Research On Fault Diagnosis Of Rolling Bearings Of Railway Trains Based On Optimized MOMEDA And Improved PSO-Elman

In recent years,with the construction of China’s eight vertical and eight horizontal high-speed railway network,the number of high-speed trains in China has increased day by day,and the safe operation of trains has attracted more attention.The working state of the running gear has a huge impact on the safe operation of the train,and the rolling bearing,as an important component of the railway train running gear,will directly affect the safety of the train once there is a problem.At present,China’s train bearing testing adopts the method of manual identification,and only based on manual experience to judge bearing failures will produce a series of problems such as high misdiagnosis rate,low efficiency and excessive repair.In order to solve the above problems,this paper will take the axlebox bearing of the railway train walking section as the research object,and carry out a systematic study on the early fault diagnosis,feature extraction and failure mode identification methods of rolling bearings,and the main research contents are as follows:(1)The common fault and vibration mechanism of train axlebox bearings is analyzed,and the calculation formula of the failure frequency of axlebox bearings and the basic characteristic frequencies generated under various fault conditions are derived and listed.This paper summarizes the research status of fault diagnosis related algorithms at home and abroad,and establishes the research ideas and contents of this paper on this basis.(2)Aiming at the problem that the parameters of the MOMEDA algorithm are difficult to select,this paper innovatively proposes a signal autocorrelation method to extract the deconvolution period and select the filter length by the envelope harmonic signal-to-noise ratio spectral entropy.Envelope demodulation and 1.5-dimensional spectral analysis of the MOMEDA processed signal can accurately extract the fundamental frequency and multiplier of the fault characteristic frequency.(3)In order to solve the problem that the traditional single domain features are not comprehensive enough for the fault description,35 feature parameters are extracted from the time domain,frequency domain,and time frequency domain to construct a multi-domain feature set.Aiming at the problems of data redundancy and low computational efficiency caused by the excessive dimensionality of multi-domain feature sets,KPCA is introduced to reduce the characteristic dimensions of multi-domain feature sets.(4)In order to improve the bearing fault identification ability,this paper constructs a model to improve the PSO optimization Elman neural network.Aiming at the problem that the PSO algorithm is prone to premature convergence and the search accuracy is low and is not conducive to later iteration,inertia weights are introduced to improve the search capability of the PSO algorithm.The improved PSO algorithm is used to iterate on the weights and thresholds of Elman neural networks to improve the speed and accuracy of its training.(5)By comparing and analyzing the training speed and classification accuracy of different training models,it is found that the classification accuracy of various bearing fault conditions is 100% by comparing and analyzing the training speed and classification accuracy of different training models,thus verifying the practicality and rationality of the proposed algorithm process.