Self validating control algorithm for plants with primary sensor failure using multiple estimators

In any closed loop system, the controller acts on an error signal; the reliability of the algorithm representing the controller depends to a large extent on the exact value of the error signal. In the event of sensor failure/inaccuracy, the reliability of the control algorithm is lost and the overall system may even become unstable. We use a supervisory expert control system (SECS) which allows intelligent decisions to be taken suited for fault tolerant control. The SECS provides a conceptual structure and captures the aspects of the process which accurately represent its behaviour. A self validating control algorithm with two estimators is presented; the reliability of the sensor is monitored continuously by comparing its output with the trained estimator output before it is presented to the controller. To take care of exogenous disturbances in the primary process variable during sensor failure, the effect of interaction between primary and secondary process variables is studied and, using the interaction present, an intelligent network is built which can estimate disturbances in the primary process variable and maintain the primary process variable at the steady state value. The two estimators help in appropriately altering the guidance laws characterizing the desired response during the time of sensor failure. Experimental results show that the system remains stable even in the case of primary variable sensor failure. The performance of the overall system is measured by comparing the MSE for two types of estimator, a neural network and a least square estimator.