Numerical dispersed flow simulation of fire-flake particle dynamics and its learning representation

Abstract

In this article, we propose methods for simulating the detailed flow of dispersed fire-flake particles in response to the movement of a flame, using chaotic advection and various buoyant flow techniques. Furthermore, we utilize these techniques to gather a synthetic dataset of detailed fire-flake particles and extend the solver to represent the movement of fire-flake particles based on learning-based approaches. Fire-flake particles not only exhibit unique and complex movements on their own, but they are also significantly influenced by the movement of the flame and the surrounding airflow. Modeling the flow of fire-flake particles realistically is challenging due to their chaotic and constantly changing nature. Instead of explicitly modeling the complex fire-flake particles in the flame based on fluid mechanics, this article efficiently approximates the chaotic motion of fire-flake particles using two approaches: 1) chaotic advection to simulate the flow and 2) controlled buoyant flow, which varies based on the temperature and lifespan of the fire-flake particles. Additionally, we collect a fire-flake dataset through this simulation and extends the solver to learn the representation of fire-flake motion using neural networks. During the advection process of fire-flake particles, a new stochastic solver is used to calculate the subgrid interactions between them. In this article, not only we propose algorithms that can express these techniques through numerical simulation, but we also extend this solver using artificial intelligence techniques to enable learning representation. By using the proposed technique, it is possible to efficiently simulate fire-flake particles with various movements in chaotic regions, and it allows for more detailed representation of fire-flake particles compared to existing methods. Unlike the typical random walk approach that adds noise randomly to the movement, our method considers the size and direction of the flame. This allows us to express fire-flake particles stably in most scenes without the need for parameter adjustments.

Keywords

• Fire simulation
• Flame
• Fire-flake particle
• Chaotic advection
• Buoyant flow
• Learning representation