Development of an integrated monitoring system for model predictive control

Yoonki Lee

Model Predictive Control (MPC) is widely used in industrial systems such as chemical processes, energy systems, and semiconductor manufacturing. It predicts future process behavior and calculates optimal control actions, allowing complex systems to operate more safely and efficiently. MPC is especially useful for multivariable systems with constraints, which is why it plays an important role in many real industrial processes. However, in practice, controller performance often degrades over time due to disturbances, process changes, plant-model mismatch, and sensor problems. Checking these issues manually requires a lot of time and experience, and it becomes very difficult when many MPC loops are running at the same time. For this reason, a monitoring system is needed to automatically evaluate whether the MPC is performing well, detect performance loss, and determine when model re-identification or controller retuning is necessary. My research focuses on developing an integrated MPC monitoring framework that not only evaluates controller performance but also detects abnormal behavior early and helps diagnose the cause. The final goal is to make MPC more intelligent and practical by building a system that can monitor itself and suggest when improvements are needed.