UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Multi-GPU accelerated lattice Boltzmann modelling of three-dimensional solid-liquid phase change at high Rayleigh numbers

Liu, X; He, YL; Luo, KH; (2025) Multi-GPU accelerated lattice Boltzmann modelling of three-dimensional solid-liquid phase change at high Rayleigh numbers. Journal of Energy Storage , 131 , Article 117522. 10.1016/j.est.2025.117522.

[thumbnail of Luo 2025 JES GPU accepted.pdf] Text
Luo 2025 JES GPU accepted.pdf - Accepted Version
Access restricted to UCL open access staff until 2 July 2026.

Download (1MB)

Abstract

Latent heat thermal energy storage (LHTES) utilising solid-liquid phase change has received much attention in energy utilisation and conversion. To gain insights into the physics of melting at high Rayleigh numbers (Ra), numerical simulations play a crucial role, besides experimental measurements. The primary challenge, however, is the high computational cost. To address this issue, the multi-GPU accelerated lattice Boltzmann (LB) method for solid-liquid phase change is developed. In this approach, the unified enthalpy-based two-relaxation-time (TRT)-LB formulation is adopted for the temperature field and moving interface identification. An incompressible non-orthogonal multiple-relaxation-time (NMRT)-LB equation for hydrodynamics is also introduced to reduce compressible errors. With this model, the multi-GPU implementation using CUDA and CUDA-aware MPI is carried out. Four CUDA streams with synchronisation are designed to tackle the interaction between different distribution functions on overlapping and inner domains. The communication is well hidden within computation in the proposed concurrency algorithm, leading to the excellent parallel performance in both strong and weak scaling tests. The computational measurements using 4 GPUs reveals that the computation speed achieve 3711.4, 3796.7, and 3807.2 million lattice updates per second (MLUPS) and parallel efficiency η equals 0.973, 0.991, and 0.997 for grid sizes of 512 × 128 × 128, 768 × 192 × 192, and 1024 × 256 × 256, respectively. Based on a careful grid study, three-dimensional (3D) melting processes at Ra = 10<sup>6</sup>, 10<sup>7</sup>, and 10<sup>8</sup> are investigated, and phase change phenomena from laminar to turbulent regimes are analysed. The melting front proceeds in a 2D manner at low Ra. With the increase of Ra, the fully 3D melting behaviour in the turbulent regime has emerged, in which both the duration of conduction-dominated melting and complete melting time are reduced. The morphology of the moving interface is also examined, revealing that an increase in Ra results in a broader interface extent and a more rough and irregular interface shape due to intensified convection. Furthermore, the melting performance at Ra = 10<sup>8</sup> with varied inclination angle θ of enclosure is investigated, indicating that varying the gravitational orientation significantly alters the interface shape and thermal flow patterns.

Type: Article
Title: Multi-GPU accelerated lattice Boltzmann modelling of three-dimensional solid-liquid phase change at high Rayleigh numbers
DOI: 10.1016/j.est.2025.117522
Publisher version: https://doi.org/10.1016/j.est.2025.117522
Language: English
Additional information: This version is the author accepted manuscript. For information on re-use, please refer to the publisher's terms and conditions.
Keywords: Lattice Boltzmann method, Solid-liquid phase change, Multi-GPU implementation, CUDA, High Rayleigh number
UCL classification: UCL
UCL > Provost and Vice Provost Offices > UCL BEAMS
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Mechanical Engineering
URI: https://discovery.ucl.ac.uk/id/eprint/10211870
Downloads since deposit
2Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item