Data-Driven Modeling for Complex Fluid Physics
A mini-symposium (MS72) on "Model Reduction for Chemically Reacting Flows: Challenges, Advances, and Benchmarks" will be held on Wednesday, 13 July, 2022 (4:00 - 6:00 pm Eastern Time), at the SIAM Annual Meeting (AN 2022). There will be four invited presentations given by Prof. Cheng Huang from the Univ. of Kansas, Dr. Elizabeth Qian from CalTech, Prof. Hessam Babaee from Univ. of Pittsburg, and Dr. Youngsoo Choi from Lawrence Livermore National Lab.
It is well-recognized that the popularity of data-driven modeling techniques is growing in the field of engineering. By “data-driven” we imply methods which build surrogate predictive models of physical systems (a) solely from data computed and learned from a limited number of high and/or low fidelity simulations, or (b) using some form of intrusive code modifications to the original solvers, based on low dimensional representations (such as projection-based reduced order models). We are interested in methods which enable an affordable deployment of high fidelity solvers which tend to be otherwise extremely expensive. Very encouraging applications are being demonstrated in applying such methods to linear or mildly nonlinear problems. In the realm of fluid mechanics, while there have been some successes in low speed and highly viscous flows, porting these methods to unsteady (turbulent) compressible flows introduces many new challenges. There is, however, a tremendous need for such methods in aerospace and related fields, where routine usage of high fidelity solvers is greatly desired, but limited by their cost. It remains challenging to use the current state-of-art data-driven modeling techniques (model reduction, machine learning etc.) for problems with complex fluid physics (multi-scale and multi-physics problems) to meet the requirements of many-query applications in industry (engineering design, optimization and UQ).
Towards this goal, we are planning a series of activities beginning with an invited panel discussion at the AIAA Scitech Forum on 01/20/2021.
This workshop will focus on data-driven models to address unresolved challenges resulting from fluid compressibility in aerospace engineering, such as: 1. Presence of dissimilar multi-scale physics; 2. Convection-dominated nonlinear dynamics; 3. Dispersed steep gradients and highly nonlinear/stiff kinetics in combustion and hypersonic flow and 4. Non-stationary chaotic features in flow physics. The following is a list of objectives for this workshop:
Bring together experts in both the field of computation science and engineering and motivate more research activities to address challenges in data-driven modeling of complex fluid physics
Assess the state of progress in data-driven techniques for complex fluid physics modeling – multi-scale, multi-physics engineering problems requiring substantial computing resources (turbulence, combustion and hypersonic flow etc.)
Identify a hierarchy of challenge problems for evaluations of different data-driven techniques using appropriate metrics for accuracy, efficiency, cost and robustness
Guide the development of improved data-driven techniques to provide fast and accurate models for many-query applications in industry (engineering design, optimization and UQ) – determine individual limitations of current approaches
Establish pathway to accelerate the transition of fundamental research in data-driven techniques to real industry applications. Primary focus will be aerospace applications such as: Unsteady turbulent reacting flows in rocket and gas turbine combustion chambers, hypersonic flows (inlets, combustion chambers)
Ramakanth Munipalli Air Force Research Laboratory
Cheng Huang University of Kansas
Benjamin Peherstorfer New York University
Karen Willcox University of Texas at Austin
Karthik Duraisamy University of Michigan
Charbel Farhat Stanford University
Jan Hesthaven EPFL
Matthias Ihme Stanford University
Irina Tezaur Sandia National Lab
Venkateswaran Sankaran Air Force Research Laboratory
Fariba Fahroo Air Force Office of Scientific Research
Please send an email to email@example.com for additional information on the workshop series or to be included on the email distribution list.
Air Force Center of Excellence on Multi-Fidelity Modeling of Rocket Combustion Dynamics: https://afcoe.engin.umich.edu/
Scientific Machine Learning for Combustion: https://kiwi.oden.utexas.edu/research/combustion-model-reduction.php
Pressio (open-source project aimed at enabling leading-edge projection-based reduced-order model): https://pressio.github.io/