Research On Fault Prediction Method Of EMU Traction Motor Based On Digital Twin Technology

The traction motor of EMU is the key component of EMU.Due to the serious consequences of the traction motor failure during the operation of EMU,the maintenance of EMU traction motor will be carried out regularly during the operation and maintenance of EMU.However,the maintenance needs more manpower.As a new method in manufacturing industry,digital twin provides a new solution for fault diagnosis of EMU traction motor.Digital twin technology can simulate the state of physical entity by constructing digital twin of physical entity and analyzing twin data,and transfer the characteristics of twin to physical entity for analysis by using transfer learning method,so as to realize fault diagnosis in the case of less physical data.Firstly,the digital twin of EMU traction motor is constructed based on digital twin technology.In theory,digital twins can simulate the data characteristics of physical entities,analyze the stress characteristics of digital twins with mathematical and physical methods,build the mathematical model of digital twins with Simulink,and run the digital twins to get the twin data.The data from the operation of the digital twin and the experimental data from the operation of the physical entity in the test-bed are sorted into the same standard form.The random window sliding is used to segment the data,and the data set is divided into several pieces of data of the same size.Then the fast Fourier transform is used to transform the time-domain signal into the frequency-domain signal,and the data label is added to form the final experimental data set,and the digital twin data analysis model is input.The experimental data are divided into source domain data and target domain data.The training network is designed based on the deep learning and migration learning methods.The feature extraction layer is designed to extract the data features.The dropout algorithm is used to optimize the over fitting problem.The objective loss function is designed based on a variety of transfer learning algorithms.The network loss function is calculated for back propagation,and the weight of each layer of nodes in the network is updated by Adam parameter optimization algorithm.The training network can extract data features.The smaller the difference between source domain features and target domain features,the better the effect of feature extractor.Then,a classifier with three full connection layers is designed to test the effect of data classification.Finally,the digital twin is regarded as the source domain,the physical entity of thetest-bed is regarded as the target domain,and the migration process is designed to move from the digital twin to the physical entity.According to the input network,the data is divided into source domain training data,source domain test data,target domain training data,target domain test data and semi supervision data.Input the training data into the training network structure of the feature extractor,and use the test data to evaluate the effect of the classifier.The experiment compares the classification effect of various models,and adjusts the super parameters according to the algorithm effect.A semi supervised scheme is proposed to optimize the experiment.The experimental results show that the fault analysis model based on digital twin technology can effectively identify the fault of traction motor.