Multi-Sensor data fusion for landslide early warning system

Landslides are a significant risk to human lives and result in great damage to the physical and organizational structures and facilities (e.g., buildings, roads, and power supplies) needed for the operation of a society or enterprise. Areas with unstable slopes are prone to landslides and thus such areas, need a monitoring system that can provide an alarm before the sliding happens. There are many landslides monitoring systems around the world, but their sphere and accuracy are limited due to the technique used to transmit data, the power sources, cost, the type of instruments they use, and the technique they used to process the data. Since the advance of low-cost sensors with low power consumption such as MEMS (micro-electro-mechanical systems) and their development in WSN (wireless sensor network) new technologies have been developed in areas such as power consumption, data transition, and communication which enable the collecting of real-time data with high rates by using remote sensing and monitoring systems. This thesis is concerned with developing an early warning system for landslides as well as discussing its features and drawbacks. The surface deformation in the landslide areas was monitored by MEMS Inertial Measurement Unit (IMU) integrated with a Global Navigation Satellite System (GNSS) and combined with other geotechnical data. A Wireless Sensor Network (WSN) was utilized for data transition between sensors and the server computer. Furthermore, to enhance the quality of data and to decrease error rates, an adapted model from Joint Directors of Laboratories (JDL) for Multi-Sensor Data Fusion was applied for landslide monitoring. A demonstration of the modified model is provided step by step with either a methodological description or simulation. Briefly, the scheme of the adapted JDL model is conceptualized by three divisions: Data Source, Data Fusion Domain, and Human/Computer Interface. In the Data Source Domain, different types of monitored data such as acceleration, angular velocity, attitude, velocity, and position from distributed sensors within the network are measured and (measurements related to soil mechanics) details on geology and tectonics are collected from geotechnical engineers. Then the collected data saved in order to be processed in the Data Fusion Domain. The Data Fusion Domain processes the raw data corresponding to a gradual hierarchy. This allows an assessment of the situation of the observed phenomenon and a prediction of the possibility of failure. Data mining techniques, numerical simulation, and statistical methods are implemented accordingly. At the end of processing, the results are reported to the Human/Computer Interface where the prediction can be displayed, and subsequent action taken. This work was not performed in the real world, but a simulated test of possible realistic implementation options was performed. Finally, discussion, outlook, and conclusion derived from this research are presented