Fault State Recognition of Rolling Bearing Based Fully Convolutional Network

To solve the problem of determining the fault damage of rolling bearings, a fault diagnosis method for intelligent classification of vibration signals with different fault locations and different damage degrees is proposed. First, the research object is the laboratory dataset. By transforming into spectrograms, this can preserve the original information of the time-domain signal to a greater extent. Then, we use a deep, fully convolutional neural network to train the dataset. It has a rapid convergence and the accuracy is up to 100%. Second, in order to verify the correctness of the model, we take the service data on the real line as the research object, and the accuracy rate is as high as 99.22%. Compared with some other machine learning algorithms, our method boasts better generalization capability and accuracy and could be applied to practical engineering.