Process-induced distortions in composite materials compromise the dimensional accuracy of components. These distortions arise from complex interactions that make them challenging to predict numerically. To address this, surrogate models provide a practical solution by enabling rapid predictions without requiring computationally intensive multiphysics simulations.
This work introduces a synthetic data generation workflow developed to support the creation of such models. The approach focuses on predicting distortion fields in L-shaped composite plates and is based on high-fidelity simulations that capture the relevant multiphysics at the mesoscopic scale. The numerical framework is implemented in Alya, a finite element solver optimised for HPC systems, and validated against experimental data to ensure the reliability of the generated dataset. PyCOMPSs is used to orchestrate task-based parallel execution across HPC nodes, ensuring efficient use of resources and scalability for large-scale data production.
A surrogate model was trained using data generated for different ply stacking sequences to demonstrate the application of the workflow. The results confirm the effectiveness of the proposed approach in reducing the computational cost associated with the design of composite structures while maintaining predictive accuracy.
A High-Fidelity HPC Workflow for Predicting Process-Induced Distortions in Composites Using Surrogate Models 