Cerfacs Enter the world of high performance ...

YouTube Link :https://youtu.be/pJThlZpYuPM

Abstract : 

Thermoacoustic combustion instabilities continue to be a major hurdle in the development of future gas turbine combustion systems. These instabilities are characterized by large-amplitude pressure oscillations in the combustor. They are undesirable as they lead to severe vibrations increasing noise and pollutant emissions, causing excessive thermal and mechanical stresses on combustor components, and even threatening structural integrity. Large Eddy Simulation (LES) has proved to be a powerful tool capable of predicting many unsteady combustor phenomena, including instabilities. However, due to the high computational costs associated with LES, it cannot be a standalone design tool to analyze all possible designs and operating conditions to which instabilities remain extremely sensitive. This is where analytical, reduced- or low-order models (ROM / LOM) tend to be valuable and complement LES well, particularly during the predesign stages of combustor development. While most available LOM tools make some important physical simplifications (e.g., linearization of acoustics, flame response), they also typically use over-simplified geometries. One primary objective is to address the latter limitation and improve existing LOM techniques to be able to handle complex realistic geometries. A major part of the work revolves around developing and validating this new acoustic network modeling tool based on modal expansions (Galerkin Series) and state-space methods (viz. STORM: State-Space Reduced Order Model) for predicting and analyzing instabilities. In STORM, a complex system to be analyzed is decomposed and represented as a network of simpler geometrical elements (subdomains), connection (coupling), flame, and impedance elements. The unique features of STORM are the recently introduced Overcomplete Frame modal expansion technique for modeling acoustics in the network subdomains and the so-called surface spectral connections methodology that was developed. Together they allow seamless interconnections between subdomains with 1D/2D/3D acoustics and construct networks representing complex industry-relevant configurations. The rational approximation methods are discussed for incorporating realistic flame/acoustic interaction models (i.e., Flame transfer functions (FTFs)) in STORM networks. The importance of a few physical constraints, particularly causality, in algorithms deriving these low-order, time- domain, state-space, data-driven flame response models from experimental or high-order simulation data are highlighted. A special type of network impedance element, DECBC (Delayed Entropy Coupled Boundary Condition), is also developed that facilitates predicting mixed entropy-acoustic instabilities. Overall, STORM presents a cost-efficient, modular and flexible tool for predicting thermoacoustic instabilities and should aid in determining stability regimes and optimum passive control strategies. In the second minor part of the thesis, the acoustic forcing of the turbulent swirling spray flame is simulated by employing the Euler-Lagrange (EL) LES approach. The objective was to compute the FTF and assess the suitability of the existing EL-LES two-phase combustion modeling framework for such a system identification problem. Recent work has demonstrated the potential of EL-LES in accurately predicting self-sustained limit-cycle instability. However, forced simulations exhibit some difficulties, as discussed. FTF retrieved numerically deviates from experimental reference values by about 20-30%. Results remain sensitive, in general, to the modeling parameters, and further investigations are required to improve the models and prediction fidelity.

Jury :