SEAT S.A., a car manufacturer based in Catalonia and a subsidiary of the Volkswagen Group, aims to develop ARCO, a tool based on the Proper Generalized Decomposition (PGD) machine learning methodology for the parametric NVH (Noise, Vibration and Harshness) analysis of a vehicle body, taking into account material and geometric design variables. This analysis is critical to ensure that the product meets noise and vibration criteria, improving comfort, quality, and customer satisfaction. The project is the natural continuation of previous developments carried out in Matlab between SEAT and LaCàN, using the ePGD library developed by inLab FIB in C++ and Python, scaled in parallel with PETSC and OpenMPI.

The first work package focuses on automating the data input process and the computation of the PGD module previously developed during the ARCO project, including automating the entire workflow from preprocessing to the final solution, establishing the logical structure of the code, classes, and methods, generating data matrices and configuration files, introducing a validator to ensure the correct generation of data extracted from Nastran, adding an optimizer for the final solution, and ensuring code maintainability and scalability.

The second work package aims to increase dimensionality, starting with the definition of the parameters to be considered and the sampling strategy across multiple dimensions, studying the spatial location of the effect of each parameter and designing an initial set of samples to verify behavior, followed by generating the matrices corresponding to the chosen sampling and constructing the separated tensors, and finally solving the parametric problem and building the PGD surrogate in multiple dimensions, with the goal of generalizing the process to other geometries and parameters in the future.