Search

In industrial sites, the running state of rolling bearings is often judged by hearing the sound with human ears. This method relies on long-term accumulation of human experience and is prone to cause occupational noise hazards. To solve this problem, this paper proposed an intelligent diagnosis method for the running state of bearings based on machine hearing. Firstly, bearing vibration signal was decomposed using an improved Variational Mode Decomposition (VMD) algorithm, by which the best mode component containing the fault characteristic was determined according to a time-frequency weighted kurtosis maximization criterion. Then, the time-domain feature indexes and psychoacoustic indexes of the best mode component were calculated to define a feature vector. Finally, the feature vectors were input into a fault classification model based on Support Vector Machine optimized by Grid Search Method (GSM-SVM) for training. The trained model was used to diagnose unknown faults of bearings. The proposed method was applied to the traction motors of EMU train for automatic bearing fault diagnose. Field test in the manufacturing factory showed that it could quickly diagnose bearing faults with a high accuracy rate.