A Multi-Fidelity Cuf-Based Framework for Surrogate Modeling in Composite Virtual Manufacturing

Composite materials are widely used in aerospace structures; however, manufacturing remains a critical and complex phase that strongly affects final structural performance. Process-induced deformations and residual stresses arise from coupled chemical, thermal, and mechanical phenomena that are difficult to predict accurately. Virtual manufacturing tools are therefore widely used, yet conventional numerical approaches still present major limitations. Accurate solution of the coupled chemo-thermo-elastic problem often requires fully three-dimensional solid models, leading to prohibitive computational costs, while simplified models may neglect relevant physical mechanisms and reduce predictive capability. Surrogate models are increasingly used to accelerate design and optimization processes. However, their reliability depends on the availability of large and consistent training datasets. High-fidelity simulations are too expensive to generate sufficient data, especially for multiphysics problems, while experimental campaigns provide only limited and costly information. This work proposes an advanced numerical framework for composite virtual manufacturing based on the Carrera Unified Formulation (CUF). CUF enables arbitrary kinematic expansions and hierarchical model refinement, allowing the systematic generation of models with different levels of accuracy and computational cost. This defines a multi-fidelity modeling environment, where low-order models efficiently populate the design space and higher-order refinements provide accurate solutions in selected regions. Closed-form solutions, Equivalent Laminate, Equivalent Single Layer and Layer Wise models can be adopted. A weighted Gaussian regression network is employed to consistently integrate multi-fidelity numerical data with, eventually, a limited set of experimental results. The weighting strategy accounts for the different reliability levels of the datasets, enabling the surrogate model to preserve the global physical trends captured by simulations while correcting local discrepancies through higherfidelity and experimental information. Results demonstrate that the proposed multi-fidelity framework delivers accurate predictions of residual stresses and distortions across the design space while requiring high-fidelity analyses only at a limited number of points. The inclusion of a small number of experimental measurements further reduces the gap between simulation and tests. The methodology provides an efficient and scalable strategy for surrogate-based design and virtual manufacturing of aerospace composite components.