A methodology to generate design allowables of composite laminates using machine learning

This work represents the first step towards application of machine learning techniques in prediction statistical design allowables composite laminates. Building on data generated analytically, four algorithms (XGBoost, Random Forests, Gaussian Processes and Artificial Neural Networks) are used to predict notched strength laminates their distribution, associated uncertainty related material properties geometrical features. focuses not only so-called Legacy Quad Laminates (0°/90°/±45°), typically aerostructures, but also newer concept double-double (or double-angle ply) Very good representations space, translating low generalization relative errors around ±10%, very accurate distributions strengths single points corresponding B-basis obtained. All algorithms, with exception show performances, outperforming others for small number while Networks have better performance larger training sets. serves as basis first-ply failure, ultimate failure mode specimens based non-linear finite element simulations, providing further reduction computational time required virtually obtain