Control of Extreme Events in Turbulent Flows with Scientific Machine Learning

Projektbeschreibung

Extreme Naturereignisse von Schurkenwellen bis zu Flammenrückschlägen vorhersagen

Der globale Wandel hin zur Dekarbonisierung bringt neue Herausforderungen mit sich, darunter auch extreme Ereignisse in der Strömungsdynamik: seltene, plötzliche Störungen mit potenziell drastischen Änderungen der Strömungsentwicklung. Ob atmosphärische Blockierung, ozeanische Schurkenwelle oder Flammenrückschläge in Wasserstoffverbrennungsanlagen – die treibende Kraft hinter diesen Ereignissen sind komplexe, nichtlineare Wechselwirkungen, die sich einer Standardvorhersage entziehen. Das ERC-finanzierte Projekt CONTEXT zielt darauf ab, dieses Problem durch die Verbindung von Deep Learning mit physikalischen Einschränkungen zu lösen. Es soll ein Rahmen zur Vorhersage und Unterdrückung dieser Ereignisse entwickelt werden. Die Methodik wird in verschiedenartigen Systemen erprobt. Im Fokus steht vor allem die Lösung solcher extremen Ereignisse bei der sauberen Wasserstoffverbrennung.

Ziel

Climate change and the race to decarbonise our society is making extreme events in fluids more prevalent. These are rare events where the flow suddenly takes extreme states far from its normal state. These can be found in any flow systems, such as in the atmosphere with atmospheric blocking causing extreme heatwaves, or in our oceans with rogue waves (waves of extreme heights) capable of capsizing boats, or in engineering flows in hydrogen-based clean combustors with flashback events where the flame suddenly moves back into the injection system.
Currently, we cannot accurately predict such extreme events due to several roadblocks. First, the chaotic nature of these turbulent flows makes them hard to predict: any infinitesimal perturbation leads to drastically different evolutions (the butterfly effect). Second, extreme events originate from complex nonlinear interactions which are very different for systems with different physical mechanisms. This makes any past development difficult to generalize across different flow systems. Third, we have very limited observations of such events.
To revolutionize how we tackle extreme events, the CONTEXT project will create a cutting-edge scientific machine learning framework that blends deep learning with physics-based techniques. CONTEXT’s framework will provide the means to (i) identify precursors and mechanisms of extreme events, (ii) forecast the flow evolution before and during extreme events and (iii) control the flows to prevent extreme events. CONTEXT’s framework will be able to handle diverse and disparate physics, with this being demonstrated across different flows of increasing complexity and with different physics, culminating in a demonstration of the practical impact of the framework on the engineering-relevant multiphysics test case of a flashbacking hydrogen combustor.
CONTEXT will provide a comprehensive framework to achieve the understanding, prediction, and prevention of extreme events in turbulent flows.