Research On Intelligent Fault Diagnosis Method Of Rolling Bearing Of Railway Train Running Part Based On Signal Processing And Deep Learning

In recent years,the scale of railway transportation in my country has been expanding,and its operational safety has gradually attracted widespread attention.As the carrier and core of railway transportation,the potential safety hazards of railway trains can easily be further expanded into serious safety accidents.Among the many components of railway trains,the rolling bearing of the running part of the railway train plays a very critical role,and its service performance has a very direct impact on the running safety of the railway train.Therefore,it is of great significance and necessity to carry out in-depth research on the fault diagnosis technology of rolling bearings in the running part of railway trains for the purpose of improving the operational safety assurance level and scientific operation and maintenance of railway trains.This paper takes the rolling bearing in the running part of the railway train as the research object,and faces the fault diagnosis task of the rolling bearing in the running part of the railway train under different application scenarios.The pattern recognition problem of automatic identification of signal health status and the small sample problem of lack of labeled samples in actual operation scenarios,this paper has carried out in-depth research work,and completed the following four main work contents:(1)Aiming at the problem of impact noise of railway trains in complex operating environment,in order to achieve better feature extraction effect,it is necessary to highlight the fault characteristics of rolling bearings in running parts.Therefore,a new method based on Adaptive Complementary Ensemble Empirical Mode Decomposition and Cyclic Spectrum Coherence is proposed.The proposed data preprocessing method achieves highlighting of fault features by reducing the shock noise in the vibration signal.and uses two rolling bearing vibration data sets to verify the effectiveness of the algorithm.(2)Aiming at the problem of fault feature extraction based on vibration signal,a multidomain fusion feature matrix based on Adaptive Complementary Ensemble Empirical Mode Decomposition and Cyclic Spectrum Coherence is proposed,which can obtain a variety of more features from the time domain,frequency domain and entropy domain.Rich feature indicators and further fusion calculation and processing to achieve comprehensive extraction and fusion of fault feature information,in order to achieve better fault feature extraction effect,the effectiveness of the proposed fault feature extraction method is verified by using two rolling bearing vibration data sets.(3)Aiming at the pattern recognition problem of automatic identification of health status based on vibration signals,a diagnostic model based on densely connected network and Transformer is proposed.The densely connected network model is used to realize the deep representation of the extracted multi-domain fusion feature matrix.The encoder structure of Transfomer realizes the deepening of the network model,and achieves better processing effect on multiple features,and finally achieves better intelligent diagnosis effect for rolling bearing faults in the running part of railway trains,and uses two kinds of rolling bearing vibration data sets to verify the proposed the effectiveness of the algorithm.(4)Aiming at the problem of small samples with few labeled samples in the actual operation scene of railway trains,a meta-learning model based on relational network is proposed that combines dense connection network and multi-head self-attention,so that the model can master the ability to distinguish the relationship between different samples,so that the model has the self-learning ability under a small number of samples,and achieves a better diagnosis effect for the small sample problem of rolling bearing fault diagnosis in the running part of the railway train,and uses two rolling bearing vibration data sets to verify the effectiveness of the proposed algorithm.