Application of graphics processing units to the Lattice Monte Carlo Method for predicting and modelling thermal conduction
AffiliationChemical and Biomolecular Engineering
Document TypePhD thesis
Access StatusThis item is embargoed and will be available on 2020-11-26.
© 2018 Dr Michael Wang
The Lattice Monte Carlo (LMC) method has been successfully applied to a range of scenarios involving pure thermal conduction/diffusion. Extensive investigations by Murch, Belova and Fiedler have exemplified the utility of the so-called quasi-steady-state mode in its ability to predict the effective thermal conductivity of arbitrary composites given knowledge only of the individual phases. The LMC method also operates in a transient mode for time-resolved conduction/diffusion studies. However, a major drawback of the LMC method relative to other methods such as Finite Elements (FE) is the long runtimes required when a large ensemble of particles (typically ∼10^6 ) is run for many iterations (typically >10^5 ) to achieve convergence. This thesis presents, for the first time, parallel processing approaches to both modes of LMC, implemented using CUDA on Graphics Processing Units (GPUs). The benchmarks show that the GPU-accelerated versions of the transient and quasi-steady-state LMC methods, called cudaLMC-trans and cudaLMC-keff, are able to achieve speed-ups of 195× and 28×, respectively, on a single NVIDIA Tesla V100 GPU relative to a modern 10-core CPU. Both cudaLMC programs were validated against analytical and FE results, and directly compared with previously-published LMC models in the literature. In most cases, the achieved accuracy was within 1%. Finally, cudaLMC-keff was applied to provide novel effective thermal conductivity (λ_eff ) predictions of two real composites: copper-infiltrated tungsten (W–Cu) with 10-wt%Cu, and porous zirconia (ZrO_2 ) with 65-70% porosity. In order to accurately capture the internal microstructures of both composites, characterisation of the porous zirconia and W–Cu specimens were respectively carried out using X-ray microcomputed tomography (μCT) and a serial, laser-sectioning technique called the ‘TriBeam’. The simulations were carried out at various temperatures ranging from 25 ◦ C to 1000 ◦ C. This thesis illustrates a complete workflow in the prediction of the effective thermal conductivity of a real material. It advances the state of the art in thermal conductivity modelling by coupling advanced experimental characterisation techniques with a powerful simulation capability provided by a novel, GPU-based LMC algorithm.
Keywordsmonte carlo; thermal; gpu; lattice monte carlo; lmc; cudalmc; nvidia; cuda; conduction; conductivity; random walk; diffusivity
- Click on "Export Reference in RIS Format" and choose "open with... Endnote".
- Click on "Export Reference in RIS Format". Login to Refworks, go to References => Import References