Sliding mode methods for fault detection and fault tolerant control

Abstract

Sliding modes in dynamical systems have been historically studied because of their strong robustness properties to a certain class of uncertainty. In feedback control systems, this is achieved by employing nonlinear control/injection signals to force the system trajectories to attain, in finite time, a motion along a surface in the state-space. The associated reduced order dynamics the system exhibits, whilst constrained to the surface, is called the sliding motion. This motion possesses strong robustness properties to so-called matched uncertainty. These nonlinear injection techniques can also be applied to observer problem formulations, and result in intriguing properties. This talk will consider how sliding mode ideas can be exploited for fault detection (specifically fault signal estimation) and fault tolerant control. In particular the talk will describe how sliding mode observers can be used for fault estimation in dynamic systems by exploiting the ‘equivalent injection’ signal necessary for maintaining sliding. The talk will attempt to demonstrate the practicality of these methods with examples of applications of these ideas to aerospace systems. This will include results demonstrating the successful real-time implementation of a sliding mode fault tolerant control scheme on a motion flight simulator configured to represent the aircraft associated with the El-AL Bijlmermeer incident.