A control performance monitor and a model performance monitor for process controllers

Scope and method of study. A control performance monitor and a model performance monitor for process controllers are developed. The control performance monitor automatically detects poor control performance. It compares the distribution of a performance index, called run length RL, to a reference RL distribution built from the data collected during a good control period. A statistical chi-square test is performed on the differences of the two RL distributions. It uses routine plant operation data only. The model performance monitor automatically detects a poor process model for a process model-based controller. It suggests model adjustment only when a statistical F test indicates that the benefit of model adjustment, i.e., the improvement in performance of the model-based controller, is significant. The process model is judged by its function, i.e., the performance of the model-based controller, not by the accuracy of model prediction. A reference performance system is constructed from the closed-loop step response to a setpoint change of the control system consisting of the initial well-tuned controller and the initial plant. A performance index, the variance ratio RV of the controlled variables of the current control system and the reference performance system, is proposed. Both monitors are evaluated by experiments on a water flow control loop and an air flow control loop and simulations.; Findings and conclusions. Experiments and simulations show that the control performance monitor can detect poor control performance, such as, oscillations, too aggressive or too sluggish responses to disturbances or setpoint changes, and hitting a constraint, and the model performance monitor can flag a poor process model for a process model-based controller due to plant changes. The model performance monitor does not flag a poor process model when the process is at steady state. Both monitors are simple to use, and are non-invasive to plant operations.