Study of Data Fusion Method for Fault Diagnosis Based on FDR Feature Selection Algorithm and HMM/SVM Model

Abstract

To effectively avoid the loss of useful information, in this paper, we extract feature information from the fault signal of rotating machinery in different aspects such as amplitude-domain, time-domain and time-frequency domain. Then for the multi-dimensional feature extraction is prone to the problem of “dimension disaster”, introduce the principles of FDR in data mining to determine the classification ability of each individual feature, and introduce the cross correlation coefficient to solve the problem that dealing with individual feature neglects the interrelationship between the features, and construct a new feature level data fusion algorithm. Finally, According to the characteristics of the HMM (Hidden Markov model), SVM (Support Vector Machine) and its hybrid model, we construct a new decision-level data fusion model.