Generating synthetic evolution of turbulent flames with an experimental data-based spatiotemporal diffusion model
Using a conditional diffusion model to generate synthetic evolution of turbulent flames, preserving key features.
Key Findings
Methodology
The study employs a conditional diffusion model combined with a pixel-based spatiotemporal transformer to generate synthetic evolution of turbulent flames. Using an x-prediction flow matching framework, the model generates entire spatiotemporal slabs at inference time, preserving flame features and statistical consistency.
Key Results
- Synthetic flames preserved key features and statistical consistency at large scales, especially in long time-span slabs.
- Successfully generated unseen flame transitions, demonstrating spatiotemporal coherence.
- Controlled transition directions and timescales using a time-varying linear combination of denoising velocities.
Significance
The study provides a new pathway for utilizing experimental data-based generative models in data-sparse environments, complementing both experimental and computational fluid dynamics approaches, addressing the complexity of turbulent flame evolution.
Technical Contribution
Technical contributions include developing a new conditional diffusion model combined with a spatiotemporal transformer, capable of generating complex flame evolution data, offering new engineering possibilities.
Novelty
First to use a conditional diffusion model for generating spatiotemporal evolution of turbulent flames, differing from prior uni-modal methods by offering multi-modal generation capabilities.
Limitations
- Deviations at high temporal frequencies and small spatial scales depend on the time-span of generated slabs.
- Uncertainty in transitions not seen during training.
Future Work
Future work includes extending the model to handle more flame states, optimizing generation quality, and exploring more application scenarios.
AI Executive Summary
This study develops a conditional diffusion model for generating synthetic evolution of turbulent flames. By combining a pixel-based spatiotemporal transformer, the model generates complete spatiotemporal slabs at inference time, preserving flame features and statistical consistency. Experimental results show that synthetic flames preserved key features and statistical consistency at large scales, especially in long time-span slabs. The model also successfully generated unseen flame transitions, demonstrating spatiotemporal coherence. This study provides a new pathway for utilizing experimental data-based generative models in data-sparse environments, complementing both experimental and computational fluid dynamics approaches, addressing the complexity of turbulent flame evolution. Future work includes extending the model to handle more flame states, optimizing generation quality, and exploring more application scenarios.
Deep Analysis
Background
In recent years, machine learning applications in combustion sciences have significantly increased, particularly in modeling complex turbulent combustion processes. Traditionally, machine learning is used in conjunction with high-fidelity simulation data, achieving significant advances.
Core Problem
The evolution of turbulent flames is complex and difficult to predict. Existing methods face bottlenecks in handling multi-modal data and generating unseen flame transitions.
Innovation
Innovations include developing a conditional diffusion model combined with a spatiotemporal transformer, capable of generating complex flame evolution data, and offering multi-modal generation capabilities for the first time.
Methodology
- �� Use conditional diffusion model for flame evolution
- �� Combine pixel-based spatiotemporal transformer
- �� Use x-prediction flow matching framework to generate slabs
Experiments
Experiments use simultaneously measured OH-PLIF and multi-component PIV data to evaluate the model's performance in generating flame evolution, especially in unseen transition states.
Results
Synthetic flames preserved key features and statistical consistency at large scales. The model successfully generated unseen flame transitions, demonstrating spatiotemporal coherence.
Applications
The model can be used for prediction and control of turbulent flames, particularly in combustor design and optimization.
Limitations & Outlook
Deviations at high temporal frequencies and small spatial scales depend on the time-span of generated slabs. Uncertainty in transitions not seen during training.
Plain Language Accessible to non-experts
Imagine a kitchen where the flame is the pot on the stove. Traditional methods are like a chef following a recipe step by step, while the diffusion model is like a smart cookware that automatically adjusts the heat based on the pot's state, ensuring perfect dishes every time.
ELI14 Explained like you're 14
Imagine you're playing a game where the flame is your character. Traditional methods are like manually controlling every move of your character, while the diffusion model is like a smart assistant that automatically plans the best route for you, making it easy to win!
Glossary
Diffusion Model
A generative model that creates data through a denoising process.
Used for generating synthetic evolution of turbulent flames.
Spatiotemporal Transformer
A neural network architecture that processes spatiotemporal data.
Used to generate spatiotemporal slabs.
OH-PLIF
Planar laser-induced fluorescence used to measure OH radicals in flames.
Part of the experimental data.
PIV
Particle image velocimetry used to measure fluid velocity fields.
Part of the experimental data.
x-prediction flow matching framework
A variant of diffusion model that directly predicts target data.
Used for generating flame evolution.
Open Questions Unanswered questions from this research
- 1 How to improve model performance at high frequencies and small scales?
- 2 How to extend the model to handle more flame states?
Applications
Immediate Applications
Combustor Design
Helps engineers optimize combustor performance and reduce emissions.
Long-term Vision
Flame Control Systems
Develop intelligent flame control systems for automated combustion processes.
Abstract
In this study, a conditional diffusion model -- a class of generative machine learning models -- is developed to generate synthetic, experimental data-based trajectories of turbulent flames. Generated experimental data corresponds to simultaneous field measurements, namely OH planar laser-induced fluorescence (OH-PLIF) fields and multi-component particle image velocimetry (PIV) fields, for attached and detached flame states in a swirl combustor configuration. This is done using an x-prediction flow matching framework combined with a pixel-based spatiotemporal transformer, which is capable of generating entire spatiotemporal slabs containing synthetic flame evolution at inference time, conditioned on the flame regime. Using this framework, synthetic flames were found to preserve key flame features and statistical consistency across space and time, particularly at the large scales -- deviations at high temporal frequencies and small spatial length scales were found to depend on the time-span of the generated space-time slabs. An extrapolation task of transition synthesis is also conducted, in which the conditional diffusion model is used to synthesize spatiotemporally coherent flame transitions (flame liftoff and reattachment) unseen by the model during training. This was accomplished using a model for the denoising transition velocity that relies on time-varying linear combinations of attached and detached denoising velocities, leading to an approach that (a) allows for control of the generated transition directions and timescales, and (b) retains sample-to-sample variability in the generated transitions in the process. Overall, this study provides a promising pathway for the utilization of experimental data-based generative models as a new means of data exploration in data-sparse environments, complementing both experiments and computational fluid dynamics-based approaches.