Quantifying shear induced joint surface damage using acoustic emission and machine learning

Rock slope failure is one of the most common geological hazards in regions with hilly terrains, causing tremendous fatalities and economic loss worldwide. Hong Kong spends approximately one billion dollars annually on landslide hazard mitigation. However, there are still thousands of rock slopes that are posing landslide hazard risks. Maintaining such a substantial number of slopes is unfeasible and uneconomical, and this situation will be worsened by the rapidly expanding urban development. The current rock slope failure monitoring and early warning methods in Hong Kong rely on the information observed from the slope surface. However, rock joints in the rock slopes may be weakened progressively over time, and different parts of the joints may experience different degrees of degradation. This process has been identified as an important mechanism of rock slope degradation leading to geological hazards. This weakening process is still not well understood, mostly due to the lack of observation and measurement of the joint surface conditions. Field observations and laboratory studies showed that the weakening process of rock joints exhibits precursory geophysical signals in the form of small seismic events (known as microseismic events in the field and acoustic emissions in the laboratory). “Listening” to these seismic events may provide critical information about the joint degradation process. However, to date, a quantitative relationship between microseismic events and joint surface conditions has not been studied. Several obstacles are preventing us from achieving this goal: (1) the heterogeneous joint surface damage and degradation processes have not been quantitatively described, (2) there is no systematic characterization of the evolving seismic signals during the rough rock joint degradation, and (3) there is no framework to quantitatively correlate seismic characteristics and joint surface damage. This project aims to improve the understanding of the rock joint degradation and weakening processes using laboratory direct shear tests and acoustic emission monitoring, with the help of machine learning data analysis. We will use a novel spatial windowing data processing technique to take into consideration the spatial heterogeneity of the joint surface damage and roughness evolution, and machine learning models will be trained to predict the joint surface damage based on acoustic emission characteristics. The outcome of this project could provide a foundation for assessing joint surface damage using microseismic monitoring. The proposed method could help to identify the hazardous parts of rock slopes, which will facilitate more accurate and more efficient forecasting of potential slope failures.