Machine-Learning-Assisted Inverse Design of Centrally Thickened Plates for Vibration, Buckling, and Mass Efficiency

This study proposes a machine-learning-assisted inverse design framework for simply supported rectangular plates with a centrally located thickened region, aiming to improve vibration and buckling performance simultaneously while minimizing structural mass. A verified finite-element model based on SHELL181 elements was used to generate a 150,000-case development database, while an additional 50,000 independently generated cases were held out for final testing. The fundamental natural frequency and first critical buckling load were normalized with respect to corresponding uniform plates, yielding the performance ratios R_f and R_b. Surrogate models were trained to predict these responses and then incorporated into a constrained inverse design procedure. The inverse search identified the lowest-mass feasible candidates within the screened design pools satisfying prescribed frequency and buckling targets, after which candidate solutions were re-evaluated using finite-element reanalysis. Because surrogate overprediction near active constraints may produce designs incorrectly classified as feasible, a conservative second search was introduced using fixed guard bands of 0.020 for R_f and 0.040 for R_b, selected slightly above the largest positive deviations observed during the initial FEM recheck. Final FEM verification confirmed the feasibility of the refined designs. The framework provides an efficient and computationally practical strategy for multi-constraint inverse design of locally thickened plates.