Integrated fault monitoring and reliable control

Successful use of a reliable controller requires a fault detector to detect sensor faults such that the sensor can be removed from the closed-loop if it is faulty. If it could be expected that a sensor would completely fail such that it would not emit any signal then a fault detector would not be required and the reliable controller should operate as it was designed for this situation. Integrating a fault detector with a controller introduces tradeoffs that have to be made between the controller and fault detector performance. This can make the H^∞ design harder since weighting choice is more crucial. Where there is more than one fault to be detected there must be frequency separation between the fault models and the other weightings which further restricts a successful design. It was shown by Niemann and Stoustrup (1997) that for uncertain systems the integrated design of a controller a fault detection filter should have a smaller fault error than separate designs. In this paper the integration of a fault detector with a reliable controller was explored. The convenience of a single H ^∞ design must be balanced against the complexity of that design and the difficulties in choosing frequency weighting functions to satisfy all the design objectives