
Andre Maues Brabo Pereira
Estimating surface relaxivity as a function of pore size via conditional diffusion models
Pore size distribution is a fundamental deliverable of Nuclear Magnetic Resonance (NMR) for reservoir rock characterization. Typically, it is determined in the fast diffusion regime via the surface-to-volume ratio and surface relaxivity (ρ). Although conventional methods rely on a uniform relaxivity assumption, it is highly unlikely that heterogeneous sedimentary rocks exhibit uniform mineralogy across all pore surfaces (Keating and Knight, 2012). To investigate how ρ varies as a function of pore size, we adopt an inverse problem framework (Benavides et al., 2017) to recover the relaxivity function whose corresponding simulated T2 distribution best matches the ground truth T2 distribution. Because standard iterative inversion frameworks require computing forward NMR simulations at each step, they may be computationally prohibitive. To overcome this limitation, we employ generative conditional diffusion models to solve the inverse problem using a data-driven approach. The model takes a rock sample’s microstructure image and its experimental T2 distribution as inputs to predict the underlying relaxivity function. To build a robust training dataset, we adopted a sigmoid prior assumption for the relaxivity functions and utilized an in-house Random Walk (RW) simulator to generate the transversal NMR relaxation signals. Importantly, since the total number of wall collisions of a walker in an RW simulation correlates with the local pore size, we parameterize ρ as a function of the collision count rather than mapping it directly to complex pore coordinates. Tested on synthetic data, the proposed framework successfully reconstructed the non-uniform relaxivity functions, providing a significantly better match to the reference T2 distributions than the uniform relaxivity assumption and demonstrating high potential for accurate porous media characterization.
1) Benavides, F., Leiderman, R., Souza, A., Carneiro, G., & Bagueira, R. (2017). Estimating the surface relaxivity as a function of pore size from NMR T2 distributions and micro-tomographic images. Computers & Geosciences, 106, 200-208. https://doi.org/10.1016/j.cageo. 2017.06.016
2) Keating, K., & Knight, R. (2012). The effect of spatial variation in surface relaxivity on nuclear magnetic resonance relaxation rates. Geophysics, 77(5), E365-E377. https://doi.org/10.1190/ geo2011-0462.1

