| Summary: | 碩士 === 國立臺灣大學 === 機械工程學研究所 === 104 === Effective physical models play important roles in efficient product development cycle. This research focuses on parameter uncertainty to improve precision between model predictions and measured system performances. The state-of-the-art methods use model calibration with Bayesian Inference to identify parameter uncertainties; however potential risks might exist in complex system analysis, namely (1) analyzing multiple parameters, resulting in high computational costs, (2) the predicted confidence levels are low, and (3) unable to infer each individual uncertainty in complex systems. This research adopts main effect analysis from Taguchi''s framework of design of experiments to select important parameters from a complex system. The uncertainty analysis is then narrowed down to those on important parameters. Bayesian updating loop is then reinforced and joint inference of multiple testing functions are used to improve the performance of model calibration. The method is demonstrated in two engineering cases: one is a steady-state test of a simple-supported beam, and the identifying error turns out to be 1.5%; while the other vehicle dynamic test under CarSim® has 17% of identifying error.
|
|---|
