Bayesian optimization-based parameter search space determination for performance enhancement of feedback controller auto-tuning

In industrial mechatronics, achieving high-speed and high-precision control of positioning mechanisms requires precise tuning of feedback controllers to meet stringent control specifications. Consequently, the demand for auto-tuning solutions that streamline the tuning process has increased. A key challenge in auto-tuning feedback controllers is defining a sufficiently large parameter search space that ensures stability and enables the identification of optimal control parameters, all without risking system divergence. This study introduces an innovative approach for determining the parameter search space, leveraging Bayesian optimization to facilitate stable and high-performance auto-tuning of feedback controllers. In the proposed method, the true stable parameter space is accurately and efficiently estimated as the search domain for auto-tuning, utilizing a frequency-domain stability criterion-aware Bayesian optimization. Compared with conventional methods based on solvability analysis, our approach enables the exploration of a broader parameter search space, thereby enhancing the potential for improved positioning performance through auto-tuning. The effectiveness of the proposed method has been validated through auto-tuning experiments on a galvano scanner with PID gain tuning, providing a fundamental validation under a specific and low-dimensional auto-tuning scenario.