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.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to use limited numerical simulation data to efficiently and accurately construct the nonlinear thermo - acoustic response of the flame in the time domain**. Specifically, traditional numerical simulation methods require a large amount of computing resources to accurately determine the complete nonlinear thermo - acoustic response of the flame to different perturbation frequencies and amplitudes. To overcome this challenge, the author proposes a method based on deep learning, in particular, a dual - path neural network model is introduced to construct the nonlinear response of the flame from limited numerical simulation data. ### Problem Background 1. **Combustion Instability**: - Combustion instability is a common problem in the operation of gas turbines and rocket engines, which may lead to equipment damage, shorten the service life, and even cause catastrophic consequences. - Combustion instability usually exhibits highly nonlinear behavior, so understanding and predicting the nonlinear response of the flame is crucial for controlling combustion instability. 2. **Limitations of Existing Methods**: - **Theoretical Models**: Only applicable to flames with simple geometric shapes. - **Experimental Measurements**: Resource - intensive and difficult to implement in actual engines. - **Numerical Simulations**: Although increasingly popular in recent years, high - fidelity numerical simulations are still very time - consuming and require a large amount of computing resources for each calculation. 3. **System Identification (SI) Methods**: - Polifke et al. proposed a system identification method to derive the flame transfer function (FTF) in the entire frequency range through broadband excitation signals, which significantly improves the efficiency of FTF determination. - However, this method is only applicable to linear flame responses and cannot determine the flame describing function (FDF) which depends on frequency and amplitude. ### Proposed Solutions To solve the above problems, the author proposes the following solutions: 1. **Dataset Construction**: - Use linear swept - frequency signals as the training dataset. These signals contain information of different amplitudes and can provide rich learnable information in a limited dataset. - The training dataset consists of linear swept - frequency velocity perturbations of different amplitudes, covering the frequency range of interest, thereby reducing the high - fidelity numerical simulation cost required to obtain single - frequency signals. 2. **Dual - Path Neural Network Model**: - Introduce a dual - path neural network model (Dual - Path Model), which includes a chronological feature path (CFP) and a temporal detail feature path (TDFP). - CFP is mainly responsible for extracting time features in the time series, using a convolutional neural network (CNN) embedding module and an LSTM layer. - TDFP focuses on extracting temporal detail features, including a temporal detail module and an encoder module. - Feature interaction between the two paths is carried out through an injector module to improve the generalization ability and accuracy of the model. 3. **Short - Sequence Sampling Method**: - Propose a sparse - to - dense short - sequence sampling method to reduce data redundancy, significantly improve the training and inference speed, while maintaining or improving the prediction accuracy. ### Result Verification Through verification, the author proves that the proposed dual - path model can accurately construct the nonlinear response of the flame and shows good generalization ability even under strong nonlinear conditions. In addition, the test results of the short - sequence sampling method on 72 single - frequency signals show that the average relative error (MRE) is reduced by 1.2% compared with the equally - spaced sampling method. ### Summary This paper aims to efficiently and accurately construct the nonlinear thermo - acoustic response of the flame in the time domain through deep learning and the dual - path neural network model, combined with the short - sequence sampling method, and solves the computational resource bottleneck problem faced by traditional numerical simulation methods.

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

Linear and nonlinear flame response prediction of turbulent flames using neural network models

A Robust Autoregressive Long-Term Spatiotemporal Forecasting Framework for Surrogate-Based Turbulent Combustion Modeling Via Deep Learning

Physics-aware neural network flame closure for combustion instability modeling in a single-injector engine

FlamePINN-1D: Physics-informed neural networks to solve forward and inverse problems of 1D laminar flames

DeepFlame: A deep learning empowered open-source platform for reacting flow simulations

Dimensionality Reduction and Dynamical Mode Recognition of Circular Arrays of Flame Oscillators Using Deep Neural Network

Explore artificial neural networks for solving complex hydrocarbon chemistry in turbulent reactive flows

A multi-scale sampling method for accurate and robust deep neural network to predict combustion chemical kinetics

Research on Flow Field Prediction in a Multi-Swirl Combustor Using Artificial Neural Network

A deep learning-based dynamic deformable adaptive framework for locating the root region of the dynamic flames

Predictive Models for Flame Evolution Using Machine Learning: A Priori Assessment in Turbulent Flames Without and with Mean Shear

Network- and CFD/CAA-modelling of the high frequency flame response in multi-jet combustors

Learning Flame Evolution Operator under Hybrid Darrieus Landau and Diffusive Thermal Instability

Exploring Active Subspace for Neural Network Prediction of Oscillating Combustion

Parametric learning of time-advancement operators for unstable flame evolution

Reconstructing the flame in supersonic combustion flow by using the distribution of the wall pressure based on deep learning

Investigation of nonlinear flame response to dual-frequency disturbances

Reconstructing temperature fields from OH distribution and soot volume fraction in turbulent flames using an artificial neural network

Real-time parameter inference in reduced-order flame models with heteroscedastic Bayesian neural network ensembles

Data-Driven Fire Modeling: Learning First Arrival Times and Model Parameters with Neural Networks