Fault diagnosis of virtually-coupled trains by adaptive observer with pattern-matched detection and reinforced identification

Virtual coupling is gaining in popularity as a promising development direction to maximize the rail-line efficiency by minimizing the headway distance among trains in the presence of potentially encountered traction engines' faults with unknown amplitude, happening time and probability, which would be huge threat to the safety of trains without proper sensing and handling. This paper considers the fault diagnosis problem for virtually-coupled multiple trains using adaptive observer design. In order to generate false-free and timely fault alarming and relieving signals, an adaptive threshold function design is firstly given using a novel pattern-matched gain technique with explicit consideration of model uncertainty. Then, reinforced regressor-based fault identification algorithm is proposed to generate precise estimation of unknown fault values, activated and powered-off by the fault alarming and relieving signals output by fault detection observer, with globally Lipschitz property and fast convergence performance. Finally, comparative and simulation results are given to demonstrate the effectiveness and advantages of proposed fault diagnosis algorithms.