Model Update and Statistical Correlation Metrics for Automotive Crash Simulations 2007-01-1744

In order to develop confidence in numerical models which are used for automotive crash simulations, results are compared with test data. Modeling assumptions are made when constructing a simulation model for a complex system, such as a vehicle. Through a thorough understanding of the modeling assumptions an appropriate set of variables can be selected and adjusted in order to improve correlation with test data. Such a process can lead to better modeling practices when constructing a simulation model. Comparisons between the time history of acceleration responses from test and simulations are the most challenging. Computing accelerations correctly is more difficult compared to computing displacements, velocities, or intrusion levels due to the second order differentiation with time. In this paper a methodology for enabling the update of a simulation model for improved correlation is presented. Fast running models are developed for the time histories of the acceleration at the measurement locations based on principal component decomposition and metamodeling. A large number of iterations are required during the model update process in order to guide the changes in the numerical model for improved correlation. The fast running models are utilized during this process instead of the actual solver for computing the time histories of the accelerations. Once the model update is completed, the fast running models are further employed for enabling probabilistic analyses that can reflect the modeling uncertainties in the simulation results. Repeatability of a test is also an issue either due to vehicle-to-vehicle variability, or due to the challenges of instrumenting a vehicle and collecting the test data. Therefore, a statistical correlation metric between the numerical solution and the test data is introduced and the fast running models are employed in the process.