Research On Intelligent Diagnosis Technology Of Traction Converter Fault Of High-speed Train Based On Multi-sensor Fusion

With the rapid development of railway technology and the continuous increase of high-speed rail operating mileage,the number of online railway trains in my country has gradually increased.The research of train condition safety supervision and fault diagnosis is becoming more and more important.Traction converter system is an important device for kinetic energy conversion of trains,which has the characteristics of complex system structure and high failure rate.The occurrence of converter faults often affects the train traction drive system and hinders the normal operation of the entire vehicle.At present,there are not many studies on fault diagnosis of traction converters,so fault diagnosis for train traction converters is a key research direction.However,the failure scenarios of train traction converters are complex.Traditional diagnosis methods mostly rely on singlesensor data,which cannot comprehensively include fault characteristics and are easily affected by environmental factors,resulting in low diagnosis efficiency and difficulty in meeting actual needs.Therefore,this paper carries out the research on data-driven fault diagnosis technology of high-speed train traction converter,which mainly includes the following research contents:(1)This article summarizes the research status of high-speed train fault diagnosis technology and focuses on the problems of traditional diagnosis methods.Then the article introduces the sensor data characteristics of the high-speed train traction converter and selects the characteristics of converter sensor data from the perspective of physical mechanism and statistics.Finally,10 characteristic variables which have great influence on converter faults are selected as the data basis for the follow-up research in this paper.(2)Traditional converter fault diagnosis is mostly based on single-signal sensors,which can only represent the abnormal state of the monitoring index and cannot accurately represent the component fault characteristics.It is susceptible to noise interference,which leads to low diagnosis efficiency.To solve these problems,this paper proposes a multi-fault intelligent diagnosis scheme based on LSTM.The model adaptively learns the data characteristics of multi-source variables and takes effective data fusion strategy to identify different types of faults.In the process of diagnosis,the model is good at extracting the long-term dependencies hidden in the time series data and can make good use of the time-domain correlation characteristics of the before and after data fragments.In addition,the multivariate fusion strategy makes full use of the spatial correlation characteristics among the data of different variables,which enables the spatial and temporal characteristics of the input data to be fully mined.Experiments show that this method can effectively diagnose most of the faults in the converter scenario.(3)This article makes an in-depth comparison of various intelligent diagnosis methods to study the respective performance advantages of different models.This article conducts sensitivity verification experiments in single sensor scenario and conducts multiple fault diagnosis experiments in a multi-sensor scenario to test the performance of different models.In addition,DCNN and LSTM are compared to analyze the process of feature extraction,and study of the impact of different training set sizes on the performance of each model.Targeted optimization strategies are put forward to improve the overall accuracy of the experiment for the difficult samples that are easily confused.Then,the scalability experiment test was carried out in the auxiliary converter scenario,and satisfactory results were obtained.(4)A high-speed train networked data fault diagnosis system is developed,which visually presents the intelligent diagnosis results in the system,and dynamically supervises the real-time sensor data with data visualization technology.In addition,the system can also effectively manage the historical sensor data.The system combines realtime signal monitoring,fault diagnosis and data management to provide technicians with intuitive and visual train operation status monitoring services.There are 48 pictures,20 tables and 63 references in the body.