CAREER: Efficient Uncertainty Quantification in Turbulent Combustion Simulations: Theory, Algorithms, and Computations

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

Practical systems for energy conversion (such as internal combustion engines and gas turbines) and hazardous phenomena (such as urban and wildland fires) are governed by turbulent combustion. Building turbulent and chemical models that make accurate predictions with known uncertainty are crucial to enable the design of cleaner and better engines as well as the prediction of disastrous fire risks to meet the urgent need for global sustainability. One of the biggest challenges in quantifying the uncertainty for turbulent combustion simulations is the computational cost associated with complex chemistry, where tens of thousands of parameters are involved. Thus, it is currently impossible to estimate the uncertainty for each parameter and to predict the effects on turbulent combustion simulations. This project aims to establish a theoretical foundation and develop a computational framework to address these challenges. The development of the open-source packages in the project will engage a broad community and provide educational opportunities to undergraduate students in the science, technology, engineering, and mathematics (STEM) fields. The current project aims at addressing two technical questions: first, how to quantify the kinetic uncertainty in turbulent combustion simulations to evaluate the soundness and guide the development of turbulent combustion models; and second, how to leverage experiments not only to validate but also inform the development of chemical models. The proposed work will validate the hypothesis of universal kinetic sensitivity direction in laminar and turbulent flames to build the theoretical foundation for reducing the dimensionality of the uncertain kinetic parameter space via the active subspace method. Furthermore, a computational framework and open-source packages will be developed, combining physics-informed active subspace estimation, auto-differentiable programming, and GPU acceleration, to enable the forward and inverse uncertainty quantification in large-scale combustion simulations with affordable computational cost. The framework will be adopted to demonstrate the feasibility to 1) constrain kinetic uncertainties based on canonical combustion measurements, and 2) quantify the effect of kinetic uncertainties on laminar and turbulent flame responses. The success of the project will provide fundamental insights on the impact of flow-chemistry interaction on the propagation of kinetic uncertainty and enable efficient and tractable uncertainty quantification in practical combustion systems and phenomena with real fuels. More broadly, the framework developed in the project will provide essential tools and insights for future integration of physical science and uncertainty quantification techniques in energy applications. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.