An Unsupervised Domain Adaptation Method for Subway Train Transmission System Fault Diagnosis

In the modern rail train system, the safety of fault diagnosis technology is very important. At present, the mainstream algorithms based on data-driven have achieved excellent results in fault diagnosis. However, the current research often focuses on a device data set itself, and its training set and test set are often in the same distribution. Due to the diversity of equipment and working conditions, it is difficult to apply it to industrial practice. To solve this problem, we propose an unsupervised domain adaptation method for subway train transmission system fault diagnosis that integrates convolutional neural networks(CNN) and domain adaptation transfer. Firstly, we extract features based on the data set of subway train transmission system with multiple working conditions and divide the training set, validation set and test set. Secondly, we use the training set to train the CNN to obtain the model weight file. After that, we train the domain adaptation model based on the weight file and the labeled training set and the unlabeled test set. Finally, we tested the accuracy of the model on the test set. Experiments show that compared with CNN without domain adaptation training, the fault diagnosis accuracy of the domain adaptation model is improved by about 27.24%, which verifies the effectiveness of our method.