Research On Fault Diagnosis System Of Metro Vehicle Gearbox Based On Oil Wear Particles Image

With the continuous development of national modernization and the continuous expansion of urban scale,urban rail transit has been developed rapidly under the vigorous promotion of various national policies.As an important part of urban rail transit,metro vehicle has become one of the first means of transportation for people’s daily travel with its advantages of comfort,convenience,safety,reliability,energy conservation and environmental protection.At the same time,the daily huge passenger flow also puts forward higher requirements for the safety of metro vehicle.Gearbox as a key component of metro vehicles,the reliability of it directly affects the safety of metro vehicles,since its poor working environment,such as long-time work and frequent start and stop,so that it becomes one of the components prone to failure.Therefore,fault diagnosis is particularly important.This paper first analyzes the common faults and their generation mechanism of the gearbox of metro vehicles,and determines the oil wear particle image detection as the fault diagnosis method of the gearbox of metro vehicles.Then the overall scheme of the fault diagnosis system is designed,and the fault diagnosis system is divided into three parts: image acquisition,image processing,feature extraction and fault diagnosis.In the part of image acquisition,the performance parameters and their influences of key components such as camera and lens are analyzed,and key components are selected to construct acquisition system for obtaining image data.In the part of image processing and feature extraction,the processing method of wear particle image and wear particle feature parameters are studied,and then the technical routes of image processing and feature extraction based on the collected pictures and monitoring video are formulated,and the corresponding programs are written to meet the requirements.Secondly,this paper focuses on the research of fault diagnosis method,using support vector machine(SVM)and Pretrained convolutional neural network(CNN)to classify the oil wear particle image of gearbox,and analyzing the factors that limit the accuracy of classification.SVM classifier requires high quality of input features,However,the Pretrained CNN model does not consider the relationship between the training set and the test set.To solve the above problems,this paper introduces MAML(Model Agnostic Meta Learning)algorithm to optimize the training of CNN model,and sets the wear particle classification problem as 4-way k-shot problem.Firstly,the feasibility of MAML algorithm is verified by using bearing vibration data.The MAML algorithm is applied to bearing fault diagnosis,and the maximum mean difference(MMD)loss with penalty coefficient is added to the loss function.The bearing vibration data of Case Western Reserve University and bearing life data of Xi’an Jiaotong University are used for training and testing.The test results show that,compare with Pretrained model,under the condition of 1 shot,The accuracy of the model optimized by MAML algorithm can be improved by 44.9%,and the accuracy of the model optimized by MAML algorithm combined with MMD loss can be improved by 55.4%,and in the state of 5 shot,the accuracy of the CNN model optimized by MAML algorithm can be improved by 30.3% and34.8% respectively,which shows that the CNN model optimized by MAML algorithm can effectively improve the accuracy in dealing with few samples.Then the CNN model optimized by MAML algorithm is applied to the classification of oil wear particles in the gearbox of metro vehicles for fault diagnosis.The test results show that the average accuracy of fault identification is 61.8% in the state of 1 shot,which is 22.9% higher than the Pretrained method,and 82.1% in the state of 5 shot,which is 13.1% higher than the Pretrained method.Finally,the interface of gearbox fault diagnosis system for metro vehicle is compiled.The image processing,feature extraction and fault diagnosis methods are integrated,and the human-computer interaction interface of the fault diagnosis system is compiled.