General Sensitivity Analysis (GSA) using CAELESTIS HPC-based workflows

### Introduction The certification of composite structures in the aeronautical sector follows a building-block pyramid approach, in which structural complexity, material uncertainties, and computational cost increase progressively across scales. To reduce reliance on extensive experimental campaigns, high-fidelity finite element simulations are increasingly employed for virtual testing. However, the propagation and management of material uncertainties in such simulations remain computationally demanding, particularly when large parametric studies are required. ### Workflow and Use-Case This example employs the `SENSITIVITY_ANALYSIS` workflow available in the [CAELESTIS Workflows repository](https://github.com/CAELESTIS-Project-EU/Workflows). The workflow performs a Global Sensitivity Analysis (GSA) using the Morris screening method on high-fidelity finite element models to identify the material parameters governing structural failure. The methodology and application are based on [Sasikumar et al. 2023](https://journals.sagepub.com/doi/10.1177/00219983231163272). The workflow operates in a fully parallel environment, using PyCOMPSs as a task-based orchestration framework and Alya Multiphysics as a parallel finite element solver. This combination enables large-scale uncertainty quantification through High-Performance Computing (HPC), allowing extensive parametric exploration while maintaining computational efficiency. ### Results and Conclusions The proposed framework successfully identifies the critical material parameters influencing the failure load of an open-hole tensile specimen. The results demonstrate that integrating Global Sensitivity Analysis with high-fidelity damage modeling and HPC-based orchestration provides an effective strategy for uncertainty management in composite structures. This approach contributes to reducing experimental effort while preserving predictive capability. Furthermore, the fully parallel workflow enables efficient distribution of simulations across computational resources, significantly reducing time-to-solution while ensuring numerical robustness and scalability. ### Acknowledgements This work has been developed within the CAELESTIS project. The project has received funding from the European Union's Horizon research and innovation program under Grant Agreement No. 101056886.