Catastrophic failure of brittle rocks - and porous material in general - is important for a number of practical problems in managing risk associated with system sized material failure. Such failure occurs when structural damage, consisting of fractures and faults, spontaneously self-organizes along localized damage zones. Failure eventually occurs along these zones, often suddenly and with devastating consequences. A great deal about how fractures nucleate, grow and coalesce is known from microstructural and field observations of damaged rocks, and from monitoring and locating acoustic emissions (laboratory earthquakes). However, the precise mechanisms involved in localisation remain elusive. How do cracks, pores and grain boundaries interact locally under stress to cause failure at a specific place, orientation and time? Why can we detect precursors to failure only in some cases?

Heterogeneity is a strong control on failure forecasting; heterogeneous materials exhibit a continuous transition to failure with precursors, while homogeneous materials fail abruptly with no precursors [1]. Here, we take you inside the black box - inside the breaking rocks and the micron-scale processes involved.

X-ray microtomography data from a novel program of triaxial deformation experiments on two samples of Ailsa Craig microgranite, imaged live and in situ at SOLEIL synchrotron, allow us to investigate the influence of heterogeneity on the evolving geometry and spatial distribution of microcracks in deforming rocks [2]. Ailsa Craig microgranite is known to be isotropic and almost entirely crack free. To control for
heterogeneity, we introduced a nanoscale microcrack network into one of these samples by means of thermal stressing, leaving the second sample pristine and nominally crack-free. Specifically, we test the hypothesis that the degree of microstructural heterogeneity in the starting material influences the phase transition order between intact and failed states. Our results show that heterogeneity does indeed affect the order and the predictability of the phase transition, with the second-order transition linked to quasi-static fault propagation in the heterogeneous (heat-treated) sample. In the homogeneous (untreated) sample, the first-order transition is related to the accumulation of increasingly distributed damage and abrupt failure. We also found that catastrophic failure initiates when the correlation length approaches the grain size, while anisotropy in the initial porosity dictates the fault orientation. The correlation length and two-point correlation dimension of the fracture population are key indicators of localization.

Our results show that the degree of starting heterogeneity has a strong effect on the subsequent evolution of the micro-crack network, the nature of the localisation process and the detection (or not) of precursors to failure. These results are important because reliable precursors to field-scale catastrophic failure events are only detected in some cases. Our observations go some way to explaining the reason for this, with significant implications for the predictability of catastrophic failure in different materials.

Acknowledgments: This work is supported by the UK's Natural Environment Research Council (NERC) through the CATFAIL project NE/R001693/1 `Catastrophic failure: what controls precursory localization in rocks?' We acknowledge the beamline PSICHE at SOLEIL for provision of synchrotron radiation facilities. We also thank the University of Edinburgh GeoSciences Workshop for their support in developing the experimental apparatus.

References
[1] J. Vasseur et al. (2015), Heterogeneity: The key to failure forecasting, Nature Sci. Rep. 5, 13259.
[2] A. Cartwright-Taylor et al. (2020), Catastrophic Failure: How and When? Insights From 4-D In Situ X-ray Microtomography, J. Geophys. Res. Solid Earth 125, e2020JB019642.