Identifying attributes for knowledge-based development in dynamic scheduling environments

Inductive learning techniques have shown inspiring success in reducing the effort for knowledge acquisiton in the development of knowledge-based scheduling systems. However, there is little research on selecting the proper attributes to facilitate development of knowledge bases and to enhance the generalization ability of resulting knowledge bases. In this study, the authors proposed a neural network based approach to identify the essential attributes for knowledge-based scheduling systems. Through the case study conducted, the proposed identification procedure is capable of selecting the essential attributes out of a given set of candidate attributes. Not only can this procedure consider several primary performance measures simultaneously, but also, by using a modularized representation, the previous results of attribute identification have the flexibility to cope with the new control strategies which may be introduced in the future. The results of experiments show that the scheduling knowledge bases developed by the set of selected attributes have superior ability to select the correct control strategy over other knowledge bases under new production conditions in a Mazak flexible manufacturing system.