata-based extension to SYSREM software engineering graphs is used. These automata have relationships to other discrete-event-system models, such as: concurrent finite-state machines, Petri nets, Markov chains, and queueing networks. They are shown to be useful for design, concise yet complete description, comparison, and analysis of general decentralized and distributed discrete-event computer-controlled systems. They are compared to formal language and mathematical logic models. Using these models, the complexity of the fully distributed Market-Driven Contract Net scheduling algorithm is shown to be ${\cal O}\ \{nm\}$. Dynamic behavior is analyzed using Kelly's queueing analysis applied to SYSREM-automata models: this analysis shows a good correlation to the results obtained in the case study, and is applicable to the analysis of general discrete-event control algorithms. Optimality is determined in terms of flow-time (lead-time), tardiness, and cost. This dissertation parallels other research in: decentralized control theory for large-scale systems, command control, MRP II systems, just-in-time manufacturing systems (JIT), parallel scheduling algorithms, hierarchical and non-hierarchical manufacturing computer-control architectures, distributed problem solving in Artificial Intelligence, object-oriented programming, and electronic data interchange (EDI).