Data-based monitoring and reconfiguration of a distributed model predictive control system

In this work, we develop a data-based monitoring and reconfiguration system for a distributed model predictive control system in the presence of control actuator faults. Specifically, we first design fault detection filters and filter residuals, which are computed via exponentially weighted moving average, to effectively detect faults. Then, we propose a fault isolation approach which uses adaptive fault isolation time windows to quickly and accurately isolate actuator faults. Simultaneously, we estimate the magnitudes of the faults using a least-squares method and based on the estimated fault values, we design appropriate fault-tolerant control strategies to handle the actuator faults and maintain the closed-loop system state within a desired operating region. A nonlinear chemical process example is used to demonstrate the approach.