Development and Implementation of a Real-Time Rockfall Warning System Using Seismic signal and machine learning analysis

Following an initial landslide in Taiwan, frequent post-failure events, primarily rockfalls with occasional debris flows, pose risks to the safety of road users on a road section next to the bare land slope. To address this issue, a comprehensive warning system has been developed. This system utilizes two seismometers strategically positioned at the crown and toe of the landslide. This configuration effectively captures the physical processes of rockfalls, with the elevation difference between the stations correlating to the time difference in their peak ground velocities. Eleven seismic parameters are employed for initial rockfall detection. Subsequently, a machine learning model, trained on over 100,000 spectrograms, is implemented as a secondary filter to minimize false alarms. Additionally, the system assesses rockfall risk levels by calculating nighttime rockfall activity (from 6 PM to 6 AM) to determine a daily risk level communicated through a traffic light concept. Furthermore, the system integrates local acceleration and rainfall data to address potential coseismic rockfalls and debris flows. This data is transmitted to local electronic boards on both sides of the landslide, displaying the corresponding rockfall/debris flow risk levels with red, yellow, and green lights. Overall, this multi-tiered approach facilitates immediate hazard alerts and proactive risk management. The system provides a robust and adaptable solution for real-time warnings and risk assessments related to rockfalls and debris flows, ultimately enhancing road safety and management efficiency in hazard-prone slopes.