event

Phd Proposal by Yingxing Li

Primary tabs

Proposal Request

Name: Yixing Li

Title: Multi-fidelity Framework for Combustion Dynamics Identification under Supercritical Conditions

Time: Mar 2nd, 12:30pm-2:00pm, 2018

Location: MK 317

Committee:

Dr. Vigor Yang (advisor)

Dr. Tim Lieuwen

Dr. Joseph Oefelein

Dr. Jeff Wu (from ISyE @ GaTech)

 

Abstract:

A primary goal of propulsion engine design is to achieve efficient mixing and combustion, yet the understanding of combustion dynamics at supercritical conditions is still limited and not sufficient to support design optimization. High-fidelity tools, such as Large Eddy Simulations (LES), can generate accurate results, but the high demand for computational resources hinders the broader application of LES. Consequently, lower fidelity approaches like Helmholtz solver are needed to work jointly with high-fidelity simulations to provide information of combustion dynamics in modest amount of CPU hours. In this work, the traditional framework is re-visited, analyzed, and improved with incorporation of machine-learning techniques. The idea of UQ is also introduced with a Bayesian framework.

The first part of this work proposes a novel flame transfer function identification methodology. The traditional Wiener-Hopf is re-examined in a statistical perspective. It is limited by estimation deficiency if ill-conditioned, and fails to capture sparsity. In order to overcome its intrinsic limitations, a two-stage method is proposed. For the first stage, an L1 regularization term is proposed to address the estimation deficiencies. For the second stage, a physics-based criterion which incorporates prior information on dominant frequencies is employed to tune regularization penalty. This two-stage transfer function approach is then applied to study the combustion dynamics of liquid-oxygen/kerosene bi-swirl injectors at supercritical conditions. The application indicates better reflection of underlying physics of the system.

For the second part, based on the traditional LES-Helmholtz joint framework, Bayesian Lasso approach is applied to model the UQ of impulse function, flame transfer function and combustion instability. After the application of prior distributions to impulse function and its variance, the distributions are gradually updated to posterior distributions through incorporation of observed data. Sampling is then drawn from the posterior distributions. The uncertainty of impulse function then propagates forward to determine the UQ of flame transfer function and combustion instability, through propagation in each sample. This framework is then applied to study combustion dynamics of liquid-oxygen/kerosene bi-swirl injectors at supercritical conditions, the framework can predict uncertainty ranges of dominant frequencies and corresponding growth rates. This approach is able to capture dominant responses of injectors. The UQ of combustion instability estimations will be used to guide future injector design.

Status

  • Workflow Status:Published
  • Created By:Tatianna Richardson
  • Created:02/08/2018
  • Modified By:Tatianna Richardson
  • Modified:02/08/2018

Categories

Keywords