Sliding modes for fault detection and fault tolerant control

This chapter will describe the use of sliding mode ideas for fault detection leading to fault tolerant control. The fundamental purpose of a fault detection and isolation (FDI) scheme is to generate an alarm when a fault occurs and to pin-point the source. Fault tolerant control (FTC) systems seek to provide, at worst, a degraded level of performance (compared to the fault free situation) in the event of a fault or failure developing in the system. This chapter will discuss how sliding mode methods for control system design and observer design, can be advantageously used for such schemes. The sliding mode observer FDI schemes seek to robustly estimate any unknown fault signal existing within the system based on appropriate scaling of the equivalent output estimation error injection signal. Both actuator fault and sensor fault problems are considered. One advantage of these sliding mode methods over more traditional residual based observer schemes is that because the faults are reconstructed, both the 'shape' and size of the faults are preserved. In the absence of modelling discrepancies, the faults would be reconstructed perfectly. In the uncertain case, the thresholds set for the reconstruction signals for alarm purposes, correspond directly to the level of faults than can (or must) be tolerated. A further benefit of this approach is that because faults are reconstructed, these signals can be used to correct a faulty sensor for example, to maintain reasonable performance until appropriate maintenance could be undertaken. This 'virtual sensor' can be used in the control algorithm to form the output tracking error signal which is processed to generate the control signal. In particular the chapter discusses recent advances which seek to obviate the traditional relative degree one minimum phaseness conditions. Also the effects of unmatched uncertainty are discussed. In all the methods proposed, efficient Linear Matrix Inequality methods are employed to synthesis the required gains. A recent application of sliding mode controllers for fault tolerant control is also presented. Here the inherent robustness properties of sliding modes to matched uncertainty are exploited. Although sliding mode controllers can cope easily with faults, they are not able to directly deal with failures - i.e. the total loss of an actuator. In order to overcome this, the integration of a sliding mode scheme with a control allocation framework is considered whereby the effectiveness level of the actuators is used by the control allocation scheme to redistribute the control signals to the 'healthy' actuators when a fault occurs.