Research on Algorithm of Mechanical Fault Detection Based on Convolution Neural Network and GMM12

With the development of modern manufacturing industry, various kinds of mechanical equipment are applied more and more widely. However, once such equipment fails, it will not only cause losses to the company, but also may cause casualties even in serious cases. At present, most fault diagnosis of devices relies on feature extraction methods, usually based on information processing technology, to extract fault features manually, to identify device faults according to the differences of fault features, and combined with classifiers. Obviously, this method needs prior knowledge and professional support, and the accuracy of fault diagnosis cannot be guaranteed. This project uses the Gaussian mixture model (GMM) which is widely used in industry and the Convolutional Neural Network (CNN) model which has advantages in image processing to detect the faults of devices. The model is trained using a sound dataset for investigation and inspection of faulty industrial machines that is publicly available on the Web. GMM and tensorflow-keras framework are built using Python language to build convolution neural network model, and the recognition accuracy of the two algorithms is analyzed. The results show that when the training data is small, the recognition accuracy of GMM is 73%, while that of convolution neural network model is 73.2%. When the training data is large, the recognition accuracy of GMM is 75.8%, and that of convolution neural network is 94.2%. When the amount of data is increased, the accuracy improvement of GMM is less than that of convolution neural network.