Sensor Fault Accommodation for a Plug Flow Reactor using an M-Estimator

Abstract

The presence of gross errors in the measurements can lead to biased state estimates when conventional Bayesian estimators are used. This can hamper the model-based monitoring and control schemes that rely on the accurate state estimates. In this work, we have developed a framework for robust estimation of the state profiles for Distributed parameter systems (DPSs), in the presence of biased measurements. The proposed approach uses an M-estimator to identify the faulty sensor. The sensor fault diagnosis is then used to augment the state estimator with an extra state that estimates the drifting sensor bias. The proposed approach has been applied to an Auto-Thermal tubular reactor system. The proposed scheme successfully isolates the biased temperature sensors and includes or removes additional bias states as and when required. The gross errors/biases are estimated and subsequently accommodated to provide accurate estimates of spatial profiles of reactor concentration and temperature.