CASUS Institute Seminar, Dr. Damar Wicaksono, CASUS
Computer simulations are relied upon for robust and safe engineering system design. Besides the obvious use for making predictions, more specific applications of computer simulations range from parametric study to sensitivity analysis, from model calibration to design optimization.
Such computer simulations typically feature dozens of input parameters (model parameters, environmental and operating conditions) whose exact values are often not known; they may thus be considered uncertain and modeled probabilistically. Uncertainty quantification (UQ) in engineering thus deals with the same types of analyses while considering the uncertainty of the inputs.
As computer simulations become increasingly expensive to run due to increasing model complexity, brute force UQ methods such as Monte Carlo techniques are usually not feasible when applied directly. Therefore, central to the application of UQ methods is metamodeling, in which an approximation of the simulation model is first constructed from a limited number of runs and then used as a faster-to-run proxy for the simulation model.
In this presentation, Damar introduces the practices of uncertainty quantification for computer simulations in engineering under a common framework. Then, he presents several examples – taken from simple benchmark problems as well as nuclear engineering simulation – to illustrate some typical analyses, including metamodeling, sensitivity analysis, calibration, and reliability analysis.