Automated Characterization of Formation Damage Using Computer Vision and Deep Learning

Abstract

Abstract Formation damage, often caused by drilling, completion, or production activities, can significantly impair reservoir permeability and reduce hydrocarbon recovery. Traditional methods for evaluating formation damage rely on core, logs and production data analysis and manual interpretation, which are labor-intensive and prone to subjectivity. This study presents a novel computer vision-based approach for the automated characterization of formation damage in subsurface rocks using machine learning techniques. This research proposes an image-based classification model that leverages machine learning using Scanning Electron Microscopy (SEM) images, capillary pressure curves, bulk mineralogy, XDR, and petrophysical data to detect and classify formation damage features such as pore blockage, phase trapping, fluid-rock interactions, and wettability alteration. A dataset comprising high-resolution SEM images of core samples from various damaged and undamaged reservoir conditions was curated and preprocessed using computer vision technology (i.e., OpenCV). The performance of four types of machine learning algorithms is compared. The model demonstrated high classification accuracy, effectively distinguishing between degrees and types of formation damage. The results underscore the potential of integrating computer vision with rock characterization workflows to deliver rapid, consistent, and scalable assessments of formation damage potential. The automated analysis approach minimizes the biases introduced during subjective interpretation in traditional methods, reduces the analysis time from days to hours, improves the accuracy, and enables standardized evaluation. A comprehensive formation damage assessment can be performed since the algorithm can process large volumes of SEM images in a short time period. This approach supports improved decision-making in well stimulation, damage mitigation, and production optimization strategies. Future work will focus on expanding the dataset, incorporating petrographic and petrophysical data, and exploring transparent AI techniques to enhance model interpretability in field applications. This study presents a novel method that employs deep learning for the characterization of formation damage. The method is more efficient and comprehensive than traditional techniques. The developed system can support rapid assessment of formation damage from SEM images, facilitating timely operational decisions.