Distributed Kalman Filter Investigation and Application to Leak Detection in Water Pipeline Monitoring Using Wireless Sensor Networks with Nonintrusive Sensors
Résumé
Wireless Sensor Networks (WSN) have found a wide range of applications in monitoring, with most deployments done in a centralized fashion. This results in high energy consumption and latency, as such centralized schemes are characterized by periodic long-distance transmissions. In this work, we demonstrate the benefits of trading off transmission for computation. We propose a solution where local and distributed computing are used to improve the accuracy of anomaly detection in physical systems without the need for long distance transmissions to some central base station. We practically demonstrate this in detecting leaks on a water pipeline testbed, since water losses via leaks is a major problem in most developing countries, including Cameroon. Unlike other works for leak detection in water pipelines available in literature, we build a low-cost sensor node, which
is feasible for deployment in developing countries from cheap off-the-shelf commercial elements. The accuracy of the measured vibrations on the surface of pipes is improved using a distributed Kalman filter, where every node independently computes the optimal state estimate used for leak detection by running a local Kalman filter to obtain an accurate local estimate from local measurements and also fusing it with those of its close neighbors. Results show that the distributed Kalman filter improves the reliability of leak detection.