AbstractAcoustic Emission (AE) refer to the release of energy that occurs due to inelastic deformation of media at the laboratory scale. Analogous to seismic data, they provide a crucial window into the analyses of energy propagation at a scale that is relatively easy to handle. In this thesis, the fracturing processes in the lead up to dynamic failure of rock samples are analysed in terms of the microfracturing source and the effects of a developing fault zone on the propagation of energy.
Two families of rock samples were selected: (1) granites, represented by Alzo Granite and Westerly Granite, and (2) sandstones, represented by Darley Dale Sandstone. The former was selected as a relatively flaw free environment in which to study the enucleation of fractures with minimal biases. Meanwhile the latter, was selected to study these processes in an environment which already had a pre-existing network of damage/porosity. In both cases, the rock types were selected for their generally homogeneous properties, further minimising any influence from bedding or foliation on deformation structure. Samples were deformed under conventional triaxial deformation conditions until dynamic failure under a range of confining pressures. During experimentation, AE were detected by an array of Piezo-Electric Transducers (PZT), recording fracturing events as discrete variations in voltage. It is from this data that the following analyses are derived.A Distributed Time Delay Neural Network is trained under semi-supervised conditions to recognise the onset of a signal in Acoustic Emission (AE) data obtained during the laboratory deformation experiments. Time series of instantaneous frequency, permutation entropy and seismic envelope are separated into simple classifications of noise and signal. The model is trained in sequential batches, allowing for an automated process that steadily improves as new data are added. To validate the approach, real AE data from a triaxial deformation experiment of Darley Dale Sandstone (fully drained conditions and a confining pressure of 20 MPa) are used to train a model of 300 waveforms that is subsequently applied to pick the onsets of the remaining data. When compared with a simple amplitude-threshold picking methodology, results demonstrate significant improvement in the number and quality of event source locations that may be used in later analyses.Source mechanism were solved using a least squares minimisation of the 3D first-motion polarity focal sphere to characterise AE as tensile (T-type), shearing (S-type) and compaction/collapsing (C-type). Samples of Alzo Granite and Darley Dale Sandstone were systematically deformed until dynamic failure at confining pressures of 5, 10, 20 and 40 MPa. Periods of fracture enucleation and growth, crack coalescence, and dynamic failure are identified from relative percentages of fracture mechanisms. Spatio-temporal trends further reveal a dependency of fault zone formation on the competition between tensile and compaction type mechanisms, with a surprisingly small amount of shearing. Finally, the occurrence of a family of low amplitude tensile events prior to sample failure point towards predictable and deterministic behaviour in the development of fault zone highlighting the potential for the forecasting of fracture coalescence.
The delay in the maximum amplitude arrival of seismic energy (peak delay) is an important attribute to map complex geology, fluid reservoirs, and faulting in the lithosphere. The parameter was measured and mapped in the frequency range 50 to 800 KHz using Acoustic Emission data recorded during triaxial deformation experiments of Westerly Granite and Darley Dale Sandstone. The highest peak delay consistently appears when energy propagates perpendicular to an acoustic impedance surface such as the deformation-induced shear zone. Measurements confirm the dominance of forward scattering and anisotropy processes, with results that are strongly influenced by the distribution of time-dependent heterogeneity and stiffness.
|Date of Award||2020|
|Supervisor||Sergio C. Vinciguerra (Supervisor), Philip Benson (Supervisor) & Luca De Siena (Supervisor)|