Research On Fault Classification Method For Key Components Of Emu Based On Deep Learning

In the development of China,the high-speed trains are in the leading position in the word.As the main part of the key components of rotating machinery of EMU,the health status of gearbox is very important.Once the machine is in fault,it will be cause inestimable cost losses.It has certain engineering significance to reduce the maintenance cost and reduce the occurrence of accidents by diagnosing the fault as soon as possible.With the development of artificial intelligence and deep learning,data driving method has become the mainstream technology of fault diagnosis.The traditional method of the vibration signal is to reduce the noise,adopt time-frequency analysis,employ the feature extraction of the signals,and finally match the fault features manually to judge whether the rotating machinery has fault and identify the type of fault.The artificial intelligence solves the difficult task of manual recognition.The algorithms of signal denoising,feature extraction and intelligent classification are studied by more and more scholars.The intelligent diagnosis model constructed by deep learning technology has achieved excellent performance in the field of fault diagnosis in time series prediction in recent years.In this paper,through the vibration signals studying,EMU rotating machinery fault classification method researching is combined with feature extraction,intelligent fault diagnosis and transfer learning algorithms.The vibration data of gearbox are non-linear and non-stationary.Empirical Mode Decomposition(EMD)can deal with the characteristics of this kind of data,and Ensemble Empirical Mode Decomposition(EEMD)can solve the problem of mode aliasing of EMD method.The gearbox vibration data is regarded as a kind of time series data,an improved Long Short-Term Memory network combined with Particle Swarm Optimization algorithm is proposed,and transfer learning method is presented to solve the practical application of lack fault data.The main contents of this paper are as follows:(1)The vibration theory of gears and bearings in the gearbox,including gear fault types and bearing fault types,and the compound fault vibration signals of the gearbox are studied.(2)In the aspect of signal preprocessing,an Improved Ensemble Empirical Mode Decomposition(IEEMD)is proposed,which effectively alleviates the mode aliasing phenomenon in EMD by adding white noise,and the endpoint effect of EEMD is solved by extreme wave extension and window function.(3)Aiming at the feature extraction and fault classification method,an improved Particle Swarm Optimization(ACMPSO)is proposed to optimize the long short-term memory neural network.The long short-term memory neural network is an end-to-end deep learning algorithm,which no need for manual feature extraction.The particle swarm optimization automatically optimizes the hyperparameters of LSTM to find the optimal parameters.At the same time,the root mean square error(RMSE)is introduced to verify the IEEMD-ACMPSOLSTM model’s performance.(4)In practical application,the EMU gearbox has the problems of less fault data and uneven distribution,transfer learning algorithm transfer the knowledge learned in the source domain to the target domain in small data set.The Maximum Mean Difference(MMD)is a function to test the fault similarity between the source domain and the target domain.After MMD and fine-tuning strategy,the model parameters are transferred.The proposed method is effective and feasible through the comparison of experimental simulation.