Gearbox fault diagnosis method based on deep convolutional neural network vibration signal image recognition

Nowadays, the internal structure of the gearbox tends to be complicated and the working environment is subject to more interference factors, so that the collected vibration signal is rich in more interference items, which makes the fault diagnosis of the gearbox more difficult. In order to find a new method to improve the efficiency and accuracy of fault diagnosis of various components in the gearbox, this paper proposes to combine the powerful image recognition capability of convolutional neural network with short-time Fourier transform to apply to gearbox diagnosis. The method transforms the one-dimensional vibration signal into a two-dimensional spectrogram by short-time Fourier transform, and performs normalization preprocessing on the image, and inputs it into the convolutional neural network through Shuffle operation to perform feature extraction to train the model. Using the operations such as Dropout makes the model training faster, and finally uses the trained model to diagnose the fault. The experimental results show that this method can effectively complete a variety of gearbox fault diagnosis and provide a possibility of a diagnostic method connected with “big data”. Compared with the traditional neural network method, the method has a higher efficiency and accuracy of about 5 percent.