Data-Driven Servomechanism Design from Noisy Data

Abstract

This paper proposes a novel method for data-driven servomechanism design based on data informativity. According to the internal model principle, the feedback controller must include an integrator so that the plant output robustly tracks a step reference signal without steady-state error. We derive a linear matrix inequality (LMI) condition characterizing data informativity for servomechanism design under the assumption where the plant’s response data is contaminated by bounded noise. The LMI condition enables us to efficiently design a servomechanism controller from noisy data, using a convex optimization technique. An important feature of the proposed LMI is that, by explicitly accounting for the aforementioned internal model structure, it is formulated directly in terms of the plant’s response data without requiring a plant model.