A flexible auction-based shopfloor control paradigm for highly-distributed manufacturing systems

This dissertation presents a new approach to distributed shopfloor control that embodies multiple auction-based control protocols and allows the system to switch between these protocols in response to changes in the demand environment. The evaluation of this flexible auction-based distributed control paradigm is performed from the perspective of traditional manufacturing system performance measures as well as communication system measures using an experimental testbed based on parallel and distributed computing. There are two aspects to this research: (i) design of a flexible auction-based control scheme, and (ii) modeling, simulation, and evaluation of auction-based control systems. With regard to design of flexible auction-based control scheme, a dynamic-programming-based optimal bid construction scheme is developed and two auction-based control protocols (namely job-initiated and machine-initiated protocols) are investigated. A generalized model of job-initiated auction protocol is developed for distributed control of multi-stage processing of jobs. Finally, an inductive learning-based protocol switching mechanism is developed to enable the use of multiple auction protocols in a changing demand environment. With respect to the second aspect of this research, an agent-based modeling and simulation testbed using distributed computing (Parallel Virtual Machine) is implemented that enables empirical study of auction-based shopfloor control. Further, a two-phase performance evaluation methodology is presented for assessing distributed auction-based control schemes from the perspective of the communication system. The results of the research can enhance the adaptability of an auction-based shopfloor control system, and achieve reduced job flow time under a dynamic demand environment. The modeling and evaluation methodologies adopted in the research will also enable the rapid assessment and prototyping of the auction-based control systems.