An approach for controller fault detection

Monitoring and maintaining of control software becomes more and more important and difficult with the increase of control software in size and complexity. In this paper, an approach for control software fault detection is proposed, which is based on the monitoring of the discrepancies between the control outputs of the actual controller and the benchmark controller, a Linear Quadratic Gaussian (LQG) controller. The discrepancies are assumed to be Gaussian distribution with a stable mean under the normal situation. Faults in the actual controller are characterized by sudden jumps in the mean of the discrepancies. The fault detection is transferred into a jump point identification problem. A detector based on Generalized Likelihood Ratio (GLR) test is employed for the jump point identification. The proposed approach is applicable to general control software even it is only illustrated through a water heater case study with a simple PID controller.