3rd ICAI 2024

International Conference on Automotive Industry 2024

Mladá Boleslav, Czech Republic

2.2 Surrogate model To describe the relationship between the input parameters mentioned above and the output parameters, which were the maximum shoulder force and the maximum upper rib compression, the HDMR approach published in (Kubíček et al, 2015) was used. It is based on the decomposition of the individual input parameters influence in the vicinity of the nominal solution. For the output values in the vicinity of the nominal solution F 0 ( x ), we can write: where x is an input parameters vector and f i , f i, j etc. are incremental functions depending on the parameters x i or its combinations. The incremental functions are zero at the point of nominal solution, and in our case the nominal solution represents the results for initial state of the side airbag. In relation (1), the first sum on the right hand side represents the effect only from the individual parameters. The second sum is the influence from the interaction of 2 parameters, etc. For our study, an interaction of at most 3 parameters was considered, and we can write the relation (1) in the form: 3. Problem Solution and results The solution procedure is schematically illustrated in Fig. 5. The basis was an FEM computational model of a side-pole crash with the default setting of the side airbag, i.e. so-called a nominal solution. At the same time, the ranges of input parameters were defined, in which the optimal solution was sought. From the nominal solution, the first set of variants was generated. The results of this set were stored in a database of FEM results, from where they were requested by an AI tool to create the first generation of the HDMR surrogate model. After performing accuracy tests, the AI tool generated another set of input parameters for the new FEM variants and ran further calculations. This cycle was repeated until sufficient accuracy was achieved. The sampling strategy and accuracy tests were based on the procedures described in (Kubíček et al, 2015). Depending on the required degree of accuracy, the resulting number of variants needed to build the surrogate model ranged from 200 to 300. The incremental functions were interpolated using Lagrange polynomials.

97