Background/Introduction <p>Variable cross-section beams are widely used in flexural wave isolation. Conventional periodic designs can only achieve vibration isolation within specific frequency bands, which limits their functionality.</p> Purpose <p>An optimized design model for broadband high-stiffness non-periodic variable-section beams is proposed using machine learning (ML) and genetic algorithms (GA).</p> Methods <p>The model integrates GA and multi-objective GA, combined with ML and the spectral stiffness matrix method (SEM), to enable efficient and precise design of variable-section non-periodic beam structures.</p> Results <p>The results show that the optimized beams can simultaneously suppress vibration across 500 –4000&#xa0;Hz and 4000 –9000&#xa0;Hz. When customizing specific band gaps, considering multi-objective optimization allows the bending stiffness of the aperiodic beam to increase by up to three times.</p> Conclusion <p>The proposed model enables efficient design of high-performance vibration isolation beams for engineering applications.</p>