Application of pattern matching techniques to example based diagnosis

Because of the demands of modern automated manufacturing systems and the serious implications of machine down time, there has been much attention focussed lately of the early diagnosis of faults. Many methodologies have been suggested and tested, however up until now little research on the applicability of new methodologies has been completed. Our research seeks to rectify this situation. This paper describes a diagnosis approach which uses a fault dictionary derived from detailed simulation modelling of a small electro-mechanical subsystem typically found in machine tool applications. Faults are then classified using one of the new methodologies and the performance evaluated. In this paper comparative results for a variety of methods including Neural Networks, Learning Vector Quantization and Statistical Data Analysis are presented.