Adaptive Sampling Strategies for Crashworthniess Applications

Abstract

In the context of surrogate metamodeling for crashworthiness applications, the implementation of adaptive sampling strategies holds great potential for overcoming the challenge of setting the optimal number of samples a priori. These adaptive strategies offer a significant advantage by avoiding the common pitfalls of underand oversampling, making them attractive for expensive-to-evaluate functions such as those commonly encountered in crashworthiness applications. Despite their potential, most current research in this area relies predominantly on static sampling strategies. Recognizing this gap, our work explores the adaptation of innovative adaptive sampling methods specifically tailored to the needs of crashworthiness applications. In this context, we describe the Multi-Query Cross-Validation Voronoi (MQCVVor) method. This approach extends the traditional CVVor technique by integrating parallel processing, thus improving the efficiency and accuracy of surrogate models, especially for small scale multi-response systems. Our method demonstrates a significant improvement over conventional static Latin Hypercube Design (LHD) in terms of convergence speed and robustness. In addition to these results, we briefly discuss the potential limitations of adaptive sampling strategies and lay the groundwork for future research aimed at refining these techniques for more complex scenarios.