LES of Spray Combustion Dynamics and Uncertainty Quantification

by Sagar Kulkarni and Wolfgang Polifke

Motivation

Air transportation is projected to grow significantly in the coming decades. To accomodate, aircraft manufacturers are required to sustain growth while keeping a check on the adverse effect of this growth on human health and environment. In Flight Path 2050, the European Union has put in place stringent norms on fuel consumption, emissions, etc.

Due to the stringent norms on emissions, aero-engine manufacturers have adopted the lean burn combustion concept. One of the critical issues affecting this kind of combustion technology is the occurrence of thermo-acoustic instabilities that may compromise engine life. Therefore prediction of thermo-acoustic behavior of the system during design phase is of primary importance. Compared to gaseous combustion one a small number of studies have been dedicated to spray combustion instability. In the MAGISTER ITN emphasis is laid on predicting the thermoacoustic behaviour of aero engine typical combustion and assess the uncertainty of the prediction due to sensitive input parameters through various Machine Learning (ML) methods (https://www.magister-itn.eu/AboutMagister/). The data generated from simulations will aid the data-driven ML algorithms to predict thermoacoustic behaviour at higher Technology Readiness Level (TRL).

One of the ways to understand the dynamic response of the flame to perturbations is through the so-called Flame Transfer Function (FTF). Determination of FTF of spray flames is more complex due to the effects the acoustic perturbations on spray specific processes. The acoustic field interacts with the fuel spray and leads to periodic variation of spray shape, droplet size distribution in turn leading to variation in evaporative and mixing processes. All these phenomena can be studied computationally to determine the response of the spray to acoustic forcing. Eventually, the FTF of spray flames is estimated from the computational data with System Identification, i.e. a form of supervised machine learning that allows to generate data-based, reduced order models [1, 2].

Because of multi-scale and multi-phase nature of the problem, predictions of the FTF will be highly sensitive to variations in geometric configuration, spray input boundary condition such as droplet diameter, size distribution, etc., and operating conditions. Therefore Uncertainty Quantification analysis is required to quantify the effect of uncertain input parameters on the system output for more reliable thermoacoustic stability analysis.

Objectives and Strategy

Interaction of acoustic with spray specific phenomena: What is the effect of acoustic perturbation on spray combustion sub-processes such as atomization and evaporation. Identify transfer functions for each sub process either analytically or numerically to understand if any time lag is introduced by each of the process and how will it affect the system dynamics.

Stability analysis from LES: In order to do a full stability analysis of a spray combustion system, Large Eddy Simulation (LES) of spray combustion is to be performed, which will generate time series and reproduce the flame dynamics needed to be captured. Broadband excitation will be applied through carefully selected signals (such as to equally excite all the frequencies of interest with low amplitudes while still having high Signal to Noise Ratio (SNR)) to identify the FTF by estimating the finite impulse response of the heat release to the velocity perturbation at the reference location through System Identification tools. The goal is then to compute LES of a spray flame and validate with experiments done in Work Package 5. 

Uncertainty Quantification: Identification of FTF from SI techniques is sensitive to model parameters and boundary condition due to nonlinear nature of the problem at hand. Additionally uncertain knowledge of operating conditions can yield different flame dynamics. Thus uncertainties gets compounded from SI and physical/operational parameters and propagate through network model affecting the prediction of growth rate and flame stability [3]. Hence, one of the main work is to quantify uncertainty of the determined FTF with respect to the length of the time series, number of spray parcels and model parameters used in the identification process. 

References

[1] Tay-Wo-Chong, L., Bomberg, S., Ulhaq, A., Komarek, T., Polifke, W., 2012. Comparative Validation Study on Identification of Premixed Flame Transfer Function. J Eng Gas Turb Power 134, 21502-1–8. doi:10.1115/1.4004183

[2] Merk, M., R. Gaudron, C. Silva, M. Gatti, C Mirat, T Schuller, and W. Polifke. “Prediction of Combustion Noise of an Enclosed Flame by Simultaneous Identification of Noise Source and Flame Dynamics.” Proceedings of the Combustion Institute 37 (2019): 5263–70. https://doi.org/10.1016/j.proci.2018.05.124.

[3] Shuai Guo, Camilo F. Silva, Abdulla Ghani, Wolfgang Polifke., 2018, Quantification and propagation of uncertainties in identification of Flame Impulse Response for Thermoacoustic Stability Analysis. J Eng for Gas Turbines and Power 141, 021032-1. doi: 10.1115/1.4041652

Acknowledgement