Research And Development Of Fault Diagnosis System For Wheel Set Tread Of Metro Vehicles

In the running process of the metro train,the wheel set not only support the whole vehicle body load,but also bear the friction in the frequent braking of the vehicle.As the metro train is in a heavily loaded and dusty working environment,the wheel tread collide with the rails continuously.The tread are subjected to alternating stress for a long period of time,resulting in frequent faults on the tread of the wheel set.Wheel tread fault will accelerate the fatigue damage of the body structure,and will also cause great impact damage to the track subgrade and even serious accidents which the train crashes.To ensure the safe operation of the train,it is of important social and economic significance to find the wheel tread fault in time.Based on the summary of the relevant research methods at home and abroad,this paper selects the wheel-rail vibration signal detection and fault diagnosis system as the research direction.The system is aimed at realizing the real-time detection and early warning of the tread fault.The main research contents are as follows:1.By comparing the advantages and disadvantages of wheel tread fault detection methods at home and in the world,and combining with the development status of domestic rail transit,the method of vibration monitoring is used to realize the fault diagnosis of wheel tread with the vibration signals as its basis.Then,the different signal feature extraction methods and pattern recognition methods are described.The fault types and production mechanism of the tread is analyzed,and the matching scheme of wheel tread and vibration signal is proposed.Finally,the collection of the wheel-rail vibration signal is conducted on the spot.2.Signal feature extraction.The principle and characteristics of fractal theory are introduced.The typical single fractal dimension and its calculation methods are described in detail.First,wavelet packet denoising is used to remove the noise of wheel-rail vibration signals,and then three single fractal dimensions under four conditions of the tread are extracted.Since the single fractal dimension can not fully characterize the wheel-rail vibration signal,the multifractal spectrum of the vibration signal is used to extract its information.According to the spectrum,the five mostrepresentative parameter values are selected as the fault feature vector.Through the comparison of single fractal dimensions and multifractal spectrum features,the latter can more character the wheel-rail vibration signal information.3.Faults identification of wheel tread.First,Least Squares Support Vector Machine(LS-SVM)is established as a classifier.It is used to identify the features of single fractal dimension and multifractal spectrum parameters respectively.Concerning there is a few number of fault vibration tread samples,a new ensemble sampling method(ESM)is proposed to address this issue.Random undersampling for the majority of the samples and SMOTE oversampling for the small number of samples to balance the data sets.And this method is extended to multi-classification,the F-measure and G-mean indexes are used to evaluate the classification performance of the method.4.System for the wheel tread treadle fault diagnosis.First,it should meet the company's needs and functional requirements,and then the overall architecture of the system is given.The emphasis is on the introduction of the system software designing process and software development results.Finally,the whole system is installed in the Guangzhou metro system and achieves great practical application effect.