Learning to traverse convective flows at moderate to high Rayleigh numbers
Journal of Fluid Mechanics · 25.07.2026 · 1039
DOI: 10.1017/jfm.2026.11802 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Fluid Mechanics, online erschienen am 15.07.2026. Verfasst von Ao Xu und 3 weiteren Autor:innen. Die Arbeit nennt 70 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
We study the navigation of a self-propelled inertial particle in two-dimensional Rayleigh–Bénard convection at Prandtl number italic probability equals 0.71 Pr = 0.71 $\textit{Pr}=0.71$ and cell aspect ratio upper Gamma equals 4 Γ = 4 $\varGamma =4$ for Rayleigh numbers upper R a R a $Ra$ ranging from 10 Superscript 7 10 7 $10^7$ to 10 Superscript 11 10 11 $10^{11}$ . A reinforcement-learning (RL) controller selects the propulsive acceleration, subject to an upper bound script upper A Subscript italic max A max $\mathcal{A}_{\textit{max}}$ , to achieve a prescribed horizontal displacement. We find that the success rate increases abruptly with script upper A Subscript italic max A max $\mathcal{A}_{\textit{max}}$ at moderate upper R a R a $Ra$ , whereas at higher upper R a R a $Ra$ the transition becomes more gradual and shifts to larger script upper A Subscript italic max A max $\mathcal{A}_{\textit{max}}$ . Moreover, although the completion time increases with upper R a R a $Ra$ , the propulsion energy required for successful traversal decreases. Proper orthogonal decomposition indicates that these performance differences are associated with reorganisation of the carrier flow. At moderate upper R a R a $Ra$ , the dominant large-scale circulation partitions the domain through persistent transport barriers, requiring a finite thrust surplus to cross them; at higher upper R a R a $Ra$ , energy is distributed across many modes, the barriers fragment and transient plume-assisted pathways emerge. Compared with a constant-heading baseline, the learned policy aligns with local currents and consumes significantly less energy. Lagrangian coherent structure analysis further suggests that the RL agent tends to cross repelling barriers and surf along attracting pathways. Finally, by mapping these behaviours onto the local Eulerian flow topology using Voronoi tessellation and the upper Q Q $Q$ -criterion, we distil an interpretable, physics-based heuristic strategy that retains robust navigability. These results connect turbulent-flow organisation with autonomous navigation under bounded actuation.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Xu, A., Wu, H., Xu, B., & Xi, H. (2026). Learning to traverse convective flows at moderate to high Rayleigh numbers. Journal of Fluid Mechanics, 1039. https://doi.org/10.1017/jfm.2026.11802
Vancouver
Xu A, Wu H, Xu B, Xi H. Learning to traverse convective flows at moderate to high Rayleigh numbers. Journal of Fluid Mechanics. 2026;1039. doi:10.1017/jfm.2026.11802
BibTeX
@article{Xu2026,
author = {Xu, Ao and Wu, Hua-Lin and Xu, Ben-Rui and Xi, Heng-Dong},
title = {Learning to traverse convective flows at moderate to high Rayleigh numbers},
journal = {Journal of Fluid Mechanics},
year = {2026},
volume = {1039},
doi = {10.1017/jfm.2026.11802}
}
EUVIMED