Artificial neural network to predict the structural compliance of irregular geometries considering volume constraints

Abstract

This study employs artificial neural networks (ANNs) to predict the structural compliance of randomly generated irregular geometries derived from Finite Element (FE) calculations. By imposing volume constraints, the scope of the study is confined to applying ANNs for learning from structural data generated by considering either multiple random walks of a circle or a set of randomly placed circles with allowed overlaps. Numerical results indicate that the learning outcomes of the former approach are more satisfactory than those of the latter. This suggests that the effectiveness of employing ANNs for predicting the structural compliance of irregular geometries is contingent upon how the random geometries are generated and the material volume ratio. The learning outcomes of irregular structures generated by the former approach with a higher volume ratio exhibit greater satisfaction due to a higher degree of structural connectivity.