Advanced applications in virtual reality for the optimization of design methodologies in car design

This Ph.D. dissertation investigates the integration of Response Surface Methodology (RSM) and Artificial Neural Networks (ANNs) within the early stages of vehicle product architecture definition. The main objective is to establish a systematic, data-driven framework (i.e. methodology) for exploring and optimizing design solutions before the detailed CAD modeling phase, thus reducing development time and computational effort while enhancing design robustness. After introducing the theoretical background of Design of Experiments (DOE), RSM, and ANN modeling, the research proposes a hybrid methodology capable of combining the statistical interpretability of RSM with the non-linear approximation power of neural networks. The resulting workflow allows designers to identify the most influential parameters, predict vehicle dynamic responses, and determine optimal or sub-optimal design configurations across multiple objectives. Several case studies are presented to validate the proposed approach, including acceleration tests and suspension system analyses with multivariable input–output relationships. Results confirm that the integration of RSM and ANN enables accurate prediction of vehicle behavior with a limited number of simulations, offering a significant improvement over traditional trial-and-error or purely numerical methods. The work contributes to the field of virtual vehicle design by providing a structured and computationally efficient methodology for product architecture optimization, setting the basis for future applications in digital twins (already existing or created ad hoc for specific project cases) and early-stage virtual prototyping.