Towards Empirical Transfer Function Estimation for Frequency Domain Closed-loop System Identification

This work introduces a method for closed-loop system identification using frequency analysis, employing Empirical Transfer Function Estimation (ETFE). By integrating optimization within a Monte Carlo framework, it enhances the precision of ETFE, resulting in minimized frequency response errors compared to actual system data. Leveraging controller information in an offline model fitting scheme, it achieves optimal realization of process dynamics. The method is evaluated on a data center rack-level cooling system, showing Bode magnitude plots of actual and estimated closed-loop and open-loop dynamics, with confidence intervals demonstrating algorithm consistency. Numerical evaluations confirm the feasibility and potential of the approach to improve offline closed-loop system identification performance in the frequency domain, beneficial for analysis and design. There will not be a comparative study for the introduced approach.