Multiple-Fault Diagnosis Method Based on Multiscale Feature Extraction and MSVM_PPA

Abstract EN

Identification of rolling bearing fault patterns, especially for the compound faults, has attracted notable attention and is still a challenge in fault diagnosis.

In this paper, a novel method called multiscale feature extraction (MFE) and multiclass support vector machine (MSVM) with particle parameter adaptive (PPA) is proposed.

MFE is used to preprocess the process signals, which decomposes the data into intrinsic mode function by empirical mode decomposition method, and instantaneous frequency of decomposed components was obtained by Hilbert transformation.

Then, statistical features and principal component analysis are utilized to extract significant information from the features, to get effective data from multiple faults.

MSVM method with PPA parameters optimization will classify the fault patterns.

The results of a case study of the rolling bearings faults data from Case Western Reserve University show that (1) the proposed intelligent method (MFE_PPA_MSVM) improves the classification recognition rate; (2) the accuracy will decline when the number of fault patterns increases; (3) prediction accuracy can be the best when the training set size is increased to 70% of the total sample set.

It verifies the method is feasible and efficient for fault diagnosis.