Sequential optimization and uncertainty propagation method for efficient optimization-based model calibration

The goal of model calibration is to improve the predictive capability of a computational model by estimating the unknown input variables of the model. Optimization-based model calibration (OBMC) is a probabilistic way to estimate the unknown input variables through the use of optimization techniques. Performing optimization in a probabilistic sense requires a high computational cost to obtain statistics about the outputs at every iteration of the optimization. To improve optimization efficiency, this paper proposes a sequential optimization-based model calibration approach that makes use of first an efficient, and then a highly accurate probabilistic assessment method, in sequence. At the earlier stage of the sequential optimizations, approximate integration methods are used to accelerate the probabilistic assessment process. As a calibration metric, the moment matching metric is devised to use the obtained statistics of the outputs. When the optimization reaches near-convergence, a more accurate method, such as a sampling method with an accurate surrogate model, is substituted for the probabilistic assessment. Thus, this paper provides an efficient and accurate procedure for optimization-based model calibration. Two engineering applications, model calibration of a shallow strip footing model and an automotive steering wheel-column model, are presented to demonstrate the effectiveness of the proposed method.