Detection and Isolation of Plant Failures in Dynamic Systems

This paper presents a method of distinguishign between actuator failures and changes in the system dynamics. The method is based upon exploitation of the system input. In particular, a class of control laws are derived that when used with conventional estimators/observers, enable isolation of actuator calibration changes from shifts in the plant dynamics. It is assumed that any calibration change or shift in the plant dynamics is detected by some form of detection scheme (e.g., a Beard-Jones Detection Filter). This paper demonstrates the subsequent procedure for distinguishing between these two events. Theorems are presented for the existence and characterization of these special inputs, referred to as designed inputs. Other issues, such as operator freedom and robustness to system noise and errors, are analyzed and discussed. To illustrate this method, an example is presented involving failures in an automobile having 4-wheel electronically controlled steering. For this example, a designed input is constructed to isolate rear wheel actuator failures from changes in the tire cornering force parameter.