Answers are grounded in this paper and cite their sources.
Structural optimization of permanent-magnet bridged reluctance actuators (PMBRAs) requires costly repeated finite-element analysis (FEA). Multimodal optimization must also preserve structurally distinct designs with comparable objective values. Here, we develop a surrogate-assisted multimodal multi-objective optimization framework combining a deep neural network (DNN) force-per-unit-depth surrogate with SPD-enhanced DN-NSGA-II. A full-factorial dataset of 4096 two-dimensional FEA samples from four structural variables, \(N_c\) , \(N_n\) , \(d_e\) , and g , was used to train a 64-32-16 DNN to predict \(q_F\) in \(kN/m\) , with actuator thrust calculated as \(F=q_F w_m\) . The surrogate was evaluated using a fixed hold-out test, deterministic five-fold cross-validation, and a 168-point intermediate-grid validation. Over a stack-depth range of 37.6–60.0 mm, the five-fold MAE and RMSE were 10.04–16.03 N and 13.20–21.06 N, while the intermediate-grid errors were 3.95–6.30 N and 5.74–9.17 N. The DNN also outperformed five low-complexity surrogate baselines in MAE and RMSE under both validation protocols. The surrogate was coupled with SPD-DN-NSGA-II, where an FE-adaptive SPD offspring branch improves preservation of multiple Pareto-set branches in decision space. On 22 CEC2019 benchmark problems under equal function-evaluation budgets, significant suite-level differences were observed for rPSP, rHV, IGDX, and IGDF, with the clearest gains in rPSP and IGDX. Applied to the mixed-integer PMBRA problem, the framework revealed trade-offs between thrust density and the copper-loss-based force-to-power ratio while retaining alternative structural solutions under different air-gap and performance constraints.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: