Research on rolling bearing fault diagnosis based on MDFF and DCNN-SVM hybrid network

In view of the large number of fault types and obvious uncertainty of rolling bearings, the single signal collected often contains various redundant information and is easy to be interfered by noise signals. In this paper, based on multi-domain feature fusion, MDFF) and DCNN-SVM for rolling bearing fault diagnosis. By collecting bearing vibration signal from multiple sensors, feature extraction is carried out by time domain, frequency domain and complete empirical mode decomposition of noise set, and the sensitive features are screened by random forest algorithm. The feature dimension is reduced, and the optimized sensitive feature values are respectively input into the DCNN network for adaptive feature extraction. DCNN network was used to change the weight value of each sensitive feature quantity, and comprehensive training was carried out to obtain the multi-domain fusion feature quantity, which was input into the support vector machine for fault diagnosis. By setting multiple groups of comparison tests, it can be seen that the recognition accuracy of the proposed method is 96.82%, which is 19.95% higher than that of artificial SVM. It can effectively realize the comprehensive diagnosis of rolling bearing fault state, and has certain application value.