Machine Fault Diagnosis Through a Recursive Optimal Pairwise Linear Discriminant Function Method

Abstract

Abstract A Recursive Optimal Pairwise Linear Discriminant Function (ROPLDF) method based on pattern recognition theory is developed for multiple class fault diagnosis. This approach does not need a priori failure probabilistic knowledge. Optimal pairwise linear discriminant function (OPLDF) is implemented to enhance the performance of the classification, and a recursive method is developed to allow the on-line updating of the coefficients of the OPLDF. Comparison results show that the proposed approach has better performance than other conventional approaches, such as the multiple linear discriminant function method and the feedforward neural network method. A tapping experimental result indicates that the proposed multiple class classification method is effective for practical use. The success rate of this approach for the tapping process condition diagnosis could be 100% with a trained diagnostic model.