In today’s competitive market environment, it is important to maximize economic performance of a processing plant in the face of slowly drifting model parameters and unmeasured disturbances. Economic nonlinear model predictive control (ENMPC) has emerged as a promising alternative for merging economic and control objectives, as the conventional combination of real-time optimization (RTO) and MPC schemes can lead to suboptimal performance due to long wait times and model-plant mismatch. However, ENMPC approaches available in the literature either assume that model parameters or unmeasured disturbances are measured or choose to update a fixed set of model parameters online. This work presents an integrated framework that combines with two recently developed intelligent Bayesian state and parameter estimation schemes, namely, moving horizon estimation (iMHE) and extended Kalman filter (iEKF), which use online fault diagnosis and identification (FDI) for automated online model maintenance. These intelligent estimators are capable of isolating the subset of actively changing faults (i.e., drifting parameters or disturbances) and compensating for these faults until they continue to change. In addition, these estimators can differentiate between faulty (biased) sensors and actuators and parameter/disturbance variations and take appropriate corrective action in estimation and optimal control as and when needed. Simulation studies carried out on the benchmark Williams–Otto reactor process demonstrate that both integrated schemes achieve accurate fault diagnosis and maintain economic objectives under faulty conditions. In particular, the iMHE was found to generate superior FDI performance when compared with iEKF. Also, comparative performance evaluation with the conventional combination of RTO and NMPC highlights the advantage of the proposed fault tolerant ENMPC in sustaining profitability, especially under dynamic disturbances and fault scenarios.
Bagla et al. (Fri,) studied this question.