Reconstruction-based interval observer dedicated to fault detection: Application to a throttle valve

Model-based fault detection is a very wide field of research. Sometimes, uncertainties within the model create deviations affecting fault detection. Several methods have been developed in order to overcome this problem. In this paper, a very well-suited observer for uncertain fault detection, called set-valued observer is investigated. However, this observer needs all the sensors to work properly. Other works have focused on the well-known Luenberger observer applied to uncertain systems, but this architecture has its own limitation. Indeed, the recursivity of the algorithm induces the wrapping effect. Here an approach coupling the Luenberger observer with the set-valued observer is proposed in order to reconstruct the state sets without divergence. The proposed observer is implemented for the purpose of fault detection on a throttle valve. The nonlinearities of the uncertain system induce a loss of information and require the use of a time-consuming subpaving method. To overcome this problem, the multimodel approach is used to obtain several linear models. In this paper, different types of faults will be addressed: actuator, system, and sensor faults. The proposed observer is implemented and computed on a real-time benchmark. The results obtained are encouraging and bring to light areas for future work. Copyright (c) 2015 John Wiley & Sons, Ltd.