Adaptive wireless sensor network for real-time monitoring of water quality

Water quality problems have appeared in many places all around the world, and have caused severe public health problems. In identify the quality of different aquatic environments, wireless sensor networks have been used for monitoring large geographic areas of interest (AOI). Among the challenges of using wireless sensor networks for water quality monitoring in large areas, sensor node deployment strategy is a key consideration since an optimal sensor node deployment strategy can ensure the most appropriate utilization of the limited monitoring resources (sensor node, incorporated sensors, power supply, monitoring rates, etc.). To tackle such problems, we in the Industrial Automation Laboratory (IAL) of the Department of Mechanical Engineering, the University of British Columbia (UBC) have developed a mobile wireless sensor network for water quality monitoring. It has mobile (dynamic) sensor nodes, which can move to best sensing locations, and the ability to sense key water quality attributes. The developed platform is equipped with multiple nodes each of which having basic water quality detecting sensor probes, supports up to six propellers, and has upgradeable wireless communication boards. Besides, we have also proposed an optimal sensor node deployment strategy called “Rapid Random exploring tree with Linear Reduction” (RRLR) for this mobile wireless sensor network. The proposed method removes redundant sensor nodes depending on the linear dependence of sensor readings at the current deployment location without losing information. In this manner, spatial-temporal correlation of sensor node deployment in large geographic AOI can be minimized. The developed platform is demonstrated to have good performance even when moving against water flow and has low packet loss rate (0.85%) while transmitting data under different types of obstacles in real-world tests. Furthermore, the developed optimal sensor node deployment strategy, RRLR, requires nearly 60% fewer sensor nodes to achieve the same estimation error as our benchmark.Applied Science, Faculty ofMechanical Engineering, Department ofGraduat