Research On Fault Diagnosis Method For Axle Box Bearing Of High-speed Train Based On Dispersion Entropy Algorithm

Due to the large scale of railway transportation in China,trains often travel in complex environment,coupled with the continuous improvement of vehicle operation speed,which makes the service conditions of train bearing worse and worse.If the structural state of bearing is degraded,it may lay a huge hidden danger for driving safety,leading to derailment,rollover and other accidents.Thus,monitoring the service status of the axle box bearing,diagnosing its early failures and warning in advance are effective methods to avoid major risks.In this paper,the fault diagnosis of axle box bearing on the high-speed train is the main research content.The key features of bearing vibration signals are extracted by dispersion entropy algorithm,and then the intelligent diagnosis models are built by combing Disp En with feature evaluation and various machine learning algorithms,which can identify a variety of bearing fault states to ensure the safe operation of the train,and has practical significance.The paper mainly completes the following tasks:(1)The experiments and algorithms needed for fault diagnosis of rolling bearing are fully elaborated.The calculation principle and steps of multiscale dispersion entropy are deconstructed in detail,compared with algorithms such as sample entropy and fuzzy entropy,it has better processing ability for short time series and higher calculation efficiency.At the same time,it optimizes many shortcomings of permutation entropy and has good feature extraction capabilities.After that,m RMR algorithm is used for feature reconstruction,and WELM,LSSVM,Fk-NN and PNN are introduced for identification and classification,finally,the intelligent diagnosis model of axle box bearing is constructed.(2)In order to effectively capture the periodic impact characteristics in the rolling bearing fault signal,the fluctuation-based dispersion entropy algorithm is introduced on the basis of the traditional dispersion entropy.This method puts more emphasis on the continuous change of signal amplitude,so its entropy contains more fluctuation characteristics,which can better describe the degraded state of the bearing.Meanwhile,in order to solve the problem that entropy fluctuation caused by the scale factor increase,a modified multiscale fluctuation-based dispersion entropy algorithm was formed by combining fluctuation-based dispersion entropy with modified coarse-graining process.Then,the applicability of MMFDE is verified by the signals of motor and axle box bearing,which has achieved higher diagnostic accuracy compared with MDE.(3)In order to further improve the feature extraction effect of dispersion entropy,the weighted dispersion entropy algorithm is proposed in this paper.This method mainly considers the problem of the lack essential information in patterns,and calculates the standard deviations of different patterns as the weight values which are assigned to various patterns.The more complex the model structure is,the higher the weight becomes,and vice versa.Finally,the entropy values are corrected through the weight probabilities and weight coefficients so as to reduce the errors of the entropy values and enhance the ability to characterize key features.Combining with the improved coarse-graining process,the improved multiscale weighted dispersion entropy is proposed,and the performance of the algorithm is analyzed to determine the optimal parameters.Finally,the experimental data of two types of bearings are used for verification.The results show that the improved multiscale weighted dispersion entropy can accurately characterize the operation status of different damage types and damage degrees of bearings,and obtain the expected recognition effect.