Model calibration, which is an important element of Verification &Validation, is the process of estimating unknown inputs in a model by improving the agreement between experimental observations and model predictions. Optimization-based model calibration has been widely used for estimating unknown variables using optimization algorithms. Among several optimization algorithms, gradient based algorithms are often used for efficiency problem. However, if an optimization problem is not formulated as a convex problem, then accuracy issue may arise.
Calibration metrics, such as likelihood function and probability residual, have been used as an objective function of optimization-based model calibration. The objective of a calibration metric is to quantify the difference of similarity between two probability distributions from simulation results and experimental results. However, we have identified that optimization-based model calibration using existing calibration metrics is not a convex optimization problem. This study shows the optimization-based model calibration using existing calibration metrics is not a strict convex optimization problem. Later, a new calibration metric is proposed to formulate a convex optimization-based model calibration.