A visual vibration characterization method for intelligent fault diagnosis of rotating machinery

Health monitoring and fault diagnosis are the keys to ensuring equipment safe operation. This work proposes a novel fault diagnosis method based on visual extraction and vibration characterization. Instead of using conventional accelerometers to obtain fault data, the visual extraction method obtains the full-field vibration information with rich texture features and produces no mass loading effect on the measured object. This method extracts the time-domain vibration information from the collected image sequences through image phase difference, and then encodes it into gray-scale images as input for a convolutional neural network model. The experimental results testing on the bearing vibration image dataset show that the proposed method can achieve superior performance in fault diagnosis. It gains superior results with high classification and recognition accuracy.