Feature Extraction for Fault Diagnosis Utilizing Blind Parameter Identification of MAR Model

摘要

This paper proposes a novel feature extraction method with the combination of variational mode decomposition (VMD) and multiple autoregressive model (MAR) for the fault diagnosis of rolling element bearings. With the collected fault signals, VMD is firstly applied to decompose the fault signal into some components, after which MAR model is built with all the components and its order is determined by Schwartz Bayes Criterion. Then, all parameters within the MAR model are identified blindly through QR decomposition. Finally, principal component analysis is applied to extract the key features, with which support vector machine is trained and used to recognize the fault mode. In order to assertain the effectiveness of the proposed method, engineering experiment and comparative analysis are conducted. The experimental result shows the superiority of the proposed method.