Evolutionary Optimisation of Production-Control Systems

This dissertation, "Evolutionary Optimisation of Production-control Systems" by Pik-yin, Mok, 莫碧賢, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: SUMMARY The production control of manufacturing systems is at the heart of the whole manufacturing process, and such control involves a series of real-time scheduling decisions regarding different discrete production operations. Thus, improving the effectiveness of production-control systems is of vital importance. The use of stochastic control theory to find the optimal or near-optimal solutions of production-control problems is greatly facilitated by using continuous-flow models to approximate the discrete flows in manufacturing systems. However, past research showed that such problems are very difficult to solve either analytically or numerically by such means. In order to simplify the problems, some restrictive assumptions (such as long-run constant demands and exponentially distributed machine failures) were made in this previous theoretical work, but even so analytical solutions were obtained only for simple systems containing single machines producing single product-types. Evolutionary algorithms are used in this thesis to develop practical solutions for these analytically intractable production-control problems in two different ways, namely, crisp-logic control and fuzzy-logic control. In the case of crisp-logic control, an evolutionary stochastic optimisation procedure is developed to estimate the optimal short-run inventory levels for finished-products and work-in-progress buffers (i.e., hedging points) for manufacturing systems with constant demands under closed-loop control. This methodology is illustrated by examples of manufacturing systems with different configurations (single machines, multiple machines, single product-types, and multiple product-types), and the results are validated by comparing them to the theoretically optimal i results (where such results are available). The evolutionarily optimised short-run inventory levels are also used to construct gain-scheduled adaptive controllers, and it is shown that such controllers can provide automatic closed-loop control for manufacturing systems with piecewise-constant demands. In the case of fuzzy-logic control, an evolutionary rule-induction design procedure is developed to optimise fuzzy-logic controllers to provide closed-loop control for manufacturing systems with constant demands. This methodology is illustrated by examples of manufacturing systems with different configurations, and the results are compared with the results obtained for manufacturing systems under crisp-logic control. Robust fuzzy-logic controllers are also developed for manufacturing systems with sets of different constant demands. The crisp-logic and fuzzy-logic control methodologies developed in this thesis are also used for the control of manufacturing systems producing prioritised multiple product- types. These evolutionary crisp-logic and fuzzy-logic control methodologies are finally applied to an industrial design example involving an actual semiconductor packaging assembly line, and the results thus obtained are found to be superior to those obtained in industrial practice. Three evolutionary algorithms, namely, genetic algorithms, non-adaptive evolution strategies, and adaptive evolution strategies, are used to optimise both crisp-logic and fuzzy-logic production-control systems in the various parts of this thesis, and their relative effectiveness is compared. It is concluded that, although adaptive evolution strategies do not always necessarily p