Multi-Sensor Data Fusion for Detection of Sensor Drifts in a Landslide Monitoring System

This paper aims to extend the lifetime of polymer dielectric-based humidity sensors used in landslide monitoring systems by mathematically adjusting the sensor's drifting value as time passes. Therefore, a strategy based on nested Kalman filters, cosine similarity, and Euclidean distance was proposed in this paper. An outer Kalman filter is used top redict humidity values, with the prediction based on an inner Kalman filter that fuses data from surrounding sensors. To adjust the process noise covariance of the nested Kalman filter, cosine similarity and Euclidean distance are used. When the sensor is in normal conditions, the process noise covariance remains constant; however, when the sensor is drifting, the process noise covariance changes depending on the angle and distance between the value from surrounding sensors and the drift. The proposed method could reduce the drifting value error by 14.85 percent, according to the experimental results.