Data driven and uncertainty aware physical modeling [electronic resource]
- Nicolas S. Kseib.
- Physical description
- 1 online resource.
- Kseib, Nicolas S.
- Iaccarino, Gianluca, primary advisor.
- Alonso, Juan José, 1968- advisor.
- Ihme, Matthias, advisor.
- Stanford University. Department of Mechanical Engineering.
- As computing capabilities grow and the amount of experimental and numerical data increases, computational strategies can be designed to automatically test and assess different modeling assumptions. We introduce a general data-driven statistical framework that bridges the gap between (numerical or laboratory) experimentation, physical modeling and uncertainty quantification. The framework enables the study of uncertainties and bias in physical models estimated from data. We differentiate between two types of modeling uncertainties and bias, the first one due to physical errors in the models and the second one due to noise introduced by the data-acquisition process. We also present different procedures to build models under different noise assumptions and propose a metric to quantify the quality of the data-driven estimations. The framework is tested in the context of combustion science and chemical kinetics and it is driven by empirical data and simple chemistry models. Why reaction rates? A combination of a rigorous application of the statistical framework as well as recently measured kinetic rates data will allow us to propose new modeling strategies for chemical reaction rates, their associated uncertainties, and how these uncertainties propagate into relevant combustion problems. This thesis also shows that the current state of the art of reporting kinetic uncertainties relevant for predictive problems in combustion sciences is incomplete and only focuses on describing the experimental variability. We propose a technique to report uncertainties in a useful manner for scientists interested in studying the predictive capabilities of their numerical simulations where chemical reaction rates are input parameters. Applications include hydrogen chemistry, explosion limits and initial mixture compositions uncertainties in gaseous mixtures. To represent as closely as possible actual experiments in our models, we will review the process of inferring reaction rates from shock tubes devices. Shock tubes are one of the most popular devices used to measure kinetic rates. We will closely examine the uncertainties of measurements inside a shock tube: 1- due to the presence of non-ideal phenomena in the real device (departures from the ideal operation sequence), 2- incomplete knowledge (unknown parameters needed to model the operation of shock tubes) and 3- sensor uncertainties. This framework can be extended to complex predictive problems relevant to turbulence, turbulent combustion and safety related applications (e.g. nuclear waste treatment, detonations etc.) - and to more complicated reaction rates and larger chemical mechanisms when both raw experimental signals and processed reaction rates become more accessible.
- Publication date
- Submitted to the Department of Mechanical Engineering.
- Thesis (Ph.D.)--Stanford University, 2016.