Fault Diagnosis based on Evolutionary Algorithm and Support Vector Machines

Abstract

A new intelligent fault diagnosis (IFD) method based on evolutionary algorithm and support vector machines (SVM) for multivariate process monitoring was proposed. A hybrid method combining feature selection and generation in a wrapper based approach via evolutionary algorithm was proposed to automatically generate feature set, and SVM was proposed to serve as an inductive learner for the evaluation of the feature set both as a classifier for the whole diagnosis system. The whole diagnosis process is in a full-automatic way. First, training stage is carried out. Original data with known features was directly sent to the IFD system and then selected features together with generated features are determined by the evolution of SVM learner. Finally, test stage is on the way. Test feature sets are put into SVM classifier, and IFD outputs current fault patterns, which terminates the whole diagnosis process. Applications in TEP data sets prove this method effective and robust.