Fault diagnosis via structural support vector machines

Abstract

Discriminative methods are becoming more and more popular on fault diagnosis systems, while they need additional strategies or multiple models to cope with the multiple classification problems. In this paper, we introduce the structural Support Vector Machines (structural SVMs) to fault diagnosis, which can indentify multiple kinds of faults with only one uniform discriminative model. We define error penalty function and select a proper kernel to make structural SVMs be appropriate for non-linear problem. Tennessee Eastman Process (TEP), a benchmark chemical engineering problem, is used to generate datasets to evaluate the performance of the propose method. Experiments show that the structural SVM reports a state-of-the-art performance on overlapping fault data and different fault type data.