Manufacturing planning and operations optimisation for mass customisation manufacturing using computational intelligence.

Abstract

This study determined whether an Advanced Manufacturing System could be optimised, more effectively than by traditional methods, using new and novel computational intelligence techniques. An Advanced Manufacturing System can be described as highly automated and highly complex systems that strive for global competitiveness. In the context of this study, these systems aim to compete in a Mass Customisation Manufacturing market. Traditional optimisation methods refer to methods based on mathematical models, experience, or industry best practice. Computational Intelligence refers to computational methods inspired by natural systems and processes. This includes, but is not limited to, evolutionary intelligence, Artificial Neural Networks, swarm intelligence, and fuzzy systems. This study investigated the optimisation of the manufacturing system from both a planning and an operations perspective. Research was carried out to identify Computational Intelligence paradigms and algorithms for Advanced Manufacturing System planning and operations optimisation. Static and dynamic simulation models of an Advanced Manufacturing System, for the respective perspectives, have been developed in order to simulate a manufacturing system designed to produce a hypothetical range of customisable men’s wristwatches on a mass scale at a competitive cost. A new Biogeography-Based Optimisation algorithm was developed to optimise an aggregate production plan using static simulation models. This algorithm was implemented to find the lowest production cost for the wristwatch production system case study. This algorithm produced a lower cost plan than a Simulated Annealing algorithm with a lower impact on workforce. A new Distributed Dynamic Selection Rule Strategy was developed for optimising production scheduling using dynamic simulation models. This new strategy was inspired by the Harmony Search principle and was based on traditional selection rules for scheduling. This strategy was able to produce statistically significantly lower average order lead times than three out of four traditional selection rules tested.