Fault Diagnosis In Running Gears Of High-speed Trains Based On Slow Feature Analysis

The running gear system of high-speed trains is a complex electromechanical system which is composed of mechanical structural components and control electrical units.It plays a key role in ensuring the flexible track change of high-speed trains,easing the impact between car body and rail,and providing the smooth operation.The safety and reliability of running gear systems has been paid more and more attention by the state.Fault diagnosis technology,as an effective method to ensure the efficient,stable and safe operation of high-speed trains,can monitor the state of running gear systems.Due to various sensors are widely used,the massive data provides the possibility for real-time monitoring the performance of running gear systems.However,the running environment of running gear systems is complex and changeable in actual conditions.Up to now,a relatively complete and mature fault diagnosis system for running gear systems has not been formed.Hence,based on the practical engineering problems,the main research contents are as follows:(1)Aiming at the fault data of traction systems to analyze the fault information,a system monitoring method based on block dynamic slow feature analysis(SFA)is proposed.The correlation between all variables is calculated by mutual information,and the variables are divided into blocks according to the correlation.Then,the SFA method is introduced.And on this basis,the dynamic SFA method is adopted to monitor the variables of each sub-block.Two test statistics are designed to analyze the local monitoring performance of systems.Bayesian inference algorithm is used to integrate all blocks to obtain the global monitoring performance of systems.Finally,the effectiveness of method is verified in the traction systems of running gear.(2)Aiming at the incipient fault diagnosis of running gear systems,a fault detection and diagnosis method based on deep slow feature analysis(DeSFA)and belief rule base(BRB)is proposed.Firstly,the incipient fault definition of running gear systems is given,then the collected data is decomposed by orthogonal method.The decomposed data is used for fault detection based on slow feature analysis,which improves the sensitivity to incipient faults of running gear systems.At the same time,BRB method is used to effectively combine expert knowledge with data.The incipient fault of running gear systems is quantified to achieve the purpose of diagnosis.The feasibility of the proposed method is verified by practical engineering data.(3)Aiming at the inaccurate health assessment of running gear systems,a health state assessment method based on parameter optimization is proposed.Principal component analysis(PCA)is adopted to screen feature variables to obtain key features,which reduces the computational complexity of BRB.In addition,due to the incompleteness of expert knowledge,it is hard to give reasonable and accurate initial expert knowledge.By using the adaptive covariance matrix adaptation evolution strategy(CMA-ES)to update it,the health state can be accurately evaluated in running gear systems.