Machine learning assisted stochastic progressive failure analysis of composite laminates in a meso-macro framework

Quantification of uncertainty in composite materials has been a challenge in terms of complexity and computation time. This is due to the nonlinear behaviour of composite materials and multiple failure mechanisms occurring simultaneously. This study develops a high fidelity surrogate model to quantify the uncertainty in matrix cracking in 90-degree plies of a composite laminate efficiently. The surrogate model is trained by continuum damage mechanics-based user subroutine (UMAT) coupled with the gaussian processes assisted finite element method. High fidelity surrogate model-based uncertainty propagation can effectively replace physics-based models and the global response of the composite laminates can be predicted accurately and cost-effectively. Using the proposed computational model, progressive failure of blunt-notched GLARE specimen is investigated considering stochasticity in applied strain following a multi-scale framework.