Data-Driven Surrogates of Rotating Detonation Engine Physics with Neural ODEs and High-Speed Camera Footage

TL;DR

This paper develops neural ODE-based surrogate models to replicate complex, multi-scale physics of Rotating Detonation Engines, using high-speed camera data to understand wave interactions and nonlinear behaviors.

Contribution

It introduces a novel approach combining Neural ODEs with latent wave representations to model RDE physics, enabling separation and analysis of multi-scale phenomena.

Findings

Successfully reproduces RDE wave behaviors

Separates and analyzes multi-scale physics

Provides insights into physical processes of RDE

Abstract

Interacting multi-scale physics present in the Rotating Detonation Engine lead to diverse nonlinear dynamical behavior, including combustion wave mode-locking, modulation, and bifurcations. In this work, surrogate models of the RDE physics, including combustion, injection, and mixing, are sought that can reproduce the observed behavior through their interactions. These surrogate models are constructed and trained within the context of Neural ODEs evolving through the latent representation of the waves: the traveling wave coordinate $\xi = x - ct + a$. Shown is that the multi-scale nature of the physics can be successfully separated and analyzed separately, providing valuable insight into the fundamental physical processes of the RDE.