Research On Fault Diagnosis Of Railway Wagon Bearing Based On Optimized Probabilistic Neural Network

The rolling bearings are the key supporting component of the railway wagon.Bearing inspection relies on the maintenance personnel to rotate the bearing outer ring,and judge whether there is any fault by hand touch or ear.The existing bearing inspection relies heavily on maintenance personnel's responsibility and work experience,which leads to many serious problems,such as over-repair,ambiguous judgment standards,low efficiency,and high cost.It lags behind the development of vehicle technology.The key to solving the problem is how to achieve accurate fault diagnosis of the bearing,and repair it according to the true "health state" of the bearing.This thesis aims at developing a intelligent and efficient bearing fault diagnosis platform.First of all,doing research on the fault mechanism of railway wagon bearing,and a new fault diagnosis method for railway wagon bearing fault is proposed,then the bearing data is collected by the railway wagon bearing fault automatic detecting device,finally,a railway wagon bearing fault diagnosis system is developed with the GUI of MATLAB,and verify the fault diagnosis system by practical data.The main contents are as follows:(1)Aiming at the low accuracy of manual detection of railway wagon bearings,a new method for fault diagnosis of railway wagon bearings is proposed.Firstly,the optimized multi-scale wavelet packet(OMS-WP)is used to denoise the bearing signal,and the signal-noise ratio,root mean square error and shannon are used as indicators.Then,the intrinsic mode function is selected based on the Hilbert-Huang transform(HHT)and the kurtosis coefficien,the multi-scale permutation entropy(MPE)of the most relevant intrinsic mode function is calculated,and combine the time domain parameters and the IMF energy entropy to form a multi-feature vector.Finally,in order to avoid dimension explosion,the reciprocal of root mean square error is taken as the fitness function.The multi-feature vector is filtered by genetic algorithm(GA),and the probabilistic neural network(PNN)parameters are optimized.It proves that the proposed method can identify bearing faults accurately.(2)Data acquisition and analysis of C64K railway wagon bearings were carried out.The bearing data is collected by the automatic bearing fault detection device of the railway wagon and the acquisition software designed by our laboratory,two vibration acceleration sensors are arranged respectively at the same position of the bearing of the railway wagon,and the manual inspection results are recorded.The collected signals are analyzed according to the fault diagnosis method of the railway wagon bearing,and the test results of the test set are completely consistent with the inspection results.(3)A fault diagnosis system for railway wagon bearings is developed based on MATLAB GUI,the system mainly includes five parts:the system login module,the signal time domain analysis module,the OMS-WPD denoising module,the HHT-MPE feature calculation module,the GA-PNN fault diagnosis module.Each module is nested with each other,the operator only need to import the data and make mouse clicks to get the fault status of the bearing.After the comparison test,the diagnostic results output from the system is the same as the actual state of the bearing,which proves that the railway wagon bearing fault diagnosis system can identify the fault effectively.