SEAT: How to optimize the design of a car-body-structure by using Machine Learning скачать в хорошем качестве

SEAT: How to optimize the design of a car-body-structure by using Machine Learning

As presented by Fabiola Cavaliere from SEAT S.A. at the 9th BEFORE REALITY Conference. Abstract: The future of the automotive industry is electric. Guaranteeing the perfect match between sustainability and performance is our challenge. A feature which plays a crucial role in this challenge is the weight of the car. Lighter structures mean higher battery autonomy, as well as lower production costs for the company. At the same time, weight reduction can lead to the deterioration of the car functional properties. Standard vehicle development projects are still driven by trial-and-error methods. Based on experience, car body designers propose tentative configurations of the structure, which are then tested by the simulation team. Inevitably, multiple configurations need to be tested before a satisfactory design is reached. Moreover, due to time constraints, only few configurations can be analyzed, which limits our understanding of the problem and ignores potentially better solutions. Artificial Intelligence (AI) and Machine Learning (ML) techniques offer new ways to push our boundaries. They are based on the idea that we do not need complex and time-consuming models to identify patterns in the behavior of a structure. The method developed in the context of this work uses this principle to maximize the stiffness and comfort behavior of a car while minimizing its weight. Based on a machine learning approach known as Proper Generalized Decomposition method (PGD), the tool self-learns how to approximate the solution of a complex problem depending on a set of design parameters (material/geometry properties of car components). It consists of three main phases. First, it parametrizes the model. Next, with only one computation, it automatically computes a parametric solution which contains the results for every possible combination of predefined design variables. This parametric solution is then used to perform fast optimization analysis. All results are uploaded to an interactive app, where users, both technical and non-technical, can explore in real-time how changes in the design variables affect the car performance and make decision accordingly, thus drastically reducing the repetitive iterations in the development process and improving the quality of the final solution. #9thBEFOREREALITY ••••••••••••••••••••••••••••••••••••••••••••••••••• ▷ Our software products: https://www.beta-cae.com/products.htm ▷ Get in touch with us: https://www.beta-cae.com/contacts.htm ▷ Follow us: FACEBOOK ►   / betacae   LINKEDIN ►   / beta-cae-systems   TWITTER ►   / betacae   INSTAGRAM ►   / beta_cae_systems