Fault detection using unknown input observers for heavy-haul trains

In this paper, we consider the problems of fault detection for heavy-haul trains equipped with electronically controlled pneumatic (ECP) brake systems. A longitudinal dynamical model which has been successfully validated is used to simulate the actual situation. Based on the model, a set of unknown input observers which are adopted to estimate locomotives' state is constructed, and observers can determine the existence and place of the faulty locomotive. Since heavy-haul trains are much longer than general passenger trains, the longitudinal dynamical model is decomposed into smaller subsystems which can be detected locally. To estimate the fault parameter after a failure occurred, a minimal extremum seeking algorithm was presented for adaptive approximation. Simulation results provide evidence of the effectiveness of the proposed fault detection scheme.