Machine learning-based inverse method for determining elastic coefficients of unsymmetric laminates

Determination of equivalent elastic coefficients in composite laminates is a fundamental problem in structural mechanics and materials engineering. This task becomes particularly challenging for unsymmetric laminates, where coupling effects between membrane, bending, and shear responses render analytical characterisation difficult and computationally expensive. In this work, a machine learning-based inverse methodology is proposed for determining the equivalent shear modulus of unsymmetric composite laminates. A high-fidelity finite element implementation of the ASTM V-notched shear test is used to generate a comprehensive dataset covering a wide range of unsymmetric stacking sequences. A Light Gradient Boosting Machine (LightGBM) model is trained as a nonlinear surrogate to map laminate design variables to the corresponding equivalent shear modulus. The trained model is subsequently employed in an inverse design framework to identify laminate stacking sequences that achieve prescribed target values of shear modulus. The proposed approach significantly reduces computational cost compared to conventional trial-and-error finite element analysis and provides new insight into the relationship between laminate architecture and effective shear behaviour.