GPU-Accelerated NMR T2 Simulator Incorporating Surface Roughness Effect

NMR T2 simulation supplements laboratory experiments that numerically predicts petrophysical properties of digitized rock samples. Recent research demonstrated that the pore shape irregularity acts as surface roughness at the macroscopic point of view and its effect on NMR T2 relaxation is overlooked. So far, neither commercial nor open-source software explicitly suppresses the surface roughness effect on nuclear magnetization decay. This work proposes an innovative GPU-accelerated NMR T2 simulator incorporating the relaxation inhibition factor (RIF) to effectively control the NMR relaxation rate. The proposed GPU-accelerated NMR T2 simulator allocates a thread to control the random movement of a particle; thousands of particles can be moved simultaneously using modern GPU devices. An image-based pore surface roughness characterization technique is applied to parameterize surface roughness into a dimensionless coefficient. The RIF is defined as a function of the roughness coefficient to suppress accelerated surface relaxation and is distributed along solid-pore interfaces. When a walker collides with a solid voxel, it accesses the local RIF that effectively controls the nuclear magnetization decay of individual particles. Numerical results show that the proposed simulator accurately models the NMR T2 relaxation so that the pore structures interpreted from NMR T2 responses agree with the ground truth. This demonstrates that the surface roughness effect has to be removed for pore structure characterization. From the perspective of efficiency, the proposed simulator could achieve two to three orders of magnitude speedup, which exceeds the efficiency of commercial software for high-fidelity simulations. The proposed simulator honors greater physical consistency than existing software by assigning surface relaxivity and RIF to solid voxels rather than treating them as particle-carrying parameters. It is the first NMR T2 simulator that explicitly suppresses the surface roughness effect in numerical simulations, which provides an integrated platform for accurate and effective characterization of pore structures from digital rocks.