Enabling the real-world application of machine-learning augmented linear reacting mean field analysis to facilitate the conversion of green hydrogen to electrical energy

• Applicant: Dr.-Ing. Thomas Ludwig Kaiser
• Subject Area: Fluid Mechanics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 530442286

Hydrogen is widely seen as a promising strategy for managing the energy transition. It makes it possible to store surplus energy from wind and solar power plants in chemical form. When needed, the stored energy can be converted into electrical energy in fuel cells or by combustion in gas turbines. The latter alternative is particularly interesting for large-scale applications because it is highly scalable and at the same time enables high power densities. It could therefore also be used in aircraft engines. However, as with combustion systems using conventional fuels, combustion noise (CN) and thermoacoustic instabilities (TAI) during operation still pose significant problems, right up to the loss of the engine, and must be avoided at all costs. However, addressing these effects through incremental improvements in long iterative cycles, as has been the case for conventional fuel gas turbines in recent decades, is not a solution given the urgency of climate change. The central goal of the IGNITION project is therefore to develop an inverse design strategy to account for CN and TAI and find the most efficient way to stabilize the engine. This strategy is based on the linearized mean field analysis (LMFA) approach. Given a flow instability in a laminar flow that causes oscillations, LMFA combined with adjoint optimization can find the most efficient change in system parameters to stabilize the flow. This way, for example, the geometry of a flow device can be optimized to mitigate the instability. However, this approach is not applicable to turbulent flows, let alone turbulent reacting flows such as those found in gas turbine combustors. This is because there are currently no reliable models that can describe turbulent mixing and combustion in the context of LMFA. In this context, the IGNITION project aims at developing such models. To this end, experimental measurements will be performed for a generic turbulent hydrogen-air jet flame. The same configuration will be studied with large-eddy simulations (LES). State-of-the-art data assimilation tools are employed to study the results of experiments and LES to improve the understanding of the interaction of turbulence with instabilities/coherent structures. From recent studies, the research community has learned that LMFA is closely related to the Reynolds-Averaged Navier-Stokes (RANS) process. Therefore, the RANS turbulence and flame models are adapted to LMFA. Since they are known to give erroneous results, they are extended for the given configuration using machine learning in the form of physically informed neural networks (PINNs). In this way, the combination of LMFA, adjoint optimization and PINNs leads to an inverse design strategy applicable to turbulent flames. It is used to find the most efficient ways, be it geometry optimization or heating/cooling of parts, to stabilize TAIs and mitigate CN. In the final phase of the project, the approach will be applied to an industrial configuration.

• DFG Programme: Independent Junior Research Groups