Aplicação de simulação numérica e aprendizado profundo utilizando imagens de micro-ct de rochas reservatório

Abstract

Carbonate rocks constitute important reservoirs of mineral resources, however, they exhibit a high degree of anisotropy and heterogeneity, with various types of pores and scales that challenge both the industry and the scientific community. Digital petrophysical studies aim to obtain intrinsic rock parameters through the 3D rendering of X-ray microtomography images. The present study aims to classify pore types and estimate petrophysical parameters from the 3D rendering, segmentation, and numerical simulation of images of carbonate rocks from the Jandaíra Formation, Potiguar Basin, Rio Grande do Norte, Brazil. The images were acquired by X-ray microtomography, with resolution of 35 μm, and represent transverse slices of the rocks. For the treatment of slices, the ImageJ program was used. The segmentation of 3D pores was performed using the multi-Otsu method with the Simpleware ScanIP software. The segmented files were discretized and imported into the Comsol Multiphysics numerical simulator with coarse mesh density. The simulator adopts the finite element method (FEM) to perform its calculations. The simulated physics was configured to be a single-phase fluid in steady-state, laminar flow, with a density of 1000 kg/m³ and a viscosity of 0.001 Pa.s. The differential pressure between the inlet and outlet of this fluid was established at 1 Pa. Comparing with literature and laboratory-obtained values, the digital results were satisfactory and consistent. Segmentation enabled the classification of well-developed stylolites and vugs among the samples. Petrophysical estimates ranged from 0.70% to 7.99% with effective porosity and simulated permeability ranged from 7.79 mD to 3000 mD. Regarding deep learning, the Alexnet model, with SGD criterion and MSE optimizer, exhibited the best performance.