Multi-scale porosity characterization of rocks via data mining and geomechanical applications

Accurately characterizing the geomechanical properties of subsurface reservoirs is critical for assessing wellbore stability, designing hydraulic fracture treatments, and predicting surface subsidence. While direct laboratory measurements on core plugs remain the definitive standard, the limited availability of physical core material necessitates the development of alternative predictive methodologies, such as microstructural image analysis. This research presents an optimized workflow for estimating porosity and dynamic elastic properties utilizing high-resolution Backscattered Electron Micrographs (BSEM).
Initial investigations utilizing moderate-magnification BSEM images yielded significant statistical scatter (R²=0.555), with measurement errors ranging from -600% to 100% in heterogeneous datasets. These inaccuracies were systematically attributed to macroscopic sample heterogeneity and resolution-dependent bias, where fine-scale pore details are unresolved during digital acquisition. To mitigate these effects, a standardized imaging protocol was established, identifying a scan speed of level 4 as the optimal balance between signal-to-noise ratio and acquisition efficiency for high-volume automated scanning.
The core methodology employs fractal geometry to address resolution limits. By applying a power-law regression to iterative 50% downscaled resolution levels, porosity values were extrapolated to a theoretical infinite magnification. This approach is theoretically grounded in the scale-independent nature of sandstone pore networks established by Thompson, Katz, and Krohn (1987). Furthermore, the study demonstrates that satisfying the Representative Elementary Volume (REV) through large-scale montages (~1 cm²) is essential for mitigating spatial variability and windowing bias.
The synthesis of optimized BSEM data with the PETGAS ultrasonic dataset provides a critical geomechanical bridge, allowing for the calculation of dynamic Young’s modulus and Poisson’s ratio. Ultimately, this thesis establishes a robust framework for integrating micro-scale image data and data mining workflows to enhance the predictive accuracy of subsurface geomechanical models.