Research on Mechanical Fault Diagnosis Algorithm Based on Sound Signal and CNN

Failure diagnosis is of great significance for the timely detection of the safety hazard of the equipment and the guarantee of the normal operation of the production. In fault diagnosis, the way based on the processing of sound signal has the advantages of strong fault sensitivity, easy acquisition, and noncontact measurement, and the way of using neural network provides a more efficient and generally applicable method for fault diagnosis efficiency. For the poor diagnostic accuracy of traditional methods, which requires manual extraction of features and poor general applicability of the model, in this paper, we propose a mechanical failure diagnosis method based on acoustic signals and CNNs. The sound signals were first sampled and features extracted by MFCC, then the data were split into training and test sets in a 6:4 ratio and input to the convolutional neural network. After adjusting the parameters for the comparison experiment, the final experimental model was able to achieve 97.05% test accuracy over 20 training test iterations.