Engineers have collected a large amount of data from monitoring systems used to detect potential landslide hazards. However, it is difficult to predict slope failure using these data and the landslide process involves complex phenomena with nonlinear time dependency and seasonal effects. Here, we apply principal component analysis (PCA) to detect abnormal events in field data collected on a slope over a period during which it collapsed. Two major monitoring indices are calculated including the T2 statistic and Q statistic. An abnormal event will cause producing deviations from control limits in at least one of these two values. The results show that the T2 statistic represents an abnormal state of the I step, whereas the Q statistic represents an abnormal state of the III and predicts slope collapse at the highest confidence limit. Therefore, the results show that the PCA can be successfully applied to the detection of abnormal events. This method is expected to be a useful tool in slope monitoring and alarm systems.
Collapse or failure of a slope cannot be predicted exactly. Therefore monitoring systems are installed. Monitoring systems provide information regarding damage and abnormal events on a slope before a collapse or failure. With technological advancement monitoring instruments are becoming more diversified and intelligent. However, there are established alert thresholds for only certain values such as surface displacement and inclination. A more unique approach using an analysis model cannot be easily applied to the data because of nonlinear time dependency and seasonal effects. Therefore, such an approach cannot be defined for modeling landslide movements. In this study, Principal Component Analysis (PCA) was applied to a slope with history of collapse. PCA, being a data-driven approach, is advantageous in modeling systems whose geomechanical properties are unknown or difficult to measure.
