Abstract:

An important problem in the control of complex systems, and one that inherently exhibits hybrid dynamics, is fault management and recovery. If a component faults in a system, the continuous behavior will suddenly change. In an autonomous system, the controller must be adaptive to ensure continued useful operation when faults occur.

Fault-adaptive control encompasses a number of hard problems: the detection of a fault, its identification and assessment, the selection of a new control algorithm, the reconfiguration of the plant to "disconnect" the failed component, and the launching of the new control algorithm. The authors propose a model-based approach for designing control systems that are capable of accommodating faults. This chapter focuses on one aspect of this approach; however, an overview of the fault-adaptive control architecture is also given.

The authors model a plant as a hybrid bond graph. Bond graphs are energy-based models, and they are extended for hybrid dynamics by representing mode transitions as controlled junctions. A hybrid observer has been developed that uses this model to track the system behavior within and across modes.

Two complementary approaches to fault detection and isolation are outlined, one based on hybrid models, the other on discrete-event models. The former makes use of  the hybrid observer, qualitative reasoning techniques, and real-time parameter estimation. The discrete approach uses failure propagation graphs. Controller reconfiguration relies on a previously developed controller library - an appropriate controller is selected based on current conditions. The authors discuss ideas for mitigating the large transients that can arise during controller switching. A two-tank system is used as an example throughout the chapter.

Download: KarsaiEtAl_SEC2003.pdf