CDS&E: Uncovering Self-Competition and Weak Asymmetry in Turbulence through Data-Enabled Diagnostics and Learning

Turbulent motion of fluids appears in many important settings, including weather systems, ocean currents, aircraft flows, wind energy, and industrial mixing. Turbulence looks extremely complicated with swirls of many different sizes. This project addresses a fundamental puzzle: why does violent, intense turbulence often result in little net energy transfer between swirls of different sizes? The reason is that energy does not move one way. Instead, it moves forward and backward between large and small swirls, resembling a tug-of-war where both sides pull hard, but the rope barely moves. The detailed mechanisms of energy transfer are not captured in simplified statistical tools often used to study turbulence. By analyzing large amounts of data from lab experiments and computer simulations with data-enabled learning, this research will uncover the hidden rules that control this balance. The results will help build better models for predicting weather and designing safer engineering systems. The project will also train students in engineering, data science and physics, and provide free software tools to the public. The project will support the use of AI and machine learning in fluid dynamics and will benefit advanced manufacturing of transportation vehicles. Current turbulence theory relies on simplified statistical tools, leaving the mechanisms of energy transfer across scales only partially understood. To address this critical gap, this project develops a data-enabled, physics-guided framework to uncover and model the processes controlling the cascade. The research focuses on two fundamental but unresolved phenomena: "self-competition," where strong physical-space advection paradoxically suppresses scale-to-scale energy transfer, and "weak asymmetry," where intense forward and backward flux events nearly cancel to produce a weak net cascade. The overarching goal is to quantify these mechanisms and translate this physical understanding into improved predictive models for Large Eddy Simulation. To achieve this, the project will analyze large, time-resolved datasets from three-dimensional Direct Numerical Simulations and high-resolution two-dimensional laboratory experiments. The methodology employs advanced data-driven diagnostics to identify the origins of the net energy flux. These insights will then be operationalized in a diagnostic-aware spatio-temporal Graph Neural Network framework to create accurate, interpretable subgrid-scale closures. Results will reveal a mechanistic understanding of “self-competition” and “weak asymmetry” phenomena and provide a principled pathway to regularize turbulence models. The project will produce open software and curated datasets, train graduate and undergraduate students at the interface of physics and data science and deliver transferable methods for other complex multiscale systems in engineering, physics, and beyond. 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.