An Intelligent Fault Diagnosis Method based on STFT and Convolutional Neural Network for Bearings Under Variable Working Conditions

The faults on rolling bearings, one of key components in various rotating machinery, are usually main source of many failures in these devices. It leads to many attentions by engineers and scholars who expect to accurately diagnosis their faults as early as possible to prevent chain accident. Many diagnosis methods are reported to process the cases under the constant speed or load, while the reality on this is often harsh and variable, which limits the accuracy of bearing diagnosis. To address this problem, an intelligent fault diagnosis model is put forward by combining the short-time Fourier transform (STFT) and the convolutional neural network (CNN), the former of which is used to transform the vibration signal in time domain to time-frequency domain and further forms inputs of the latter. Experimental data accumulated from six bearings under two conditions are applied to verify the effectiveness and accuracy of the diagnosis model. The damages on the bearing outer or inner race are actually generated during the accelerated life time tests and are still at the early stage, which are quite different from artificial damages and make the accurate diagnosis harder. Analyses and comparisons of the experiment results demonstrate the feasibility and higher diagnosis accuracy of the intelligent diagnosis model.