TY - JOUR

T1 - Deformation Precursors to Catastrophic Failure in Rocks

AU - McBeck, J. A.

AU - Aiken, J. M.

AU - Mathiesen, J.

AU - Ben-Zion, Y.

AU - Renard, F.

PY - 2020/12/28

Y1 - 2020/12/28

N2 - Forecasting the timing of catastrophic failure, such as crustal earthquakes, has been a central concern for centuries. Such forecasting requires identifying signals that evolve or accelerate in the precursory phase leading to failure, and the subset of signals that may be detected in the crust. We develop machine learning models to predict the proximity of catastrophic failure in synchrotron X-ray tomography triaxial compression experiments on rocks using characteristics of evolving fracture networks. We then examine the characteristics that most strongly influence the model results, and thus may be considered the best predictors of the proximity of macroscopic failure. The resulting suite of predictive parameters underscores the importance of dilation in the precursory phase leading to catastrophic failure. The results indicate that the evolution of the strain energy density field may provide more robust predictions of the proximity of failure than other existing metrics of rock deformation.Plain Language Summary What controls the timing of large earthquakes? Estimating the conditions conducive to the next large earthquake can help mitigate seismic hazard and save significant societal and economic costs. A prerequisite for such estimates includes determining what measurable and detectable signals change in a systematic manner in rocks approaching catastrophic failure. Machine learning analyses of data acquired by synchrotron X-ray experiments on rocks provide robust means of identifying the evolving fault network characteristics that best predict the proximity of catastrophic failure of the rocks. Translating these fracture network characteristics to geophysical signals may help scientists detect such precursors within crustal fault systems preceding large earthquakes.

AB - Forecasting the timing of catastrophic failure, such as crustal earthquakes, has been a central concern for centuries. Such forecasting requires identifying signals that evolve or accelerate in the precursory phase leading to failure, and the subset of signals that may be detected in the crust. We develop machine learning models to predict the proximity of catastrophic failure in synchrotron X-ray tomography triaxial compression experiments on rocks using characteristics of evolving fracture networks. We then examine the characteristics that most strongly influence the model results, and thus may be considered the best predictors of the proximity of macroscopic failure. The resulting suite of predictive parameters underscores the importance of dilation in the precursory phase leading to catastrophic failure. The results indicate that the evolution of the strain energy density field may provide more robust predictions of the proximity of failure than other existing metrics of rock deformation.Plain Language Summary What controls the timing of large earthquakes? Estimating the conditions conducive to the next large earthquake can help mitigate seismic hazard and save significant societal and economic costs. A prerequisite for such estimates includes determining what measurable and detectable signals change in a systematic manner in rocks approaching catastrophic failure. Machine learning analyses of data acquired by synchrotron X-ray experiments on rocks provide robust means of identifying the evolving fault network characteristics that best predict the proximity of catastrophic failure of the rocks. Translating these fracture network characteristics to geophysical signals may help scientists detect such precursors within crustal fault systems preceding large earthquakes.

KW - STRESS-INTENSITY FACTORS

KW - SEISMIC VELOCITIES

KW - ENERGY BUDGET

KW - MINIMUM-WORK

KW - EARTHQUAKES

KW - PREDICTION

KW - CALIFORNIA

KW - EVOLUTION

KW - GROWTH

KW - WEDGES

U2 - 10.1029/2020GL090255

DO - 10.1029/2020GL090255

M3 - Journal article

VL - 47

JO - Geophysical Research Letters

JF - Geophysical Research Letters

SN - 0094-8276

IS - 24

M1 - e2020GL090255

ER -