Research On The Method Of Early Fault Recognition Of Axlebox Bearing Of EMU Based On Shapelets Learning Algorithm

It is of great practical value for the early fault detection and identification of rolling bearings failure,which is one of the key rotating parts of the running part.Under longtime high-speed and heavy-load operating environment,the working conditions of high-speed EMU running parts are becoming increasingly stressful.The rotating parts of the running parts fail frequently in such conditions.The two current fault diagnosis methods have their own advantages.The interpretability of the diagnosis method based on traditional signal processing is high,the fault features extracted from which is provided with Current signal-processing-based fault diagnosis methods underperform intelligence and with clear physical meanings.The diagnosis method based on big data processing technology guarantees extremely high diagnosis accuracy and reduces time consuming at the same time.Although diagnosis method based on big data processing technology has become mainstream and various new methods are constantly emerging,their problems have also been prominent for a long time: that is,the interpretability of diagnostic methods is poor.It completely transforms an actual engineering problem into a problem of establishing a mathematical model,sacrificing the essence of the rolling bearing fault signal as time series.This paper focuses on the shortcomings of intelligent diagnosis methods,taking the nature of rolling bearing fault signals into full consideration,and from the perspective of time series classification,the newly developed shapelets-based time series method has been introduced into the field of bearing fault diagnosis for the first time.Through the high-speed EMU rolling vibration test bench test,an EMU axle box bearing fault data set was compiled,which eliminated the diagnostic method’s dependence on the data set of Case Western Reserve University(CWRU)and targeted early identification of the EMU running gear bearing early faults.The main research contents of this article are as follows:(1)The shapelets discovery algorithm is introduced into the field of fault diagnosis.In response to the shortcomings of the traditional time-consuming shapelets brute force discovery algorithm,a learning algorithm based on machine learning shapelets is introduced,which greatly reduces time consumed by traditional shapelets discovery algorithm and achieves a good-enough fault recognition result on CRWU data set.(2)Based on experiments carried out on the high-speed EMU rolling vibration test bed of the High-speed Wheel-Rail Relations Laboratory of the Chinese Academy of Railway Sciences,a method for collecting signals of bearing fault of the running gear was developed.Faults corresponding to common types of fault of rolling bearing are simulated by manual application,and finally the bearing fault signals are collected.(3)Through the signal classification method based on the shapelets-learning algorithm and the bearing fault signal from the bench test,the unbalanced data set is divided according to the failure ratio in the actual working condition.In order to solve the over-fitting problem of the classification model,the regularization scheme is deeply explored,and the high-precision performance of the diagnosis model is realized both on training set and test set.