Combining acoustic emission and unsupervised machine learning to investigate microscopic fracturing in tight reservoir rock

Abstract

We use the acoustic emission (AE) and unsupervised machine learning to investigate the influence of bedding structures on the tight rock fracturing at the microscale, aiming to uncover the macro failure mechanisms relevant to oil and gas production engineering. We compared the AE characteristics of typical tight rocks, specifically tight sandstone and shale, under uniaxial loading both perpendicular and parallel to the bedding structure. Additionally, we applied unsupervised machine learning to cluster AE waveforms to analyze microscopic fracturing. The clustering results, constrained using the elbow method and silhouette score, revealed that a consistent number of three clusters was suitable for categorizing all samples. We then used the classification results, together with other AE parameters, to interpret the fractures influenced by bedding structures. Our results revealed that AE waveforms could be classified into three clusters, corresponding to microscopic fracturing, including tensile, shear, and mixed cracking types. Cracks formed under low-stress conditions tend to exhibit tensile failure modes, transitioning into shear fracturing before reaching peak compressive stress. Tight sandstones exhibited higher strength in their bedding structures compared to shale, possibly due to differences in pre-existing microcrack structure characteristics. This study advances our knowledge of tight reservoir rock failure mechanisms and provides valuable guidance for tight reservoir development engineering.