Surrogate-assisted adaptive controller tuning for BLDC motors: an experimental evaluation

Adaptive controller tuning via metaheuristic optimization provides robust compensation against parametric uncertainties and external disturbances, but its high computational cost limits its practical applicability in fast-dynamic systems. This work proposes an experimentally validated surrogate-assisted framework for metaheuristic-based indirect adaptive controller tuning, reducing computational cost while maintaining effective control performance. The methodology adopts a surrogate-assisted integration framework that enables the interchangeability and comparison of surrogate-model strategies. In this work, experimental validation is performed on a brushless direct current motor using three surrogate model techniques: response surface method (RSM), radial basis function (RBF), and Gaussian process regression. A statistical comparative analysis across experiments of increasing complexity evaluates trade-offs between control performance and computational cost relative to a conventional metaheuristic-based tuning strategy. Results confirm computational savings without degrading closed-loop performance. In particular, the RSM variant achieved up to a 43-fold reduction in computational burden, while the RBF variant delivered superior dynamic performance. The results demonstrate the interchangeability of the evaluated surrogate-model strategies within the proposed framework and highlight the practical trade-offs associated with their selection in fast-response adaptive control tuning.