This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada
In the last years, the automotive engineering industry has been deeply influenced by the use of « machine learning » techniques. However, some specific engineering aspects like optimization study still require the development of suitable high-performance learning approaches involving FE simulation result data.
The reduction of weight on a car body is a crucial matter, which better the environmental impact and the cost of the product. The actual optimization process at Renault SA uses numerical Design of Experiments (DOE) to find the right thicknesses and materials for each part of the vehicle, that guarantees a reduced weight and still a good behavior of the car body, measured by sensors on the body. DOE methodology uses between 3 and 10 times the numbers of parameters of the study (which means, for a 30 parameters-study, at least 90 simulations).
During the last 2 years, Renault’s teams strived to develop a disruptive methodology to conduct optimization study using massive extraction of data from the FE crash solver, and high-performance computing (HPC). For instance, in the last study presented by S. Assou et al. , it took 10 runs to find a solution of a 34-parameters problem.
In order to improve this method, we must extract more knowledge from the simulation results (correlations, spatio-temporal features, explanatory variables) and process them to find efficient ways to describe the car crash dynamics.
One of the improvements made in the last months is the use of EIM (Empirical Interpolation Method ) to identify the few time instants and spatial nodes of the FE-mesh that “explain” the behavior of the body during the crash, within a dimensionality reduction approach. The EIM method replaces a former K-Means algorithm, which was processed online, for every ROM. Instead, the computation of EIM method is made offline, once for all, for each simulation. This new method allows us to compute a ROM quite faster, and to reduce the number of features that we use for regression (about 100 approximatively). The nonlinear regression step is achieved by a Random Forest (RF) algorithm, as presented in .
Another improvement of the method is the computation of numerical features describing the shape of the body, at a nodal scale. The characteristics of orientation of the elements surrounding a node must be taken into account to describe the behavior of the node during the crash. The actual method integrates some numerical features, computed from the orientation of the elements around each node, to explain the node behavior.