Parameter Estimation Of High Speed Train Running Bogie Using RAO-Blackwellised Particle Filter

In this thesis, parameter estimation for railway vehicle running bogie is discussed. To support condition-based maintenance based on fault diagnosing of vehicle, a CRH380A high-speed railway vehicle lateral state space model is built. The Rao-Blackwellised Particle Filter(RBPF)-based method is used for parameter estimation. However, the standard RBPF-based method does not adapt to non-Gaussian noise when verified using the real track irregularity as the input of model instead of Gaussian noise. An improved RBPF estimation method is introduced, which can estimate parameters with real track irregularity.First of all, the method of Kalman filter is introduced and the model of a high-speed railway vehicle is transferred into state-space equations. Then we try to estimate the parameters of the model using Extended Kalman Filter, the results of which show that Extended Kalman Filter can not handle the estimation perfectly when the initial state was set with large deviation value.Thus, a more efficient particle filter called Rao-Blackwellised particle filter is used in this study. The RBPF method is based on the theory of marginalization that combines a particle filter to compute distribution of the parameter with a bank of Kalman filters to calculate the distribution of the state vectors. The advantage of this strategy is that it can reduce the size of the sample space drastically.However, simulation results show that the RBPF is ineffective when subjected to non-Gaussian track irregularities. We propose the use of time-series theory and state augmented technique to improve the RBPF method for parameter estimation and fault diagnosis in the high-speed railway vehicle system. Apparently, the results of the new algorithm show a good capability to deal with non-Gaussian noise.