A manufacturing problem solving environment combining evaluation, search, and generalisation methods

This paper develops a general environment for suggesting good operating strategies for specific factory conditions at the time the strategies are needed. The characteristics of the problems addressed do not allow analysis of the alternatives at the time the suggestions are needed. This requires the analysis to be done beforehand. However, by performing the analysis before the suggestions are needed, the future factory condition is unknown. With a large number of possible factory conditions, it is not possible to analyse all the possible states beforehand. We develop an environment that characterises this problem in terms of search, evaluation, and generalisation. This environment is characterised by several components working together. To aid understanding of the tasks of each component, we characterise their actions in terms of vectors and spaces. To demonstrate the operation of this environment we choose specific search and generalisation techniques and apply the environment to a specific factory problem. We will discuss the choice of these methods and how they work together, the results of a specific application, and a discussion of further extensions. The methods used in the test problem are: discrete event simulation, genetic algorithm (GA) search, and neural network generalisation. We will also point out where recent work by others has addressed segments of the problem presented and where these efforts fit in the proposed structure, and how current methods of knowledge extraction and data mining relate to this model.