Conditional adversarial domain adaptive bearing fault diagnosis method based on embedding discrimination

Abstract

The invention discloses a conditional adversarial domain adaptive bearing fault diagnosis method based on embedded discrimination. The method comprises the steps of: collecting first vibration signalsand second vibration signals of a high-speed rail traction motor bearing operating under different working conditions, enabling the first vibration signals to serve as source domain data, and enabling the second vibration signals to serve as target domain data; establishing a conditional adversarial domain adaptive network comprising a feature extractor F, a label predictor G and a domain classifier D; inputting source domain data with labels and target domain data without labels at the same time to train the network; and after completing training, inputting the target domain data with labelsinto the trained conditional adversarial domain adaptive network so as to be subjected to forward propagation, and outputting a fault prediction result by the trained conditional adversarial domain adaptive network to realize high-speed rail traction motor bearing fault diagnosis. According to the method, the diagnosis accuracy of the model on the target domain data can be effectively improved, inter-domain difference is reduced, and the discrimination of feature distribution is enhanced, so that the robustness of the model is improved.

Description

technical field [0001] The disclosure belongs to the field of mechanical fault diagnosis, and in particular relates to a bearing fault diagnosis method based on embedded discriminative condition confrontation domain self-adaptation. Background technique [0002] In recent years, deep learning has been widely used in intelligent fault diagnosis due to its powerful feature extraction ability and ability to process large data. However, due to frequent changes in the working conditions of high-speed rail traction motors, the assumption that the training samples and test samples of deep learning are subject to independent and identical distribution is broken, and the diagnostic accuracy of deep learning is greatly reduced when it is applied to the fault diagnosis of high-speed rail traction motors. The existing depth-domain adaptive methods cannot fully reduce the inter-domain differences because they do not use label information. Moreover, the traction motors of high-speed railw...

AI Technical Summary

Problems solved by technology

However, due to frequent changes in the working conditions of high-speed rail traction motors, the assumption that the training samples and test samples of deep learning obey the independent and identical distribution is broken, and the diagnostic accuracy of deep learning is greatly reduced when it is applied to the fault diagnosis of high-speed rail traction motors.

The existing depth-domain adaptive methods cannot fully reduce the inter-domain differences because they do not use label information. Moreover, the traction motors of high-speed railways need to withstand the shock and vibration generated by the dynamic action of locomotive wheels and rails, resulting in noise and noise in the collected vibration signals. There are many interference components, so a model with strong robustness is needed

Images