Uncertain System State Estimation And Its Application To Adhesion Control

Traction force which is dependence on adhesion between wheel and rail is the ultimate motive force to drive locomotion. In order to increase the utilization of traction, the level of locomotion adhesion need be improved that is achieved mainly by re-adhesion control. The detection of adhesive state is a difficult point. It is affected by the conditions of locomotive design, environment, locomotive operation, and many other factors. Because of adhesion between wheel and rail showing strongly nonlinear and time-varying characteristics when locomotive is running, estimating adhesion state accurately is the precondition of optimize re-adhesion control.Subject to many factors, adhesion is a complex process with great uncertainly. So it is difficult to be described accurately by mathematical model. Classical state observer or standard Kalman filter can't be effectively used in such systems. In this thesis, state estimation in uncertain system and its application to adhesion state estimation was studied. The main contents of this thesis can be concluded as follow:Due to introduction of the fading matrix in traditional Strong Track filter, the covariance of prediction error becomes asymmetric in recursive update which may cause the phenomenon of filtering divergence. In order to solve this problem, an improved Strong Track filter and proof process is proposed. The improved algorithm ensures symmetric of predicting error variance by changing the manner of multiple fading factor. Then, the strong track performance of improved algorithm is proved. The updating algorithms based singular value decomposition is used to ensure convergence.An adaptive Kalman filter algorithm with fading matrix is proposed which is used for the system with un-modeled dynamics. By taking the modeling error as virtual noise, the noise statistical of system is modified online by adaptive method. By introduction of fading factor, algorithm has the ability to track the mutation parameters, and proof process is proposed. At last the updating algorithms based singular value decomposition is used to ensure convergence.Based on the above two kinds of state estimation methods, locomotive adhesion state is estimated. Locomotive speed is estimated when only single-bearing wheel rotate speed can be obtained, and reference wheel speed is estimated when multiple bearings rotate speed signals can be obtained. Disturbance torque dynamic model based, the optimal adhesion coefficient estimation is studied.A decoupled uncertain system filter algorithm which is alternating state estimation and parameter estimation is used to estimate reference wheel speed. Based on multi-rate theory, the optimal adhesion coefficient estimation is studied which provide reference for optimal re-adhesion control.