Multi-fidelity probabilistic failure onset analysis of composite structures under uncertainties

Advancements in composite manufacturing have enabled innovative design strategies to enhance stress distribution, stiffness, and overall performance of composite structures. However, uncertainties related to material variability, load fluctuations, and manufacturing defects continue to pose serious challenges for structural reliability and early failure prediction. This study presents a novel multi-fidelity probabilistic framework for analyzing composite laminates under uncertainties. The methodology integrates low- and high-fidelity structural theories generated via the Carrera Unified Formulation (CUF). An adaptive Gaussian Process Regression (GPR) model is employed to construct a probabilistic surrogate that selectively uses high-fidelity theories only where needed, based on uncertainty-driven learning. Compared to conventional Monte Carlo Simulations (MCS) based entirely on high-fidelity models, the proposed framework achieves comparable accuracy, with R2 > 0.99 and nRMSE < 1%, while reducing the computational cost by a factor of two to ten. Convergence is obtained with only 20–300 high-fidelity simulations, against 400–800 required by the reference benchmark. The approach is applied to a composite plate, a free-edge laminate, and an open-hole configuration, where the adaptive multi-fidelity model accurately captures through-thickness stresses and failure indices.