A scheduling methodology for flexibile manufacturing systems using adaptive fuzzy logic and genetic algorithms

Abstract

Production scheduling for a flexible manufacturing environment as found in many factory models today must satisfy multiple and often conflicting criteria. Practical considerations coupled with the customer service orientation prevalent in modern organisations require goods to be delivered on time whilst still maintaining efficient use of resources. The multi-criteria problem is difficult to solve using existing heuristics as they are typically singular in their objective requiring unrealistic assumptions to be made concerning the manufacturing system. Conversely, traditional opttm.1sation techniques cannot be applied to flexible manufacturing systems due to the complexity of the problem domain. Consequently, the time and computing power required to solve to optimality precludes a global optimisation. However, where the solution is of very high value, then an optimisation approach can still be justified. Recently, solution techniques belonging to the class of meta-heuristics have attracted considerable interest and greater acceptance amongst researchers as both valid problem solving techniques and also being commercially viable. Among these are fuzzy logic and genetic algorithms. The former is able. to generate solutions of high quality although optimality cannot be guaranteed. Genetic algorithms are an evolutionary search technique that more often than not is able to converge upon an optimum solution far more rapidly than conventional search methods. Used on its own, a genetic algorithm may converge upon a local optima rather than the global optimum. However, if the starting point for the search routine can be located in the region of the global optimum then the likelihood of converging upon it is much improved. Neither technique is perfect in all respects but used together they offer considerable advantages over other existing techniques. The use of fuzzy logic proved to be an ideal way to constrain the problem domain while the genetic algorithms were able to be initialised with solutions of already high quality, improving their effectiveness markedly. The computational overhead of either meta-heuristic is minimal by comparison with traditional optimisation techniques and the solution quality is consistently high. Visibility to the end user is vastly improved through the use of a linguistic rule base that captures the decision logic exhibited by human schedulers and common sense business rules. Consequently, the expertise required to progress from conception to implementation, is generally more in the field of systems engineering than mathematics - an important consideration given the organisations and industries that manufacturing scheduling applications target. This dissertation details the development and implementation (using a computer based virtual factory modelled on a real manufacturing plant) of a novel scheduling solution combining adaptive fuzzy logic with genetic algorithms: Development proceeded in three stages: the initial fuzzy logic based scheduling application, addition of adaptability, and finally incorporation of genetic algorithms to refine the solution further. The low computational load presented by this approach allows real time operation. Results from simulation models indicate that the solution technique proposed offered considerable advantages over conventional scheduling heuristics as used predominantly in industry. Incorporating adaptability ensured the long term applicability and extensibility of the solution. Whilst not a commercial software project per se, considerable care was taken to ensure that the solution proposed was able to be integrated into existing factory information systems. Much of the development necessary for a real implementation was undertaken throughout the research to verify the feasibility of the solution. Future avenues for research in this field would include the investigation of more complex genetic algorithm .routines or perhaps combining fuzzy logic with other search methods such as a tabu search. The current combination of fuzzy logic with genetic algorithms offers a good compromise however, between optimal performance and suitability to real industrial implementation.