Two approaches for the isolation of plant failures in dynamic systems
Résumé
One of the most significant issues in fault detection and isolation using analytical redundancy is the problem of robustness to modelling errors. The purpose of this communication is to present and examine design methods for isolating plant failures from sensor and actuator failures. Failures in actuators are readily detected and isolated provided the plant dynamics remain consistent. However, any changes in the plant such as that due to change in operating condition on component confuse the isolation of actuator failures. To first approximation plant dynamic, changes can be represented by a small change in the state transition matrix of the linearized model. This paper identifies a method of distinguishing this change in plant dynamics from the actuator failure. Two methods are proposed: the first one which consists to design the inputs of the process for the particular task of isolating plant failures and the second which consider the failure directions as unknown inputs which are eliminated from the state equation by projection