State And Parameter Estimation Based Fault Diagnosis Scheme For ECP Brake System Of Heavy-Haul Train

Abstract:The fault diagnosis for ECP (Electronically Controlled Pneumatic) brake system of heavy-haul train, which is an important measure to enhance the train running safety, and there has a great significance in solving the safety braking problem and enhancing the transport capacity of heavy-haul train. In this paper, the ECP system which is a complex distributed nonlinear system should be considered, and its control model will be established, then a fault diagnosis structure based on state and parameter estimation is proposed. An adaptive estimation based fault detection strategy and a neural network based fault isolation and identification scheme are designed, to realize the fault detection, isolation and identification of ECP brake system, and to provide security for the system running.Firstly, a fault detection strategy based on adaptive estimation is proposed, which improves the fault detection performance, and solves the problems such as diagnosis accuracy is not enough and fault false alarm happens due to the system uncertainties. The uncertain signal online estimate method is proposed based on adaptive approximation technique. Then a parametric model with adaptive parameter is used to describe uncertain signal, and an adaptive parameter estimator is constructed by using the base function and approximation error. Through adjusting the parameter adaptive law by a designed update mechanism, the online estimate for uncertain signal such as modeling uncertainty and disturbance is came true. The fault detection state estimator is constructed with the estimate of uncertain signal introduced, to reduce the influence of uncertainty acting on residual and generate robust residual signal, then an adjustable time-varying threshold which can ensure the fault detection performance is derived through the Lyapunov stability theory.Secondly, a fault isolation and identification scheme based on neural network is proposed. Through constructing a neural network parameter estimator, and the weights are updated based on gradient descent method with the goal of minimizing the instantaneous output estimation error, then the iterative learning of unknown fault influence estimation is realized. Using the estimated fault information, the fault isolation state estimator is designed to obtain the estimation or prediction of system state and output, and the condition of suitable isolation threshold selection is derived, then a kind of fault isolation decision-making is proposed based on multiple residual generators to solve the problem of weak influence between faults, finally the fault isolation and identification tasks are realized.Finally, the simulation environment for ECP brake control system is built by using Matlab simulation tool, and then the validity and reliability of the proposed specific fault diagnosis algorithms are verified by the simulation. The hardware platform and software architecture are outlined based on the overall structure, to verify the practical of the algorithms.