A Non-Dominated Sorting-Based Multi-Objective Neural Network Optimization With Constraints for H_∞-Control Tuning of Rotor–AMB Systems

Tuning model-based controllers for complex multivariable systems is inherently a multi-objective problem involving competing performance and robustness requirements. Data-driven and learning-based optimization methods have recently emerged as promising alternatives to classical evolutionary approaches; however, their applicability to practical control problems is often limited by the lack of explicit constraint handling. This paper proposes a constrained non-dominated sorting-based multi-objective neural network optimization algorithm (MONNA)-based framework for multi-input multi-output (MIMO) control design. The proposed framework is demonstrated by considering of aH_∞–based control design for a radial active magnetic bearing (AMB) rotor system, where multiple conflicting objectives and constraints are taken into account during controller synthesis. The obtained results are compared against a multi-objective genetic algorithm (MOGA) based on NSGA-II. Controllers selected from the Pareto front are evaluated through simulations and experimentally validated, demonstrating that the proposed constrained MONNA method effectively identifies feasible, high-performance controllers while maintaining robustness and closed-loop stability.