Dataset related to "Laboratory Insights into Fracturing Processes Using Deep Learning Acoustic Emission Catalogues"

The application of machine learning has the potential to revolutionize our understanding of rock damage, ranging from small-scale fractures in geoengineering projects to those associated with earthquake genesis. Laboratory research has been fundamental in elucidating the fracturing processes linked to rock failure and fault preparation. This study explores how machine learning techniques can enhance our perspective on rock damage and failure through the analysis of acoustic emissions (AEs). We developed two independent catalogues from a continuous AE record during a triaxial failure test on an intact sandstone specimen: (1) a standard catalogue utilizing traditional AE data reduction methods and (2) a "deep" catalogue employing the quakephase deep learning phase detection algorithm (Shi et al., 2024). We conducted a comparative analysis of these catalogues using advanced seismological tools applicable across scales from centimetres to kilometres. The deep catalogue revealed a significant enrichment of the fracturing process, identifying twice as many events in areas experiencing compaction and subsequent mobilized shear deformation prior to failure. This augmentation was uniform across magnitudes Mw -9 to -7, resulting in no overall change in b-value; however, trend analysis indicated a small drop before failure in the deep catalogue that was missed in the standard catalogue. Additionally, the spatiotemporal distribution of events aligns with a stationary and homogeneous Poisson field, demonstrating enhancement within the deep catalogue presented here