Data-Driven Optimal Tuning of BLDC Motors With Safety Constraints: A Set Membership Approach

Abstract

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Field-Oriented Control</i> (FOC) is among the most popular control architectures for brushless BLDC motors, employed in several mechatronic applications. Data-driven strategies allow for model-free, optimal tuning of FOC parameters, optimizing a quantitative performance index. While fast, noniterative data-driven techniques, such as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Virtual Reference Feedback Tuning</i> (VRFT), are sensitive to the choice of the training experiment and the desired closed-loop behavior. On the other hand, iterative data-driven techniques represent a more robust approach, with less critical experiment design and the ability to account for the presence of nonlinearities. However, commonly used iterative algorithms, such as Bayesian <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Optimization</i> (BO), are often computationally expensive, and require <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">caution</i> in the selection of the parameters to avoid instabilities in closed-loop experiments. The contribution of this work is to formulate the tuning problem of FOC parameters as a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">model reference</i> optimization problem suitable to be solved with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Set Membership Global Optimization-<inline-formula><tex-math notation="LaTeX">$\Delta$</tex-math></inline-formula></i> . This novel, iterative algorithm allows one for the specification of safety constraints and is computationally more efficient than BO. An extensive experimental analysis on a real setup confirms the effectiveness of the proposed approach, and shows that a safe warm start based on VRFT yields faster convergence to the optimal parameters.