Next-gen gas turbine combustors

Due to increasing concern about global climate change, next-generation gas turbine combustors are being developed to meet stringent norms. Cleaner, quieter combustors operating on flexible fuels (not restricted to hydrocarbon fuels) are the immediate requirement. Such combustors encounter undesirable high-amplitude flow oscillations, popularly known as combustion instability. They not only produce excessive noise but also cause premature or catastrophic engine damage. Despite decades of research, a clear prescription for avoiding or mitigating oscillations has yet to be established. The difficulty lies in the complex interaction among combustion, acoustics, and turbulence. Experiments or modeling alone cannot provide a solution.

We therefore combine experiments, theory, numerical simulations, and, recently, machine learning methods on academic combustors that mimic a few features of the actual engines. We understand the mechanism of combustion instability in this process and provide/demonstrate prescriptions to identify and avoid it at the design stage. In particular, we work on the following problems.

Physics-informed machine learning

The challenge in mitigating combustion instability is partly due to the fact that theoretical and experimental investigations are performed without a synergy component between them. Since the flow is complex, it is not possible to theoretically represent all its features. At the same time, experiments cannot be performed with all the geometrical details and power ratings reproduced, as they become prohibitively expensive. 

As an alternative, a framework can be developed that combines the physics-based theoretical models with inputs from lab-scale experiments to generate hybrid models. These models are more likely to represent the actual system, thereby making the predictions of instability more accurate. More importantly, these models can be made modular for individual components, such as flame holders and liners, thereby building an accurate database. It can also discover hidden variables that are either challenging or impossible to measure experimentally. Such a machine learning method is successfully employed to develop a quantitatively accurate model and reconstruct the hidden variables in an academic combustor. We are currently applying the technique for practical combustors. 

Mohan, Balasundaram, and Sathesh Mariappan. "Self-excited intermittent thermoacoustic fluctuations in an annular combustor exhibiting flame transient phenomena: Physical mechanisms and modeling." Physics of Fluids 35.11 (2023).