Mechanical and Civil Engineering Seminar: PhD Thesis Defense

Abstract:

Induced seismicity – earthquakes driven by injections of fluids into the subsurface – is of growing societal importance in its impact on clean energy technology. Advancements central to the world's transition to a greener economy such as geothermal energy and long-term geologic storage of CO2 are hampered by a lack of understanding and control of the associated seismic hazards. Frictional processes in the presence of fluids are difficult to model given the challenges of studying frictionally unstable material in a controlled environment. Yet, unstable gouge material is commonly found along faults in nature, due to the breakage (comminution) of brittle rock into granular layers called 'gouge'.

This thesis will approach the challenge at two different scales: 1. at the scale of the localized shear layer in gouge where we model laboratory earthquakes in the presence of pressurized fluids, and 2. at the scale of a reservoir where we model the rate of earthquakes given the injection/extraction schedule. Laboratory earthquakes are driven in granular quartz gouge with small grain sizes as typically encountered in nature. Such materials exhibit complex deformation history that is difficult to interpret using conventional methods. We employ a Bayesian inversion framework to infer the rate-and-state frictional properties of the unstable quartz material and observe their dependence on the slip history and the presence of pressurized fluids. The inversions indicate that the presence of pressurized fluids increases the frictional stability of the gouge material, beyond what can be explained by Terzaghi's effective stress principle. The inversions also reveal, by exhaustive search of the parameter space, that the current formalism of the constitutive law for friction cannot reproduce the nucleation and arrest behavior observed in the laboratory. The frictional behavior in the gouge is extrapolated to the reservoir scale, where we model the seismicity catalogue of the 2018 geothermal well stimulation in Helsinki, Finland. We first use the popularized Dieterich model to successfully reproduce the observed seismicity rate history. Then, we use a discrete fault network model to identify the biases caused by the model's assumption that the population of earthquake sources is infinite. We connect the biases directly to field observations, namely, the co-injection back-propagation front and the decreasing rate of smaller events with time in Soultz-Sous-Forêts. The results at both scales highlight physical mechanisms whose incorporation into future models for frictional processes in the presence of fluids may help us better understand the hazards posed by induced seismicity.