Online Course Catalog
AE 598 - Uncertainty Quantification
|Reinforcement Learning||ORL||70885||ONL||4||-||Huy Trong Tran|
Title: Simulation prediction with quantified uncertainty Advances in computational techniques and resources have made predictive simulations an indispensable tool across engineering and science, with integration of continually more physical models into simulation tools to represent increasingly complex phenomena. This increases the challenge of both validating and quantifying the predictive uncertainty of such simulations. For true predictions there is no corresponding experimental data to check against, so quantification of predictive uncertainty increases their utility and can target pacing sources of uncertainty for reduction. This course will introduce technique for quantifying the uncertainty of simulation predictions. After the predictive science challenge is introduced and motivated with examples, we will: review basic statistical tools and distributions; discuss probability as measures of belief; examine the strengths, limitations, and design of experiments for calibration and validation; introduce quantitative model selection and hypothesis testing for the design and evaluation of physical models; and present methods to propagate known uncertainties through a predictive simulation to the quantity of interest. Inverse adjoint-based sensitivity methods will be discussed for use in validation and uncertainty quantification. Mechanics based examples will be used throughout for motivation.
BS degree in engineering or science from an accredited college in the United States or an approved institution of higher learning abroad and experience in (1) numerical methods (minimally TAM 470, AE 370 or equivalent), (2) mathematics (TAM 541/2 or equivalent), and (3) experience with fluid and/or solid mechanics.