Double Direct Shear Experiments as an interacting two fault system: insights from laboratory seismic cycles on fault interaction

Double direct shear experiments serve as established methods for delving into the physics of laboratory earthquakes. Using a biaxial shearing apparatus with dual fault configurations, these friction experiments simulate real Earth faults' behaviors during loading and failure. 

Despite the presence of distinct layers, double direct shear experiments are commonly perceived as a unified fault system, where the evolution of fault zone properties captured through passive or active seismic imaging can be correlated with the instantaneous stress state affecting both layers uniformly. To further explore the physics of seismic cycles generated in this setup, we perform friction experiments aiming to independently monitor the behavior of each fault layer. In our experiments, we use granular quartz (medium grain size 40 µm) to simulate fault gouge, amd we vary the normal load and shear velocity, allowing us to modify the apparatus's loading stiffness, which relies on the critical fault rheologic stiffness (kc). In the Rate-and-State framework, increasing the normal load results in an  increase of kc, pushing the system towards instability, occurring when k/kc <1, where k is the fault stiffness. Under 50 MPa of  normal loads and 10 µm/s of loading rate, these conditions result in highly non-cyclic seismic cycles marked by significantly variable stress drops and recurrence times. This situation offers an exceptional opportunity to investigate stress partitioning between the two layers and understand their interactions. Experiments are monitored using high-frequency calibrated piezoelectric sensors with a sampling rate of 6 MHz, placed on each of the two forcing blocks. Such a sampling rate allows us to clearly distinguish the time delay between the Acoustic Emissions (AEs) generated from microslip events in different layers. Phase arrivals are detected using retrained, Deep Learning-based algorithms. By associating these phase arrivals using the DBSCAN clustering algorithm, we classify events as occurring on a single gouge layer or on both layers. Subsequently, we analyze the catalog of AEs,, and single seismic waveforms, in terms of general characteristics and frequency content, to look for differences in the physical sources generating them. Unsupervised clustering may help identify classes of AEs linked to specific stages within seismic cycles. By potentially using established supervised Machine Learning technique, it would be possible to verify the relation between AEs variance for each layer and macroscopic apparatus features, like instantaneous friction or time to failure. All of these techniques reveal differences in acoustic energy released before failure for each layer, observations that can be associated to changes in fault physical properties, asperity scale processes and/or  grains sliding or fracturing. In conclusion, our findings demonstrate that double-direct shear experiments can emulate a system of interacting double faults. In such a context, the continuous monitoring of AEs can provide insights into the stress partitioning between the two layers, a process that may guide the nucleation of major slip events as well as the long term behavior of the system. Additionally, our analysis may be helpful to investigate processes like fault interactions, faults synchronization, static, and dynamic stress triggering.