Abstract

This work represents a first study towards the application of theory-guided machine learning techniques in the prediction of design allowables of notched polymer composite laminates accounting for material and geometric uncertainties. Building on data generated analytically, using either phase-field methods or finite fracture mechanics, and reduced representations of composite lay-ups, four machine learning algorithms are used to predict the strength of composite laminates with notches of several geometries and the corresponding statistical distribution, associated to material and geometrical variability.

Excellent representations of the design space (relative errors of around ±10%) and very accurate representations of the distributions of notched strengths and of the corresponding B-basis allowables used in aircraft structural design are obtained. Gaussian-based models proved to be the most reliable approach as a result of its continuous nature, accuracy, and fast training process. This work serves as basis for the prediction of first-ply failure, ultimate strength and failure mode of composite laminates based on non-linear finite element simulations across different length scales, providing significant reductions of the computational time required to virtually certify composite aircraft structures.