Research On Fault Diagnosis Of High Speed Train Traction System Based On SFA

High speed trains are an important component of modern transportation systems.As the core of high-speed trains,the smooth operation of the traction system is of great significance for the safety of the entire vehicle.However,as the working hours of the traction system increase,internal electrical and mechanical components are prone to aging,wear,and other issues,affecting train operation,and even leading to serious faults,which will bring huge safety hazards to the safety of passengers’ lives and property.Therefore,in order to ensure the reliability of trains,it is necessary to conduct research on fault diagnosis of high-speed train traction systems.The structure of the traction system is complex and the operating environment is variable,and traditional fault diagnosis methods directly applied to the traction system have limited effectiveness.This article focuses on the system characteristics of high-speed train traction systems and proposes a fault detection scheme using Slow Feature Analysis(SFA)based on Multivariate Analysis(MVA)method.This article mainly studies the following aspects:(1)To balance the fault detection performance of traction systems and reduce the probability of false positives,a fault detection method based on Probability Related Slow Feature Analysis(PRSFA)is proposed.Introducing Kullback Leibler divergence(KL divergence)into slow feature analysis methods can fully utilize the characteristic that fault information can lead to abnormal probability distribution of slow feature.Extract slowness features from traction system data using SFA analysis method,further quantitatively analyze the probability distribution difference of slowness features between offline and online data using KL divergence,and construct an evaluation function to complete fault detection.Through experimental simulation,it has been verified that this method can detect faults in a timely manner,and has better detection performance compared to existing similar algorithms.It achieves a lower missing alarm rate while achieving an acceptable false alarm rat.(2)A fault detection model based on KL divergence and Deep Kernel Slow Feature Analysis(DKSFA)is proposed for small faults detection in nonlinear traction systems.Firstly,the kernel method is used to map the low dimensional linearly indivisible traction system data to the high-dimensional space,making it linearly separable in the feature space.Then,the slowness features that change slowly are extracted through the dimensionality reduction of the slow feature analysis method.Secondly,through the deep kernel slow feature analysis method,the obtained slow features are multi mapped to deeply mine small fault information that is easily masked by noise and improve sensitivity to small faults.Finally,the evaluation function constructed through KL divergence is used for detection tasks,ensuring the detection performance of small faults.Through experimental simulation and comparative analysis,it has been verified that this method can effectively process nonlinear traction system data,deeply mine small fault information,detect small faults that are easily masked by noise,and have better detection performance.