WMCSCCNN Method for Intelligent Diagnosis of Transmission Faults

In order to overcome the disadvantage of missing information when using single-channel convolutional neural network for fault diagnosis, this paper combines image conversion method and multi-sensor data fusion technology to establish a new self-calibrated convolutional neural network model and apply it to the fault diagnosis of gearbox. Firstly, the vibration acceleration signals of multiple sensors are converted into gray images by image conversion method; Secondly, the self-calibrated convolution structure is introduced on the basis of the traditional multi-channel convolutional neural network to replace the traditional convolution layer to extract deeper feature information; Thirdly, the fusion layer is constructed by combining the multi-sensor fusion technology, and the correlation function weighting method is used to fuse the data collected by multi-sensors, which can adaptively fuse the feature information extracted from multiple channels; Finally, the effectiveness of the proposed method is verified by using multiple sets of data on the planetary gearbox simulation fault test bench of our research group, and compared with other fault diagnosis methods. The experimental results show that the algorithm can achieve more than 90% fault recognition accuracy, and has a good fault diagnosis effect at different speeds, which is feasible for practical application.