Laboratory Earthquakes Research Articles

Physics informed neural network can retrieve rate and state friction parameters from acoustic monitoring of laboratory stick-slip experiments

Various machine learning (ML) and deep learning (DL) techniques have been recently applied to the forecasting of laboratory earthquakes from friction experiments. The magnitude and timing of shear failures in stick-slip cycles are predicted using features extracted from the recorded ultrasonic or acoustic emission (AE) signals. In addition, the Rate and State Friction (RSF) constitutive laws are extensively used to model the frictional behavior of faults. In this work, we use data from shear experiments coupled with passive acoustic (variance, kurtosis, and AE rate) interleaved with active source ultrasonic monitoring (transmitted wave amplitude) to develop physics-informed neural network (PINN) models incorporating the RSF law and AE rate generation equation with wave amplitude serving as a proxy for friction state variable. This PINN framework allows learning RSF parameters from stick-slip experiments rather than measuring them through a series of velocity step experiments. We observe that when the stick-slip cycles are irregular, the PINN models outperform the data-driven DL models. Transfer learning (TL) PINN models are also developed by pre-training on data collected at one normal stress level followed by forecasting shear failures and retrieving RSF parameters at other stress levels (i.e., with different recurrence intervals) after retraining on a limited amount of new data. Our findings suggest that TL models perform better compared to standalone models. Both standalone and TL PINN-estimated RSF parameters and their ground truth values show excellent agreements thus demonstrating that RSF parameters can be retrieved from laboratory stick-slip experiments using the corresponding acoustic data and that the transmitted wave amplitude provides a good representation of the evolving frictional state during stick-slips.

Off-fault deformation feedback and strain localization precursor during laboratory earthquakes

Recent large-scale seismological observations have shown that off-fault strain localization and foreshock migration could serve as an early warning of an impending earthquake. However, this process is still largely unknown. In this study, state-of-the-art friction experiments were conducted in a oil-confined biaxial shear apparatus to investigate the link between stick-slip nucleation and off-fault deformation. Our findings indicate that there is a direct link between stick-slip nucleation and off-fault deformation, provided that the fault is conditionally unstable (a − b < 0). Inelastic off-fault deformation may trigger unstable slip by decreasing the stiffness of the surrounding rock volume, which favors earthquake nucleation. Additionally, the study presents laboratory observation of precursory strain localization around a fault during stick-slip cycles. These findings suggest that volumetric deformation processes could be a main factor in the nucleation of large ruptures and strain localization could be a reliable harbinger of large earthquakes.

Machine Learning Predicts the Slip Duration and Friction Drop of Laboratory Earthquakes in Sheared Granular Fault

Machine Learning Predicts the Slip Duration and Friction Drop of Laboratory Earthquakes in Sheared Granular Fault

Development of a loading and diagnostic system to study laboratory earthquakes

Development of a loading and diagnostic system to study laboratory earthquakes

Experimental study on the rupture behaviors of orthogonal faults: effects of stress state and rupture initiation location

Summary Fault rupture dynamics is expected to be significantly affected by the geometry of fault system, especially for orthogonal faults. However, the rupture behaviors of orthogonal faults especially the coseismic interactions are far from fully understood. Here, we present experimental results from a series of laboratory earthquakes to elucidate the effect of the stress state and initiation location on the rupture behaviors of orthogonal faults. Our results reveal a phase diagram of rupture behaviors of orthogonal faults, which is collectively controlled by stress state and rupture initiation location. For events initiating from the main fault, the rupture cannot jump to the branch, which may be due to the clamping effect or the inhibited shear stress accumulation on the branch. On the contrary, events initiating from the branch can persistently trigger ruptures of the main fault. This difference highlights the directional effect associated with the rupture of orthogonal faults. Further, the rupture length of triggered ruptures on the main fault is controlled by the stress state of the fault system. With the increase of the ratio between the shear stress and normal stress, the rupture length of the main fault increases. Our results reproduce the rupture behaviors of orthogonal faults, which may provide insights into the rupture characteristics of natural earthquakes.

Laboratory Earthquakes Simulations—Emergence, Structure, and Evolution of Fault Heterogeneity

AbstractSeismic faults are known to exhibit a high level of spatial and temporal complexity, and the causes and consequences of this complexity have been the topic of numerous research works in the past decade. In this paper, we investigate the origins and the structure of this complexity by considering a numerical model of laboratory earthquake experiment, where we introduce a fault with homogeneous mechanical properties but allow it to evolve spontaneously to its natural level of complexity. This is achieved by coupling the elastic deformability of the off‐fault medium (and therefore allowing for heterogeneous stress fields to develop) and the discrete degradation and gouge formation at the fault plane (and therefore allowing for structural heterogeneity to develop). Numerical results show the development of persistent stress, damage, and gouge thickness heterogeneities, with a much larger variability in space than in time. Strong positive correlations are found between these quantities, which suggest a positive feedback between local normal stress and damage rate, only mildly mitigated by the mobility of the granular gouge in the interface. For a wide range of confining stresses, after a sufficient number of seismic cycles, the fault reaches a state of established disorder with a constant roughness, a certain amount of periodicity at the millimetric scale, and a power law decay of the Power Spectral Density at smaller spatial scales. The typical height‐to‐wavelength ratio of geometrical asperities and the correlations between stress and damage profiles are in good agreement with previous field or lab estimates.

Machine Learning Predicts Earthquakes in the Continuum Model of a Rate‐And‐State Fault With Frictional Heterogeneities

AbstractMachine learning (ML) has been used to study the predictability of laboratory earthquakes. However, the question remains whether or not this approach can be applied in a tectonic setting where one may have to rely on sparse earthquake catalogs, and where important timescales vary by orders of magnitude. Here, we apply ML to a synthetic seismicity catalog, generated by continuum models of a rate‐and‐state fault with frictional heterogeneities, which contains foreshocks, mainshocks, and aftershocks that nucleate in a similar manner. We develop a network representation of the seismicity catalog to calculate input features and find that the trained ML model can predict the time‐to‐mainshock with great accuracy, from the scale of decades to minutes. Our results offer clues as to why ML can predict laboratory earthquakes and how the developed approach could be applied to more complex problems where multiple timescales are at play.

Laboratory Earthquake Ruptures Contained by Velocity Strengthening Fault Patches

AbstractMany natural faults are believed to consist of velocity weakening (VW) patches surrounded by velocity strengthening (VS) sections. Numerical studies routinely employ this framework to study earthquake sequences including repeating earthquakes. In this laboratory study, we made a VW asperity, of length L, from a bare Poly(methyl methacrylate) PMMA frictional interface and coated the surrounding interface with Teflon to make VS fault sections. Behavior of this isolated asperity was studied as a function of L (ranging from 100 to 400 mm) and the critical nucleation length, , which is inversely proportional to the applied normal stress (2–16 MPa). Consistent with recent numerical simulations, we observed aseismic slip for &lt; 2, periodic slip for 2 &lt; &lt; 6, and non‐periodic slip for 10 &lt; . Furthermore, we compared the experiments where L was contained by VS material to standard stick‐slip events where L was bounded by free surfaces (i.e., L = the total sample length). The free surface case produced ∼10 times larger slip during stick‐slip events compared to the contained fault ruptures, even with identical . This disparity highlights how standard, complete‐rupture stick‐slip events differ from contained events expected in nature, due to both the free surface conditions and the heterogeneous normal stress along the fault near the free ends, as confirmed by Digital Image Correlation analysis. This study not only introduces the Teflon coating experimental technique for containing laboratory earthquake ruptures, but also highlights the utility of as a predictive parameter for earthquake behavior.

Predicting Stick-Slips in Sheared Granular Fault Using Machine Learning Optimized Dense Fault Dynamics Data

Predicting earthquakes through reasonable methods can significantly reduce the damage caused by secondary disasters such as tsunamis. Recently, machine learning (ML) approaches have been employed to predict laboratory earthquakes using stick-slip dynamics data obtained from sheared granular fault experiments. Here, we adopt the combined finite-discrete element method (FDEM) to simulate a two-dimensional sheared granular fault system, from which abundant fault dynamics data (i.e., displacement and velocity) during stick-slip cycles are collected at 2203 “sensor” points densely placed along and inside the gouge. We use the simulated data to train LightGBM (Light Gradient Boosting Machine) models and predict the gouge-plate friction coefficient (an indicator of stick-slips and the friction state of the fault). To optimize the data, we build the importance ranking of input features and select those with top feature importance for prediction. We then use the optimized data and their statistics for training and finally reach a LightGBM model with an acceptable prediction accuracy (R2 = 0.94). The SHAP (SHapley Additive exPlanations) values of input features are also calculated to quantify their contributions to the prediction. We show that when sufficient fault dynamics data are available, LightGBM, together with the SHAP value approach, is capable of accurately predicting the friction state of laboratory faults and can also help pinpoint the most critical input features for laboratory earthquake prediction. This work may shed light on natural earthquake prediction and open new possibilities to explore useful earthquake precursors using artificial intelligence.

On the b-value dependency of injection-induced seismicity on geomechanical parameters

Variability in the b-value, which describes the frequency distribution of earthquake magnitudes, is usually attributed to variations in differential stress in the setting of natural and laboratory earthquakes. However, differential stress is unlikely to explain b-value variations on the reservoir scale of injection-induced seismicity cases where significant differential stress variability can hardly be expected. We investigate the responsible geomechanical parameters for the b-value reduction observed in injection-induced seismicity at the Basel EGS field in Switzerland, for which a measured in-situ stress model and fault orientations of numerous microseismic events are available. We estimate the shear and normal stresses along faults, differential stress, pore pressure increase at failure, and the Coulomb failure stress, for each event. Event magnitude and shear stress display the most systematic and clear correlation between each other, while other parameters do not show a clear correlation with event magnitude. We further examine the relationship between the b-value and these geomechanical parameters. We discover that the b-value systematically decreases with increasing shear stress. Again, other geomechanical parameters do not show a clear correlation with the b-value. We conclude that b-value variability is explained by variations in shear stress in the injection-induced seismicity setting, where near constant differential stress conditions are expected. Furthermore, we observe that b-value reduction with time also correlates with an increasing number of events along faults having high shear stress, which strongly supports our conclusions. Thus, we discovered a profound physical mechanism behind b-value variation in injection-induced seismicity beyond general understanding of b-value variation.

Nucleation Mechanism and Rupture Dynamics of Laboratory Earthquakes at Different Loading Rates

The loading rate of tectonic stress is not constant during long-term geotectonic activity and significantly affects the earthquake nucleation and fault rupture process. However, the mechanism underlying the loading rate effect is still unclear. In this study, we conducted a series of experiments to explore the effect of the loading rate on earthquake nucleation and stick–slip characteristics. Through lab experiments, faults were biaxially loaded at varying rates to produce a series of earthquakes (stick–slip events). Both shear strain and fault displacement were monitored during these events. The findings indicate a substantial effect of the loading rate on the recurrence interval and the shear stress drop of these stick–slip events, with the recurrence interval inversely proportional to the loading rate. The peak friction of the fault also decreases with the increasing loading rate. Notably, prior to the dynamic rupture of earthquakes, there exists a stable nucleation phase where slip occurs in a quasi-static manner. The critical nucleation length, or the distance required before the dynamic rupture, diminishes with both the loading rate and normal stress. A theoretical model is introduced to rationalize these observations. However, the rupture velocity of these lab-simulated earthquakes showed no significant correlation with the loading rate. Overall, this study enhanced our comprehension of earthquake nucleation and rupture dynamics in diverse tectonic settings.

Laboratory Earthquake Rupture Interactions With a High Normal Stress Bump

AbstractTo better understand how normal stress heterogeneity affects earthquake rupture, we conducted laboratory experiments on a 760 mm poly (methyl‐mathacrylate) PMMA sample with a 25 mm “bump” of locally higher normal stress (∆σbt). We systematically varied the sample‐average normal stress () and bump prominence (). For bumps with lower prominence () the rupture simply propagated through the bump and produced regular sequences of periodic stick‐slip events. Bumps with higher prominence () produced complex rupture sequences with variable timing and ruptures sizes, and this complexity persisted for multiple stick‐slip supercycles. During some events, the bump remained locked and acted as a barrier that completely stopped rupture. In other events, a dynamic rupture front terminated at the locked bump, but rupture reinitiated on the other side of the bump after a brief pause of 0.3–1 ms. Only when stress on the bump was near critical did the bump slip and unload built up strain energy in one large event. Thus, a sufficiently prominent bump acted as a barrier (energy sink) when it was far from critically stressed and as an asperity (energy source) when it was near critically stressed. Similar to an earthquake gate, the bump never acted as a permanent barrier. In the experiments, we resolve the above rupture interactions with a bump as separate rupture phases; however, when observed through the lens of seismology, it may appear as one continuous rupture that speeds up and slows down. The complicated rupture‐bump interactions also produced enhanced high frequency seismic waves recorded with piezoelectric sensors.

Explainable machine learning for labquake prediction using catalog-driven features

Recently, Machine learning (ML) has been widely utilized for laboratory earthquake (labquake) prediction using various types of data. This study pioneers in time to failure (TTF) prediction based on ML using acoustic emission (AE) records from three laboratory stick-slip experiments performed on Westerly granite samples with naturally fractured rough faults, more similar to the heterogeneous fault structures in the nature. 47 catalog-driven seismo-mechanical and statistical features are extracted introducing some new features based on focal mechanism. A regression voting ensemble of Long-Short Term Memory (LSTM) networks predicts TTF with a coefficient of determination (R2) of 70% on the test dataset. Feature importance analysis revealed that AE rate, correlation integral, event proximity, and focal mechanism-based features are the most important features for TTF prediction. Results reveal that the network uses all information among the features for prediction, including general trends in high correlated features as well as fine details about local variations and fault evolution involved in low correlated features. Therefore, some highly correlated and physically meaningful features may be considered less important for TTF prediction due to their correlation with other important features. Our study provides a ground for applying catalog-driven to constrain TTF of complex heterogeneous rough faults, which is capable to be developed for real application.

Fabric controls fault stability in serpentinite gouges

SUMMARY Serpentinites are polymineralic rocks distributed almost ubiquitously across the globe in active tectonic regions. Magnetite-rich serpentinites are found in the low-strain domains of serpentinite shear zones, which act as potential sites of nucleation of unstable slip. To assess the potential of earthquake nucleation in these materials, we investigate the link between mechanical properties and fabric of these rocks through a suite of laboratory shear experiments. Our experiments were done at room temperature and cover a range of normal stress and slip velocity from 25 to 100 MPa and 0.3 to 300 µm s−1, respectively. We show that magnetite-rich serpentinites are ideal materials since they display strong sensitivity to the loading rate and are susceptible to nucleation of unstable slip, especially at low forcing slip velocities. We also aim at the integration of mechanical and microstructural results to describe the underlying mechanisms that produce the macroscopic behaviour. We show that mineralogical composition and mineral structure dictates the coexistence of two deformation mechanisms leading to stable and unstable slip. The weakness of phyllosilicates allows for creep during the interseismic phase of the laboratory seismic cycle while favouring the restoration of a load-bearing granular framework, responsible of the nucleation of unstable events. During dynamic slip, fault zone shear fabric determines the mode of slip, producing either asymmetric or Gaussian slip time functions for either fast or slow events. We report rate/state friction parameters and integrate our mechanical data with microstructural observations to shed light on the mechanisms dictating the complexity of laboratory earthquakes. We show that mineralogical and fabric heterogeneities control fault slip behaviour.

Foreshock properties illuminate nucleation processes of slow and fast laboratory earthquakes

Understanding the connection between seismic activity and the earthquake nucleation process is a fundamental goal in earthquake seismology with important implications for earthquake early warning systems and forecasting. We use high-resolution acoustic emission (AE) waveform measurements from laboratory stick-slip experiments that span a spectrum of slow to fast slip rates to probe spatiotemporal properties of laboratory foreshocks and nucleation processes. We measure waveform similarity and pairwise differential travel-times (DTT) between AEs throughout the seismic cycle. AEs broadcasted prior to slow labquakes have small DTT and high waveform similarity relative to fast labquakes. We show that during slow stick-slip, the fault never fully locks, and waveform similarity and pairwise differential travel times do not evolve throughout the seismic cycle. In contrast, fast laboratory earthquakes are preceded by a rapid increase in waveform similarity late in the seismic cycle and a reduction in differential travel times, indicating that AEs begin to coalesce as the fault slip velocity increases leading up to failure. These observations point to key differences in the nucleation process of slow and fast labquakes and suggest that the spatiotemporal evolution of laboratory foreshocks is linked to fault slip velocity.

Using a physics-informed neural network and fault zone acoustic monitoring to predict lab earthquakes

Predicting failure in solids has broad applications including earthquake prediction which remains an unattainable goal. However, recent machine learning work shows that laboratory earthquakes can be predicted using micro-failure events and temporal evolution of fault zone elastic properties. Remarkably, these results come from purely data-driven models trained with large datasets. Such data are equivalent to centuries of fault motion rendering application to tectonic faulting unclear. In addition, the underlying physics of such predictions is poorly understood. Here, we address scalability using a novel Physics-Informed Neural Network (PINN). Our model encodes fault physics in the deep learning loss function using time-lapse ultrasonic data. PINN models outperform data-driven models and significantly improve transfer learning for small training datasets and conditions outside those used in training. Our work suggests that PINN offers a promising path for machine learning-based failure prediction and, ultimately for improving our understanding of earthquake physics and prediction.

Laboratory earthquakes decipher control and stability of rupture speeds

Earthquakes are destructive natural hazards with damage capacity dictated by rupture speeds. Traditional dynamic rupture models predict that earthquake ruptures gradually accelerate to the Rayleigh wave speed with some of them further jumping to stable supershear speeds above the Eshelby speed (~sqrt{2} times S wave speed). However, the 2018 Mw 7.5 Palu earthquake, among several others, significantly challenges such a viewpoint. Here we generate spontaneous shear ruptures on laboratory faults to confirm that ruptures can indeed attain steady subRayleigh or supershear propagation speeds immediately following nucleation. A self-similar analysis of dynamic rupture confirms our observation, leading to a simple model where the rupture speed is uniquely dependent on a driving load. Our results reproduce and explain a number of enigmatic field observations on earthquake speeds, including the existence of stable subEshelby supershear ruptures, early onset of supershear ruptures, and the correlation between the rupture speed and the driving load.

The Effect of Shear Displacement and Wear on Fault Stability: Laboratory Constraints

AbstractFaults are composed of a damage zone and a fault core with wear material that has accumulated during sliding. Fault zone width and structural complexity have been suggested as key factors that dictate the stability of fault slip and the transition from stable to unstable motion. Here we address the role of fault rock wear rate and the presence of gouge on the stability of frictional sliding. We performed biaxial shear experiments on quartz‐rich bare rock surfaces having different resistances to abrasion (i.e., wear production with slip), as well as simulated quartz gouge. To characterize the influence of shear displacement on fault stability, constant velocity and velocity step experiments were performed to displacements up to 30 mm. Mechanical data are analyzed within the rate‐and‐state framework, and integrated with post‐mortem microscopic analyses of the samples. For initially bare surfaces a shear displacement threshold is required to transition from stable to unstable sliding. Stick‐slip events (laboratory earthquakes) evolve systematically as a function of fault slip. Shear displacement has a clear effect on the rate‐and‐state parameters (a‐b) and Dc. For all our tests, (a‐b) decreases with increasing shear displacement, implying enhanced velocity weakening and potential unstable behavior for larger fault slip. For high wear rates and simulated gouge, Dc decreases with increasing slip. However, for low wear rate faults, changes in Dc are negligible. Our results demonstrate that, fault stability varies systematically with fault slip and in particular that shear displacement and strain localization are dominant parameters controlling fault stability.

Influence of Viscous Lubricant on Nucleation and Propagation of Frictional Ruptures

AbstractFluids are pervasive in the Earth's crust and saturate fractures and faults. The combination of fluids and gouge layers developing along faults can generate fluids of different viscosities. Such viscous fluids were found to influence the reactivation, frictional stability of faults, and eventually the dynamics of propagating earthquake ruptures. We reproduced laboratory earthquakes on analog material (PMMA) to study the influence of viscous lubricant on fault frictional stability, rupture nucleation, and propagation under mixed lubrication conditions. Experiments were conducted in dry conditions, and with fluids presenting a viscosity ranging from 1 to 1,000 mPa.s. Through photoelasticity, high‐frequency strain gauge sensors, and accelerometer measurements, we obtained new insights about the influence of lubricant on a characteristic nucleation length, rupture propagation velocity, and local slip and slip rate evolution during the reproduced frictional ruptures. Our experiments show that the presence of a lubricant generating mixed lubricated conditions along the fault induces, (a) a reduction of the frictional resistance, (b) an increase in nucleation length, (c) a decrease in the fracture energy. In addition, the larger the viscosity of the fluids, the larger the reduction of frictional strength and the increase in the nucleation length. Moreover, ruptures occurring under mixed lubricated conditions showed a pulse‐like rather than crack‐like behavior, suggesting that viscous lubrication can induce the transition from crack‐like to pulse‐like rupture along natural faults. We demonstrate, supported by existing theory, that this transition is mainly governed by the stress acting on the fault at the onset of nucleation, which is drastically reduced in presence of a lubricant.

Nucleation of Laboratory Earthquakes: Quantitative Analysis and Scalings

AbstractIn this study we use the precursory acoustic emission (AE) activity during the nucleation of stick‐slip instability as a proxy to investigate foreshock occurrence prior to natural earthquakes. We report on three stick‐slip experiments performed on cylindrical samples of Indian metagabbro under upper crustal stress conditions (30–60 MPa). AEs were continuously recorded by eight calibrated acoustic sensors during the experiments. Seismological parameters (moment magnitude, corner frequency and stress‐drop) of the detected AEs (−8.8 ≤ Mw ≤ −7) follow the scaling law between moment magnitude and corner frequency that characterizes natural earthquakes. AE activity always increases toward failure and is driven by along fault slip velocity. The stacked AE foreshock sequences follow an inverse Omori type law, with a characteristic Omori time c inversely proportional to normal stress. AEs moment magnitudes increase toward failure, as manifested by a decrease in b‐value from ∼1 to ∼0.5 at the end of the nucleation process. During nucleation, foreshocks migrate toward the mainshock epicenter location, and stabilize at a distance from the latter compatible with the predicted Rate‐and‐State nucleation size. Importantly, the nucleation characteristic timescale also scales inversely with applied normal stress and the expected nucleation size. Finally, we infer that foreshocks are the byproducts of the nucleation phase which is an almost fully aseismic process. Nevertheless, the seismic/aseismic energy release ratio continuously increases during nucleation, highlighting that, the nucleation process starts as a fully aseismic process, and evolves toward a cascading process at the onset of dynamic rupture.