MAE Seminar: Expecting Turbulence: Flow-Informed Machine Learning for Prediction and Control

Abstract: Advances in data-driven and learning-based methodologies have enabled unprecedented autonomy in many engineered systems, but intelligent modeling and control of fluid flows have yet to be realized. Motivated by the enormous potential of intelligent interactions between engineered systems and fluid flows across many applications, my research seeks to bridge the gap between data-driven, computational, and experimental fluid mechanics. In this talk, I will explore three research thrusts demonstrating the potential of flow-informed systems and technologies in the real world.

First, I will present applications of reinforcement learning for aerodynamic control in a gusty, turbulent experimental setting. These results highlight the value of flow measurements for informing model-free control and illustrate the benefits of flow physics domain expertise in designing novel model-based learning controllers. Second, I will discuss data-driven modeling for faster-than-real-time forecasting of experimentally measured turbulent flows. Here, numerical simulations can be integrated to reduce experimental data requirements, enhance model performance through fine-tuning, and improve generalization across geometries. Third, I will introduce a novel framework for real-time reconstruction of wind fields for flow-informed flight path planning in urban environments. The resulting system, trained on a large library of simulated wind fields, is demonstrated via flight-tests in a large-scale wind tunnel facility. Finally, I will discuss how the development and deployment of data-driven modeling may shape the future of fluid mechanics in engineered systems, with applications ranging from drag reduction for transportation to vision-based microclimate modeling.

Bio: Dr. Peter I. Renn is a staff scientist at the California Institute of Technology, where he conducts research at the intersection of artificial intelligence and experimental fluid mechanics. Peter earned his PhD in Aeronautics from Caltech in 2023 under the mentorship of Professor Morteza Gharib, receiving the William F. Ballhaus Prize for an outstanding doctoral dissertation.

Peter’s work focuses on data-driven modeling and intelligent control of fluid flows in real-world engineered systems. Prior to his current role, he served as a quantitative strategist at Virtu Financial, where he developed high-frequency algorithmic trading strategies and furthered his expertise in real-time data-informed decision making. Peter’s work has been supported by organizations such as the National Science Foundation, RTX, Verizon, and the Technology Innovation Institute.