Failure detection of control actuators using CKF based multimodels: An experimental study

Fault detection is a concern in many industrial applications. Several approaches have been used. One based on a single reference model that describes the expected nominal dynamic of a healthy system around a certain operating point is largely used. The system is considered healthy if its measured signals are around the nominal behavior, a fault is triggered otherwise. The first challenging task is how ot determinine the acceptable deviation margins. But most importantly, such approach may lead to the detection of false failures due to several operational reasons such as a nnatural move from one operating region to another in the operating space, etc. In this study, a multimodel based fault detection, diagnosis, and isolation is considered. The proposed approach has several imprtant features. A combination of several local model types can be used. The approach can be applied to several type of control actuators failures. Experimental application of the proposed approach to control valve failure demonstrate the practicality of the proposed method.