Designing complex engineering systems using computer simulations and physical experiments requires quantification of various sources of uncertainty. Parameter uncertainty and model uncertainty account for the differences between the responses of the computer simulations and physical experiments. The former form of uncertainty is caused by unknown parameters of the computer model while the latter derives from missing physics, numerical approximations, and other inaccuracies of the computer model even if all of the parameters are known. Knowledge of these two sources of uncertainty is obtained by combining data from computer simulations, usually abundant, and physical experiments, usually limited. Accounting for both of these sources of uncertainty by statistical adjustment is referred to as calibration. Identifying these two sources of uncertainty can be difficult and this paper proposes two methods to improve quantifying these sources of uncertainty: (1) multiple response (MR) calibration and (2) design of experiments for the purpose of model calibration. By including many different responses (distinct different responses or one response that varies over time and space) that are mutually dependent on the same parameters, valuable information is obtained about parameter and model uncertainty. Furthermore, obtaining data at different input settings for the experiments can enhance the quantification of uncertainty using a minimum amount of experiments. After quantifying the uncertainty, prediction of the experimental process can potentially be used for the purpose of design under uncertainty. This paper explores the use of multiple responses and design of physical experiments for calibration to enhance identifiability of uncertainty and provides insights into the calibration process.
|Title of host publication||Proceedings of the 9th World Congress on Structural and Multidisciplinary Optimization|
|State||Published - 2011|
|Event||9th World Congress on Structural and Multidisciplinary Optimization - Shizuoka, Japan|
Duration: Jan 1 2011 → …
|Conference||9th World Congress on Structural and Multidisciplinary Optimization|
|Period||1/1/11 → …|