High-Pressure LPRE Combustion Dynamics: Low-Cost Computation and Stochastic Analysis

Abstract

The work has focused on improvements for modeling, descriptions, and understanding for liquid-propellant-rocket-engines (LPRE) with regard to turbulent combustion dynamics and instability, chaotic description, and improved sub-grid closure for reduced-order models. For the combustion dynamics, the use of the flamelet model for closure continues and has been extended to our three-dimensional code, saving computational resources compared to chemical kinetic modeling and better representing the sub-grid multi-physics by includingsub-grid mixing with the chemistry. Deep learning neural networks (NN) have been used to represent the flamelet closure model and replace the look-up table, which is more costly in memory requirements and less accurate in interpolation. Successful modeling of the Continuously Variable Resonant Combustor (CVRC) combustion chamber for both initial transient behavior and the asymptotic dynamic equilibrium with near-constant amplitudes has been achieved. Optimization studies determine the number of deep NN layers, number of neurons per layer, and the use of separate NNs modeling each of the desired output variables. The training of the NN has been optimized by focusing on the most important parameter domains. Three-dimensional combustion instability has been examined in several multi injector engine configurations with 10-, 19-, 30-, and 82-injectors. Both one-step kinetics and the flamelet model have been used and compared. Some agreement with the Rocketdyne experimental results by Jensen et al. [16] is found. Both premixed and non-premixed flame structures are predicted through the combustor. Three-dimensional results show the determination of the dominant mode of oscillation is very sensitive to parameter choice. Accordingly, chaotic behavior can result in self-initiated change of the dominant instability mode occurring in an alternating fashion. The Flamelet Progress Variable subgrid model has been applied to our multi-injector computations.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 11, 2022
Accession Number
AD1230311

Entities

People

  • William A. Sirignano

Organizations

  • University of California, Irvine

Tags

Fields of Study

  • Physics

Readers

  • Combustion science or combustion engineering.
  • Computational Fluid Dynamics (CFD)
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks