||Edwards Air Force Base, CA 93524
The size and scope of aerospace propulsion CFD simulations and experiments have increased tremendously in the past decade to a point where considerably more data is being generated than can be realistically consumed. There is an urgent need to expand ways in which data obtained from computer modeling and experimental testing is stored, studied, and utilized. Scientific machine learning, and reduced order modeling are showing impressive results towards this end in many complex problems. There is not yet an universal algorithm or methodology by which large scale simulations can be made much more affordable, or physics specific insight can be automatically and effectively extracted from experimental or computational data. Specifically in the area of rocket and gas turbine propulsion, analytical models and experiments attempt to understand turbulent combustion and heat transfer where there are complex nonlinear interactions between many physical phenomena. There is active interest in deploying data science methods based on machine learning and formal model order reduction in the following areas: (a) Improving speed and efficiency of high fidelity simulations, (b) Tools to better handle large scale data sets while extracting physically the most important information from them, (c) Galerkin-projection based reduced order models in unsteady turbulent reacting flow simulations, (d) Statistical machine learning and other data driven approaches to learn system behavior from large reacting flow data sets, and (e) Techniques such as Data Assimilation which synthesize multiple input data streams to improve model predictions. This is a rapidly evolving field at this time, and presents many important opportunities to explore, which can be transitioned to real-world usage relatively quickly.
Swischuk, R., Kramer, B., Huang, C., Willcox, K., “Learning Physics-Based Reduced-Order Models for a Single-Injector Combustion Process,” AIAA Paper 2020-1411 (2020)
Raissi, M., Perdikaris, P., Karniadakis, G.E., “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” J. Comp. Phys., V. 378, pp. 686-707 (2019)
Yu, H., Jaravel, T., Ihme, M., Juniper, M.P., Magri, L., “Data Assimilation and Optimal Calibration in Nonlinear Models of Flame Dynamics,” J. Eng. Gas Turbines Power, V. 141 (12): 121010 (2019)
Combustion; Gas Dynamics; Turbulence; Numerical Methods; Reduced Order Modeling; Machine Learning; Data-Driven Analysis; Experimental Methods