Abstract:
This study evaluates and compares three metamodeling techniques—the Successive Linear Response Surface Method (SLRSM), Updated Neural Network (NN), and Kriging—for their effectiveness in crashworthiness optimization. The methods are benchmarked against a sequential random search control procedure. Through three distinct crashworthiness examples, including a complex full-vehicle model, the paper demonstrates that all three metamodeling techniques achieve a converged result with comparable efficiency. Notably, the NN and Kriging methods are highlighted for their ability to be updated, allowing them to construct more accurate global approximations and provide a better representation of the design space, particularly near the optimal solution.
