Table of Contents
Fetching ...

A Dual-Path neural network model to construct the flame nonlinear thermoacoustic response in the time domain

Jiawei Wu, Teng Wang, Jiaqi Nan, Lijun Yang, Jingxuan Li

TL;DR

The paper tackles the costly task of modeling nonlinear flame thermoacoustic responses by leveraging frequency-sweeping training data and a Dual-Path neural network that separately captures chronological and detailed temporal features. The proposed short-sequence sampling further reduces data and compute requirements while maintaining or improving accuracy. Validation shows robust performance across a range of nonlinearities and test conditions, with notable efficiency gains and improved generalization, even under strong nonlinearity. These advances offer a practical pathway to rapidly predict flame responses in thermoacoustic systems, enabling faster design and control iterations.

Abstract

Traditional numerical simulation methods require substantial computational resources to accurately determine the complete nonlinear thermoacoustic response of flames to various perturbation frequencies and amplitudes. In this paper, we have developed deep learning algorithms that can construct a comprehensive flame nonlinear response from limited numerical simulation data. To achieve this, we propose using a frequency-sweeping data type as the training dataset, which incorporates a rich array of learnable information within a constrained dataset. To enhance the precision in learning flame nonlinear response patterns from the training data, we introduce a Dual-Path neural network. This network consists of a Chronological Feature Path and a Temporal Detail Feature Path. The Dual-Path network is specifically designed to focus intensively on the temporal characteristics of velocity perturbation sequences, yielding more accurate flame response patterns and enhanced generalization capabilities. Validations confirm that our approach can accurately model flame nonlinear responses, even under conditions of significant nonlinearity, and exhibits robust generalization capabilities across various test scenarios.

A Dual-Path neural network model to construct the flame nonlinear thermoacoustic response in the time domain

TL;DR

The paper tackles the costly task of modeling nonlinear flame thermoacoustic responses by leveraging frequency-sweeping training data and a Dual-Path neural network that separately captures chronological and detailed temporal features. The proposed short-sequence sampling further reduces data and compute requirements while maintaining or improving accuracy. Validation shows robust performance across a range of nonlinearities and test conditions, with notable efficiency gains and improved generalization, even under strong nonlinearity. These advances offer a practical pathway to rapidly predict flame responses in thermoacoustic systems, enabling faster design and control iterations.

Abstract

Traditional numerical simulation methods require substantial computational resources to accurately determine the complete nonlinear thermoacoustic response of flames to various perturbation frequencies and amplitudes. In this paper, we have developed deep learning algorithms that can construct a comprehensive flame nonlinear response from limited numerical simulation data. To achieve this, we propose using a frequency-sweeping data type as the training dataset, which incorporates a rich array of learnable information within a constrained dataset. To enhance the precision in learning flame nonlinear response patterns from the training data, we introduce a Dual-Path neural network. This network consists of a Chronological Feature Path and a Temporal Detail Feature Path. The Dual-Path network is specifically designed to focus intensively on the temporal characteristics of velocity perturbation sequences, yielding more accurate flame response patterns and enhanced generalization capabilities. Validations confirm that our approach can accurately model flame nonlinear responses, even under conditions of significant nonlinearity, and exhibits robust generalization capabilities across various test scenarios.
Paper Structure (10 sections, 7 equations, 14 figures, 1 table)

This paper contains 10 sections, 7 equations, 14 figures, 1 table.

Figures (14)

  • Figure 1: The configuration of numerical simulation.
  • Figure 2: The schematic of frequency-sweeping signals with different amplitudes.
  • Figure 3: Overall architecture of the neural network. "Avg-Pooling" means the averaged pooling.
  • Figure 4: (a) Flame dynamic response to the step function signal. (b) Schematic diagram of short sequence sampling method; $n$ and $n_{s}$ are the lengths of the original sequence and the sampling short sequence, respectively.
  • Figure 5: Comparison of prediction results between different neural network models when $A=0.25$. From (a) to (d), the perturbation frequencies are 200 Hz, 400 Hz, 600 Hz, and 800 Hz, respectively. The original input sequence length $n$ is 6000, and the short sequence sampling length $n_{s}$ is set to 1000. $T$ and $A$ denote the period and amplitude of each signal respectively.
  • ...and 9 more figures