RFR-GA-BLS: A Feature Selection and Parameter Optimization Method for Fault Diagnosis of Rolling Bearing Using Infrared Images

To overcome the problems of low machine learning fault diagnosis rate and long consumption time deep in rolling bearing diagnosis, an RFR-GA-BLS model is proposed. The validated by infrared images bearings to find most representative features, suitable parameters best diagnostic rate. Based on pre-processed thermal faulty bearing, 72 second-order statistical features were obtained as information for diagnosis. RFR considered robustness new sequences obtained. BLS was optimized GA New sequence added sequentially, one at a time. After satisfying conditions, appropriate number selected first 20. search results feature nodes, node windows enhancement nodes 24, 19 544, respectively, 98.8889% achieved. According comparison with CFR-GA-BLS, BLS, PSO-BLS Grdy-BLS, our proposed more advantageous performance. accuracy higher compared SVM RF. speed 207 times faster than 1DCNN 10,147 2DCNN.