Stellantis

The development of numerical models for crash simulation in the automotive industry, and in particular the meshing phase, still occupies a large proportion of the time devoted to a vehicle development project. What's more, it systematically introduces significant geometric approximations on certain sensitive structural elements of the body-in-white, of complex geometric shape. These approximations are frequently a source of error, making the vehicle's CAE model less predictive, prompting engineers to resort to very fine meshing, generating very heavy numerical models that penalize in terms of memory capacity and computing time.

The fundamental idea behind iso-geometric modeling (IGA) is to bring finite element analysis closer to CAD modeling by doing away with the meshing operation and directly using the geometric model as a support for the calculation. This can be achieved through the development of new types of finite element models using the same shape functions as those used in geometric definition models (B-splines, NURBS or Bézier tiles).

The technical problem encountered in industrial practice concerns the CAD definitions usually generated by designers, which are not suitable for direct calculation using the iso-geometric technique. Typically, in order to keep up with the abrupt changes that can occur in geometric curvatures, the original NURBS and/or Bézier tile surfaces are partitioned and/or merged with other neighboring surfaces, thus generating complex new CAD definitions. Unfortunately, these new geometry changes require adaptation and therefore reworking, which in turn requires development time. To the best of our knowledge, there is currently no software that can perform this task automatically, despite fundamental academic advances and research (several theses exist on the subject, notably from the Stellantis group).

Note, moreover, that in the early stages of a new vehicle project (phase advance), the CAD concept is not fully compatible with mechanical strength requirements, and optimization iterations to modify the CAD definition are very rarely carried out for lack of time. Consequently, surface partitioning, adaptation and readout automation will enable us to make full use of EGO (Efficient Global Optimization) methods based on Gaussian conditioning and evolution strategies, better suited to time-consuming finite element calculations. Non-intrusive POD-type model reduction methods will also be exploited, especially for the linear static equivalent [8].

It should also be noted that to perform this optimization, the expertise of crash engineers is still required to evaluate the complex results often taking into account various antagonistic criteria. However, with the recent advances in artificial intelligence (AI) algorithms, it is becoming appropriate to synthesize all the business knowledge (rules) of the crashworthiness of certain parts of the body-in-white. Thus, in the second phase of the project, an AI computational procedure based on deep learning techniques and artificial neural networks will be developed to automate the feasibility of optimal solutions for certain essential elements of the body-in-white.

The aim of the project is to develop an innovative numerical methodology by coupling the iso-geometric method and artificial intelligence techniques for the shape optimization of body-in-white elements in an industrial context. This objective is broken down into several stages:

• Develop a numerical procedure for automatic decomposition of the CAD definition, enabling phase advance models to be prepared for explicit dynamic and implicit static IGA simulation.
• Develop a coupling procedure between IGA simulation and AI algorithms based on deep learning techniques and artificial neural networks for the optimization of body-in-white structural elements.
• Validation of the numerical methodology through shape optimization applications of body-in-white structural elements, to improve stiffness, vibration and shock (buckling) test performance.