Early Fault Detection Of High-Speed Train Wheel Set Based On Metal Magnetic Memory Effect

As the key component of the high-speed train running part,the wheel is directly related to the safe operation of the train,while the local stress concentration and micro-crack are the main reasons for the sudden failure of the wheel pair.At present,X-ray flaw detection,ultrasonic flaw detection and magnetic particle flaw detection are mainly used to detect wheel faults in China,but those methods only can detect the existing wheelset cracks and have no ability to early faults in advance.Metal magnetic memory detection can effectively detect the stress concentration zone and microcracks of the wheelset by collecting the self-leakage magnetic field signal on the surface of the wheelset.It is an early quantitative diagnosis process of “predicting-forecast-evaluation”,which has important practical significance and application prospects.Metal magnetic memory is a weak geomagnetic signal and is highly susceptible to external interference and noise,which affects the reliability of early diagnosis.As a result,it is difficult to denoise magnetic memory signals and extract accurate early fault features.At the same time,the characteristics of the magnetic memory signal and the damage state of the wheelset are not a simple linear relationship,which limits the qualitative and quantitative fault assessment of the high-speed wheelset in engineering application.This paper focuses on the above two problems,the main contents are as follows:(1)The formation mechanism of magnetic memory phenomena is studied from the perspectives of electromagnetics and energy balance.The magnetic dipole model and energy balance model of stress concentration region are constructed respectively.The validity of the model is verified by simulation,which provides effective theoretical support for practical application.(2)According to the principle of correlation,the dissimilarity of the multi-channel magnetic memory signals is eliminated after dissimilarity processing,which improves the reliability of detection.The multi-scale signal decomposition is performed base on the stationary wavelet transform.Then an autocorrelation-based wavelet threshold denoising and three-scale correlation adaptive denoising method are proposed to accurately determine the stress concentration region of the wheelset.The denoising results of test data show that the magnetic memory signal after threescale correlation adaptive denoising has smoother signal and higher signal-to-noise ratio,which better retains the key fault information of the wheelset and improves the reliability of subsequent qualitative and quantitative evaluation wheel faults.(3)In view of the qualitative evaluation of magnetic memory of high-speed rail pairs,the shortcomings of commonly used normal gradient extreme value evaluation are first described.Secondly,the Lissajous graph method,which is composed of the tangential component and the normal component of magnetic memory,is analyzed.The research shows that the method cannot accurately determine the location of the stress concentration zone.To overcome the limitations of the existing method,a new normal component phase trajectory method is proposed to semiquantitatively(qualitatively)evaluate the stress-free concentration,elastic deformation,plastic deformation and fracture state of high-speed train wheels.The validity of the method is verified by experimental data.(4)Faced with the problem that current magnetic memory detection cannot accurately evaluate the early damage of the wheelset,the multi-features of the wheel-pair magnetic memory signal is firstly established and then use the small sample advantage of the support vector machine to establish a support vector machine evaluation model based on the principal component analysis method.The support vector machine adopts the grid search method for parameter optimization.The assessment model has good robustness as the recognition rate of early faults reaches 90%.