A Domain Mapping Transfer Learning Method for Fault Diagnosis Based on Current Signal

Deep learning technology has been widely used in fault diagnosis to ensure the stable operation of equipment. However, the excellent performance of deep learning-based fault diagnosis highly depends on a large number of high-quality training samples. And most of existing methods are based on vibration signals, which are difficult to obtain. Therefore, a domain mapping transfer learning method is proposed to achieve high precision fault diagnosis based on current signals under variable operating conditions. Firstly, the distribution alignment of cross-domain sample is achieved by optimizing a mapping function, so as to reduce the influence of domain offset effect on the accuracy of the model. And this process is thought of as mapping the source domain sample to the target domain. Secondly, the one-dimensional convolutional neural network model is trained using mapped source domain sample, achieving high diagnostic accuracy on target domain. The experiment on gearbox datasets shows feasibility and superiority of the proposed method.