Search

Distributed acoustic sensing (DAS) sees increased utilization in the seismological community in recent years and various applications are investigated for the usage of DAS in different branches of seismology.Strong-motion seismology uses records of earthquakes of engineering concern (MW>4.5) with hypocentral distances within few hundreds of kilometers. This demands dense networks over a wide area and installation of typical strong-motion instruments (accelerometers) can be achieved quickly and at a reasonable budget, compared to other network types. For DAS, installation and operation are more involved, and deployment is very still limited. Consequently, DAS recordings of nearby large events are still very unlikely and rare compared to accelerometers.On September 18, 2022, a shallow earthquake sequence with a MW 6.9 mainshock struck near Chishang (Taiwan) and was recorded by DAS in Hualien city, appr. 100 km north. Shaking of the mainshock and several aftershocks were noticeable in Hualien, though not damaging with PGA recorded at 0.28 m/s^2 nearby the DAS site. The DAS campaign was originally conceptualized as a test suite with different fiber installations: including buried, within a gutter (as in commercial fiber installation) and loose within a basement. The test site is in an urban area affected by surface rupturing during the 2018 Hualien earthquake.The presented recordings provide not only an unprecedented insight how strong-motion appears on DAS but also how effective different installation techniques are for this kind of event. The waveforms are also compared to records of a collocated broadband seismometer and an accelerometer 1 km away., The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)

The wide adoption and significant computing resource consumption of attention-based Transformers, e.g., Vision Transformer and large language models, have driven the demands for efficient hardware accelerators. While electronic accelerators have been commonly used, there is a growing interest in exploring photonics as an alternative technology due to its high energy efficiency and ultra-fast processing speed. Optical neural networks (ONNs) have demonstrated promising results for convolutional neural network (CNN) workloads that only require weight-static linear operations. However, they fail to efficiently support Transformer architectures with attention operations due to the lack of ability to process dynamic full-range tensor multiplication. In this work, we propose a customized high-performance and energy-efficient photonic Transformer accelerator, DOTA. To overcome the fundamental limitation of existing ONNs, we introduce a novel photonic tensor core, consisting of a crossbar array of interference-based optical vector dot-product engines, that supports highly-parallel, dynamic, and full-range matrix-matrix multiplication. Our comprehensive evaluation demonstrates that DOTA achieves a >4x energy and a >10x latency reduction compared to prior photonic accelerators, and delivers over 20x energy reduction and 2 to 3 orders of magnitude lower latency compared to the electronic Transformer accelerator. Our work highlights the immense potential of photonic computing for efficient hardware accelerators, particularly for advanced machine learning workloads., The short version is accepted by Next-Gen AI System Workshop at MLSys 2023

Understanding fault zone dynamics in multi-scale is important to embrace the complexity of the earthquake behavior and its natural system. However, the opportunity to map and observe the fault zone behavior at depth with high spatial resolution are rare as also the challenge itself on targeting and identifying the fault zone at depth. We placed a 3D cross-fault fiber array with a downhole loop from surface to depth of 700m for Hole-A (Hanging wall site, crossing fault at depth), after drilling and coring to a frequent slip fault, Milun fault in a plate boundary zone, which ruptured during the 6 February 2018 Mw6.4 Hualien earthquake, and resulted in severe damage to several tall buildings with tens of casualties and injuries. Then, the surface segment crosses the surface fault rupture zone using commercial fiber, and to another downhole loop of 500m fiber for Hole-B (Footwall site). The high spatial resolution from distributed acoustic sensing (DAS) allows us to characterize the fault zone feature together with the retrieved core and geophysical logs after drilling through this frequent slip zone. This 3D route includes the experiment of using commercial fiber to the future application of surface rupture zone identification for seismic hazard mitigation. The project successfully retrieved the fault core associated with Milun fault zone, which could be also seen in geophysical logs with low velocity and resistivity, and mapped using Optical Fiber Sensing technique of the downhole fiber. Within the Milun fault zone, while a 20m thick fault core with grey and black gouge was discovered, a distinct seismic feature associated with this 20m fault gouge was found by its amplification of the strain records from DAS. This amplification ratio is about 2.5-3 when compared to the channels at deeper depth related to a consolidated rock material. This amplification factor was frequency and azimuth independently, as genuinely observed from all events (e.g. local, and teleseismic earthquakes) with similar amplification factor. Our study shows that the amplification from this 20m fault gouge zone is mainly from the nature of the heterogeneous medium in elastic constant while crossing the fault zone, especially the fault core. Similar feature at surface but with wider surface rupture zone (~ 200m) was found in DAS data as well although less evidence using commercial fiber, while could be validated from the densely deployed geophones crossing the surface rupture of the 2018 Hualien earthquake. Through the depth, a high-resolution asymmetric feature of this active fault was evidenced from the downhole optical fiber and cores. This fault zone behavior would be hardly seen or confirmed without continuous viewing of the wavefields to this high spatial resolution to meter scale. Although the narrow fault gouge, the nature of its amplification in strain due to its strong material contrast from fault gouge was intriguing, and requires intensive attention to consider the contribution of the fault zone heterogeneity in the medium. This might give hints on the understanding of the observation of earthquake dynamics triggering reported worldwide after the occurrence of a mega-earthquake.

Distributed Acoustic Sensing (DAS) is used to record high-spatial resolution strain-rate data. For ground motion observation, the DAS data can be converted from strain rate to acceleration or velocity by array-based measurements with coherent plane waves. DAS provides an opportunity to map high-resolution shaking patterns near faults. We installed collocated geophones and optical fiber in Hualien City (a very seismically active area in Taiwan) from the end of January to the end of February in 2022. Earthquakes with magnitudes (Mw) between 3.2 and 5.4 have been recorded. These records illustrate the typical magnitude-distance dependence of ground-motion but also show saturation for higher magnitudes and/or at shorter distances (e.g for an earthquake of Mw 5.2 earthquake recorded at 100 km). For frequency-based analyses, clipped signals on DAS result in challenges not present in classical instruments (seismometers). The upper limit in dynamic range of seismometers results in easily identifiable trapezoidal signals. The dynamic range of DAS interrogators is limited by gauge length, sampling frequency, and wrapped phase in the interferometric phase demodulation. We observe that clipped DAS signals not only affect time series but also contaminate their spectra on all frequencies, due to the random nature of clipping in DAS—contrasting to the flat plateaus in clipped time series on seismometers. Therefore, the identification of the start and end points of clipped DAS records poses a major challenge, which we aim to resolve with a neural network. This approach enhances the efficiency for quality control of massive DAS datasets., The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)

The Taiwan Milun fault zone located at the boundary between the Eurasian and Philippine Sea plates. This fault slips frequently and produced large earthquakes, as for example the Mw6.4 Hualien earthquake (6 February 2018). We map and observe the fault zone and its behavior at depth by high spatial resolution dynamic strain sensing with optical fiber. In 2021-2022, we drilled and cored the fault, and deployed a 3D multi-cross-fault fiber array comprising a borehole loop with a depth of 700 m (Hole-A, Hanging wall site, crossing the fault at depth), a surface array crossing the fault rupture zone using commercial fiber, and a second borehole loop of 500m fiber (Hole-B, Footwall site). The high spatial resolution from distributed acoustic sensing (DAS) and the retrieved core combined with geophysical logs allow us to characterize the structure on meter-scale. Within the Milun fault zone, we identified a 20-m wide fault core comprised of gray and black gouge in the core sample. DAS strain-rate records associated with the same depth as the fault core show a distinct amplification. The amplification ratio of 2.5-3 is constant as for all types of events (local, teleseismic ), when compared to DAS channels at larger depth, related to a consolidated rock material. Although the fault gouge is narrow, the nature of the amplification in strain is due to its strong material contrast from fault gouge. This result may shed the light on the understanding of fault-zone dynamics in terms of remote earthquake triggering and near-fault ground motion., The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)