A fuzzy rule-based methodology for dynamic kanban control in a generic kanban system
Abstract
This research studies the dynamic control of a production system. A particular production control scheme in flow shops, the generic kanban system, is under study. A generic kanban system that originates from the Toyota Kanban system is for production control under dynamic environments with variable demands and processing times. To ensure good production system performance, an appropriate number of kanbans employed in the system is essential. Dynamic kanban control using fuzzy rule-based systems is proposed to dynamically adjust the number of kanbans. The proposed methodology includes two major modules: example extraction and fuzzy system generation. Examples are generated from system simulation and the simulated annealing algorithm is adopted to extract desired examples. From desired examples, a general methodology to generate fuzzy systems is also proposed. We aim at generating fuzzy systems with good mapping ability and generalization ability as well. Experiments are conducted to evaluate the general methodology of generating fuzzy systems and the methodology of dynamic kanban control by using fuzzy rule-based systems, respectively. Experimental results show that the generated fuzzy systems have good system performance for iris data and data recorded from Sugeno and Yasukawa's work. Furthermore, fuzzy systems are generated for dynamic kanban control in a generic kanban system. Generated fuzzy systems are evaluated on the training process from the extracted examples and on the feasibility of applications in general situations. Finally, system performance obtained from the fuzzy system approach is compared with that from the simulated annealing approach under the same system status. Their performance does not differ much. The fuzzy system approach, however, can be applied more generally in situations where the simulated annealing approach does not work. This justifies the feasibility of the fuzzy system approach to dynamically control the number of kanbans in a generic kanban system.
Degree
Ph.D.
Advisors
Yih, Purdue University.
Subject Area
Industrial engineering|Operations research|Artificial intelligence
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.