GNGTS 2021 - Atti del 39° Convegno Nazionale

149 GNGTS 2021 S essione 1.3 USING AE BASED MACHINE LEARNING APPROACHES TO FORECAST RUPTURE DURING ROCK DEFORMATION LABORATORY EXPERIMENTS Sergio Vinciguerra* 1 , Thomas King 1,2 , Philip Benson 3 1 Department of Earth Sciences, University of Turin, Italy 2 Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania, Italy 3 Rock Mechanics Laboratory; School of Earth and Environmental Sciences, University of Portsmouth, United Kingdom The ability to detect precursors of dynamic failure in brittle rocks has key implications for hazard forecasting. Parametric analysis of laboratory Acoustic Emission (AE) data during rock de- formation laboratory experiments has revealed periodic trends and precursory behaviour of the rupture source mechanisms as crack damage enucleates, grows and coalesces into a fault zone, suggesting that strain localization is somehow repeatable and forecastable. However, due to the inherent heterogeneity of rocks and the range of effective pressures, finding a full prediction of rupture mechanisms from AE analysis is still an open goal. Time Delay Neural Networks (TDNN) have shown promise in forecasting failure when using AE derived parameters. However, input pa- rameters can significantly vary over different experimental conditions. The objective of this work is to quantitatively classify phases of deformation/fracturing on physico-mechanical parameters and develop a model that functions over a range of confining pressures. We consider the AE rates and the derived source mechanisms to constraint the stress-strain regime and the scattering and seismic velocity structure to define the evolving medium state as the most important attributes for the neural network model to learn. The former is key in understanding what is the phase of deformation/fracturing (e.g. fracture nucleation, crack coalescence), whilst the latter is impor- tant for characterising the ongoing damage- 4x10 cm samples of Alzo granite, a homogeneous Fig. 1 - a) Stress-strain curves (line) and counts of located acoustic emission (AE, histogram) at increasing confinement. b) Velocity and incremental strain. The red line represents the average. Normalisation is required for more robust training. c) Source mechanisms type ande percentage vs. increasing strain. Compaction and dilatancy dominate phases. d) Scattering ratio and incremental strain. Coalescence of structure is marked by rapid increases in scattering values.