Monitoring Model Drift in Feedback-Controlled Systems via Efficacy and Efficiency Metrics

Abstract

This work explores the possibility of monitoring model drift inside a control loop by leveraging efficacy and efficiency metrics. The methodology is based on a fundamental energy–error relationship that links tracking performance to the control effort required to sustain it.
The proposed strategy is explicitly designed to deal with feedback-controlled systems and relies only on basic loop signals and simple indicators on time windows: a control-energy metric and an error-performance metric. Based on the similarity between model drift and fault detection problems, the method enables early detection and tracking of progressive faults or drifts that would otherwise remain hidden. The monitoring relies on residuals obtained by comparing the joint evolution of the two metrics against a nominal-condition baseline model. The approach is demonstrated through a MATLAB/ Simulink simulations across various degradation scenarios, showing consistent sensitivity to drifts and enabling their timely detection well before loss of performance becomes apparent at the output level. These results support a lightweight, explainable pathway for model drift monitoring in feedback-controlled systems without requiring additional sensors, the development of complex machine learning/high-fidelity physics models, or structural modifications to the control architecture.