Modeling acoustic emissions in heterogeneous rocks during tensile fracture with the Discrete Element Method

Abstract

A computationally efficient and open sourced methodology designed for the investigation of rock matrix heterogeneities and their effect on pre-and post-fracture Acoustic Emission (AE) distributions is presented. First, an image analysis method is proposed for building a statistical model representing rock heterogeneity. The statistical model is generalized and implemented into a discrete element contact model where it efficiently simulates the presence of defects and locally tough regions. The coupling of the heterogeneity model, discrete element model, and acoustic emission model is demonstrated using a numerical three point bending test. The shape parameter of the statistical model, which controls heterogeneity magnitude, is found to control the spatial width of the acoustic emission distribution generated during failure. The same acoustic emission distribution trend is observed in literature for rocks containing various magnitudes of heterogeneity. Further analysis of the numerical AE activity reveals that larger AE events are located directly along the fracture and they are linearly related to their number of constituent interactions. As such, an AE interaction count threshold is identified to distinguish between fracture and damage AE activity. These results demonstrate the ability of the presented methodology to investigate the location and energy release associated with large fracture events for various levels of heterogeneity.

Prediction of the post‐failure behavior of rocks: Combining artificial intelligence and acoustic emission sensing

“…The recent results in Caulk 23 suggest that, for rock failures under tensile loading, the heterogeneity of the rock properties should be explicitly reflected in the local distribution of bond properties so as to realistically capture the distribution of AE leading to failure. The focus of this study is on compressive loading, where the results of DEM are deemed to be representative of the laboratory measurements qualitatively, if not quantitatively 23,49 …”

“…Given the relatively narrow range of particle sizes, and hence less variation in the contact properties, the energy released during a given strain interval is found to be proportional to the number of broken bonds. The recent results in Caulk 23 suggest that, for rock failures under tensile loading, the heterogeneity of the rock properties should be explicitly reflected in the local distribution of bond properties so as to realistically capture the distribution of AE leading to failure. The focus of this study is on compressive loading, where the results of DEM are deemed to be representative of the laboratory measurements qualitatively, if not quantitatively.…”

“…The focus of this study is on compressive loading, where the results of DEM are deemed to be representative of the laboratory measurements qualitatively, if not quantitatively. 23,49 (A) (B)…”

“…Recent studies have shown how DEM simulations can be used as a powerful tool to study the characteristics of AE events in rocks (and concrete) before, during, and after failure. [23][24][25][26][27] The complex and somewhat stochastic post-failure response and its correlation with the pre-failure characteristics evade the current rock-mechanics frameworks. This is the domain where artificial intelligence (AI) and machine learning (ML), and particularly DL methods 28 can provide powerful tools for capturing temporal correlations in AE and stress-strain data to predict the post-failure responses.…”

“…In particular, it is possible to track the energy release in the form of AE when a cohesive bond breaks. Recent studies have shown how DEM simulations can be used as a powerful tool to study the characteristics of AE events in rocks (and concrete) before, during, and after failure 23–27 …”