A reduced model using random forest: application on car crash optimization
-
S. Assou [1] ; Y. Tourbier [2] ; E. Gstalter [3] ; M. Charrier [1] ; O. Dessombz [1] ; L. Jézéquel [1]
-
[1] Ecole Centrale de Lyon
Ecole Centrale de Lyon
Arrondissement de Lyon, Francia
-
[2] Goupe Renault
-
[3] France UTC
-
- Localización: SeMA Journal: Boletín de la Sociedad Española de Matemática Aplicada, ISSN-e 2254-3902, ISSN 2254-3902, Vol. 78, Nº. Extra 2, 2021, págs. 193-212
- Idioma: inglés
- DOI: 10.1007/s40324-019-00208-8
- Texto completo no disponible (Saber más ...)
-
Resumen
The development of the metal structure of a car is carried out throughout the vehicle project using numerical simulation. The simulation of the crash, which is representative of the most serious accidents, is the most expensive computation of the design process. Each simulation takes about 10 h on a high performance computer. The engineers must regularly carry out optimization studies using their crash model. The optimization method used is based on a Design Of Experiments and requires from 3 to 10 crash computations per parameter that are varied in the study. Studies with dozens of parameters, often cost hundreds of computations, that gives a study delay of several weeks. The statistical models model some scalar variables, which are objectives and constraints of the optimization study. This paper focuses on the use of a reduced crash model to perform an optimization study representative of the standard studies, replacing the statistical models created from the DOE method. The proposed method is based on Regression-CUR which combines features selection and a multiple linear regression model, here replaced by a Random Forest model. The aim is to model all the fields simulated by the crash calculation. The replacement of the linear regression by Random Forest has several advantages such as non-linearity, low memory footprint and efficient computing.
