| The rapid development of high-speed train has been facilitated by the huge demand for travel,and high-speed train has gradually become one of the necessary means of transportation.As the core rotating part,rolling bearing plays an important role in the automatic production line of high-speed train and ensuring the high-speed and stable driving of the train.Its running state directly determines the state of related equipment.Fault diagnosis using vibration signal data is one of the most effective methods.However,the traditional method based on manual analysis is complex and the results are not satisfactory.There may be cross-domain problems,such as cross-working condition fault diagnosis,that is,the bearing operating conditions are complex and changeable,and it is difficult to collect the labeled data under all operating conditions for training model.In view of the above problems,this paper introduces the algorithm of deep learning into the fault diagnosis of rolling bearings,and proposes a set of fault diagnosis methods for rolling bearings based on transfer learning and deep convolution.Specific research content is divided into the following three aspects.(1)A fault diagnosis model based on multi-scale one-dimensional convolutional neural network is proposed for the bearing fault diagnosis under constant working conditions,that is,the training set and the test set are from the same distribution.The model takes one-dimensional vibration signal as input and extracts features through multi-scale convolution kernel with different sizes,so as to enhance the capability of network feature expression and improve the prediction accuracy of the model.The effect of model fault diagnosis is verified on the single working condition data set.Then the experiment was carried out on the data with Gaussian white noise,and the recognition accuracy was also higher.The model of this chapter lays a foundation for the research of the following two chapters.(2)In view of the problem that the prediction accuracy is decreased due to the change of data distribution in the fault diagnosis of bearing in cross-domain,deep transfer learning is introduced for optimization.We use MK-MMD and CORAL to calculate the data distribution differences between source and target domains and as part of the loss function to achieve the effect of domain adaptation.Fault diagnosis experiments in crossworking condition and cross-scene tasks verify that the model has good domain adaptive ability.(3)The idea of deep adversarial network is introduced to further optimize the task of cross-domain diagnosis.A cross-domain fault diagnosis model based on deep adversarial network is proposed.Using Wasserstein distance as a criterion to measure the difference of data domain distribution.While optimizing the edge distribution difference of source domain and target domain,we can optimize the conditional distribution difference of two data domains by using the pseudo-label generated by classifier to target domain,so as to reduce the intra-class difference and increase the distance between classes.Finally,the model has higher recognition accuracy and better cross-domain diagnosis ability in cross-working and cross-scene tasks. |
