Topology-preserving shape optimization for inverse modal design using surrogate-assisted memetic algorithms

Abstract This study presents a fully integrated CAD–FEM–ANOVA–ML–Memetic framework for topology-preserving inverse modal design of mechanical components. Natural frequencies are tuned through dimensional shape optimization while strictly maintaining geometric topology and manufacturability. The approach combines parametric CAD modelling, finite element eigenvalue analysis, ANOVA-based feature selection, Support Vector Regression (SVR) and Partial Least Squares (PLS) surrogate modelling, and a Memetic Algorithm (GA + SQP) for global–local optimization. Unlike existing surrogate-assisted frameworks, the proposed method introduces a mode-resolved ANOVA reduction strategy specifically adapted to the highly nonlinear and skewed sensitivities of inverse modal problems, yielding a compact, physically interpretable subset of dominant geometric parameters. Surrogate-assisted optimization reduces the FEM evaluation count by four to five orders of magnitude while achieving sub-percent accuracy in matching all target frequencies. The optimized shaft–disk geometries preserve modal shape integrity and avoid spurious mode switching, demonstrating the physical consistency of the topology-preserving updates. Results confirm that the proposed framework provides a scalable, high-fidelity, and computationally efficient solution for precision-driven modal tuning, with clear potential for extension to multi-objective design, and multi-fidelity modelling.