EGU22-3183
https://doi.org/10.5194/egusphere-egu22-3183
EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Using AE based Machine Learning Approaches to Forecast Rupture during Rock Deformation Laboratory Experiments

Sergio Vinciguerra1, Thomas King1,2, and Philip Benson3
Sergio Vinciguerra et al.
  • 1University of Turin, Department of Earth Sciences, Turin, Italy (sergiocarmelo.vinciguerra@unito.it)
  • 2Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania, Italy
  • 3Rock Mechanics Laboratory; School of Earth and Environmental Sciences, University of Portsmouth, United Kingdom

Parametric analysis of laboratory Acoustic Emission (AE) during rock deformation laboratory experiments has revealed periodic trends and precursory behaviour of the rupture source, as crack damage nucleates, grows and coalesces into a fault zone. Due to the heterogeneity of rocks and the different effective pressures, finding a full prediction of rupture mechanisms is still an open goal. We consider the AE rates and the derived source mechanisms to constrain the stress-strain regime, while scattering and seismic velocity structure define the evolving medium state as the most important attributes for the neural network model to learn. 4x10cm samples of Alzo granite were deformed at confining pressures of 5-40 MPa, whilst AE are recorded. Source mechanisms, as well as AE rates with relation to incremental strain, highlight distinct pre-failure phases. Scattering and seismic velocity measurements indicate the evolving mechanical conditions. A 10MPa simulation test on a model trained with data from 5, 20 and 40 MPa highlights good accuracy when predicting sample failure.

It remains a challenge to generate a ‘generic’ model that can be applied over all experimental conditions. Nonetheless, estimation of parameter importance has highlighted that some physical parameters are better for predicting strain, whilst others are better at stress. This importance can vary in time, suggesting a strong sensitivity of AE properties to the dynamic conditions of the fault zone. Small input changes can strongly affect output, therefore multiple models need to be trained in order to confirm the stability of the forecast. We aim to improve the understanding of the analysis through the search of repeating trends and the identification of consistent variations in key time-varying trends. Seismic scattering shows an early relevance, interpreted as due to the breakup of low frequency surface waves as microcracks begin to coalesce. However the reduction of importance at the later phases of deformation is less obvious. Further investigations are needed to identify at which deformation stage individual parameters are more important and segment time series accordingly.

How to cite: Vinciguerra, S., King, T., and Benson, P.: Using AE based Machine Learning Approaches to Forecast Rupture during Rock Deformation Laboratory Experiments, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3183, https://doi.org/10.5194/egusphere-egu22-3183, 2022.

Displays

Display file