Research And Implementation Of Fault Diagnosis Method For EMU Gearbox Based On LSTM And Transfer Learning

As a key component to drive the emu to realize high speed,the normal operation of the gearbox is related to the operation safety of the emu.Once the gearbox fails,it is likely to bring huge economic losses and casualties,so it is necessary to carry out condition monitoring and fault diagnosis for the gearbox.Fault analysis of equipment based on vibration signal is a very effective method,and vibration signal collection of gearbox in the actual operation of emu is also being carried out successively.Because the maintenance of key components of emu often adopts the strategy of periodic replacement,the failure data of gearbox in actual operation is few,and it is difficult to carry out effective failure analysis.In view of the above problems,this paper firstly establishes a gearbox fault diagnosis model based on the vibration signals of gearbox under several fault states measured by the test bench.According to the characteristics of time correlation of vibration signals as well as the comparison of relevant algorithms,the LSTM is selected as the basis of the fault diagnosis model.Then the fault diagnosis model is studied on the cross-domain diagnosis of small sample target domain.Based on the research results,the application of the fault diagnosis model to the gearbox data of emu is determined.This paper mainly conducts the following research:(1)IAPSO algorithm is proposed to search the optimal hyper-parameter combination of LSTM network,and IAPSO-LSTM algorithm with higher classification accuracy is obtained.This algorithm automatically optimizes the hyper-parameter combination of LSTM network by using swarm intelligence strategy of particle swarm optimization algorithm,which solves the problem of low efficiency in manually adjusting the hyper-parameter of LSTM network.On the basis of the traditional adaptive weight particle swarm optimization(APSO),the division of particles is further refined,three intervals are established according to the fitness of particles,and the weight updating formulas are designed according to the characteristics of particles in the three intervals.The above operations form the IAPSO algorithm in this paper.The experimental results show that the IAPSO algorithm has a significant effect on improving the search efficiency.Based on IAPSO-LSTM algorithm,a gearbox fault diagnosis model was established and a comparison experiment was conducted.(2)Based on the study of model cross-domain diagnosis and the analysis of thepractical application demand of gearbox of emu,a migration strategy of gearbox fault diagnosis model of emu is proposed.On the basis of model migration,the strategy maximizes the source data domain,uses cross validation and early stop method to improve the generalization ability of the source domain pre-training model and the training speed and fitting effect of the target model.(3)The results of IAPSO-LSTM algorithm and model migration strategy are verified by experiments on the gearbox data set of emu.The classification effect of the target model is evaluated under the two practical application scenarios of imbalanced fault data and complex operating conditions,and proves that the migration strategy proposed in this paper is feasible in practical application.