Fast-QSGS: A GPU accelerated program for structure generation of granular disordered media

Abstract

We present Fast-QSGS, a GPU-accelerated program for granular disordered media generation. Based on vectorization, Fast-QSGS is accelerated by modern GPU thanks to the NumPy-compatible API provided by CuPy. We also introduce a variable growth probability function and seed spacing control to improve the speed and accuracy of the original QSGS method. Computational performance benchmarks are conducted on both consumer-grade and professional-grade GPUs. Generation of disordered media of size 4003 can be completed in 30 s on A100 and 110 s on RTX4060, achieving a speedup of over 400 compared with the serial version. Physical benchmarks on the reconstruction of Fontainebleau sandstone and hydrated cement are conducted. Our results demonstrate that the permeability of the reconstructed Fontainebleau sandstone falls within the range of experimental values. Additionally, the average relative error of the volume fraction of the unhydrated cement and capillary porosity of hydrated cement is 1.9 % and 3.4 % compared with Powers’ law, respectively. Program SummaryProgram Title: Fast-QSGSCPC Library link to program files:https://doi.org/10.17632/gswbd7tf3g.1Licensing provisions: MIT licenseProgramming language: PythonNature of Problem: Fast generation/reconstruction of disordered granular media.Solution method: The QSGS method is vectorized and implemented in Python. Via CuPy, the current implementation of QSGS can be easily accelerated by modern GPUs.