Predictive Control of Extreme Events in Turbulence
Faculty: Prof. Konduri Aditya (CDS) and Prof. Balachandra Suri (BE)
Proposal: Turbulence is widely regarded as one of the greatest unsolved problems in classical physics. In fact, there is not one, but many fundamental questions in turbulence that have eluded clear and tractable solutions till date, both conceptually and practically. One such is the seemingly random occurrence of extreme or anomalous events (and associated statistical signatures) in both transitionally and fully developed turbulent flows. Such extreme events are characterized by intermittent large fluctuations in flow “observables” compared to their standard deviation around a statistically steady mean. Canonical examples of extreme events include sudden dissipation bursts in turbulent shear flows, random and unpredictable regime transitions between meta-stable turbulent states, and formation of rogue waves which affect ships. In addition to these examples from hydrodynamic turbulence, extreme events are also observed in turbulent thermal convection and combustion. Since extreme events can potentially cause significant adverse effects in real-world scenarios (e.g., climate dynamics, fluid structure interactions), recent research has focused on developing frameworks to preemptively and effectively mitigate the occurrence of such events in turbulence and allied problems (e.g., combustion, convection). From a fundamental physics perspective, identifying the dynamical mechanisms driving extreme events in moderately turbulent flows has proven successful in both predicting and mitigating such events. Such approaches, often rooted in the nonlinear dynamical systems framework, have not been tested for fully-developed turbulence and in laboratory experiments. In contrast, purely data-driven and machine learning based approaches have proven effective at identifying precursory signatures foreshadowing extreme events in both transitional and highly-turbulent flows, across diverse flow configurations. Yet, such approaches do not yield physical insights into the origin of extreme events and possible control strategies. Moreover, the vast majority of data-driven and AI/ML based techniques are tested using numerical data where the stringent requirement for real-time prediction and control required/feasible in experiments is overlooked.
The goal of this project is to bridge the dynamical systems and data-driven/ML-based approaches for identifying the origins of extreme events, predicting their imminence, and mitigating their occurrence using real-time closed loop control in turbulent flow experiments. As test cases, we shall study canonical laboratory flow analogs of geophysical circulations where we measure the turbulent flow, detecting signatures of extreme events, and develop control strategies. Additionally, we shall perform direct numerical simulation of turbulent flows and combustion phenomena to develop and test various techniques that aid prediction of extreme events in turbulence.
The PhD student should have a strong background in fluid dynamics, dynamical systems and an aptitude for data science and scientific computing.
Learning Outcomes G Skills from PhD project:
- Strong foundation in fluid dynamics and dynamical systems.
- Rigorous training in high performance parallel computing
- Expertise in turbulence and machine learning
- Expertise in fluid dynamics and turbulence
Relevant Publications:
- Aditya, K., Kolla, H., Kegelmeyer, W.P., Shead, T.M., Ling, J. and Davis IV, W.L., 2019. Anomaly detection in scientific data using joint statistical moments. Journal of Computational Physics, 387, pp.522-538.
- Floris,M., Sai, T.S., Nayak,D., Langella, I., Aditya, K. andDoan, N.A.K., 2024. Data-driven identification of precursors of flashback in a lean hydrogen reheat combustor. Proceedings of the Combustion Institute, 40(1-4), p.105524.
- Suri, B., 2024. Predictive framework for flow reversals and excursions in turbulence. Physical Review Letters, 133(15), p.154002.