Fault Diagnosis Method For Running Gear Systems Of High-Speed Trains Based On Kalman Filter

Running gear systems is an important role to ensure the stable operation of highspeed trains.With the high intensity service,some key parts of the train are prone to aging,wear and tear.If the fault can not be discovered and dealt with in time,it will lead to train outages and serious accidents,which will bring great hidden dangers to the safety of passengers.Based on the background of high-speed train running gear systems,this paper completes the fault diagnosis task for running gear systems using the modelbased method.For solving the problems of practical diagnostic applications of Kalman filter framework,this study has been accomplished the following parts:(1)In the first part of the paper,the difference of multi-sensor monitoring information has solved by proposing a distributed state estimation filter.In the process of information collection,the impact of other nodes around target node is focused for improving the consistency of multi-sensor systems.In addition,this study considers the incomplete information of initialization and colored noise of sensor measurement for increasing the accuracy of state estimation.At the end of this part,the method of multisensor filter is verified by a numerical simulation model of the additive fault.(2)In view of the degradation characteristics of the high-speed train running gear systems,the optimal state estimation under the process of long-distance correlation is studied.The Hurst index is used to extract the long-range correlation part of the data,and the correction value is designed to realize the deviation correction of the actual system state estimation.Then,the distributed fault diagnosis is realized based on the distributed state estimation model proposed in the first part.The actual temperature data of the running gear systems is used in this study,for demonstrating the effectiveness of the method.(3)For the state coupling problem of multi-source monitoring in the running gear systems,the temperature control model of the motor is established.The Sigma-fused unscented Kalman filter is proposed for estimating the state of coupling characteristics.Moreover,the model of performance degradation is constructed based on Levy,and the decomposition method of jump-diffusion process is given.An the end of this section,the effectiveness of the method is verified by the temperature data of traction motor in high-speed trains.(4)For the problem of fault identification in dynamic systems,this study proposes a collaborative deep learning framework.Fault detection is accomplished by the first neural network and fault identification is finished by the second neural network.The Kalman filter method build a bridge of neural networks which running cooperatively by the residuals.At last,the effectiveness of the method is verified by the platform of traction control systems.