|Ph.D Student||Sankary Nathan|
|Subject||Inclusion of Uncertainty in Sensor Network Optimal|
Operation and Design for Water Distribution
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Avi Ostfeld|
|Full Thesis text|
The primary goal of a water distribution system is to deliver clean, potable high quality water to consumers at their desired volumes. A fault in the WDS, weather of malicious intent or accidental is immediately felt by the connected population through discontinued service, or delivery of poor quality water and depending on the characteristics of the fault, large populations may be affected.
With increased interest in securing WDS operation, and increased focus by governments, commercial entities, and academia, monitoring water quality within the WDS has greatly progressed during the last 15 years. This Ph.D. is a continuation of this effort by employing a newly developed inline mobile sensor, previously designed to operate within WDS pipes, carried by flow, while simultaneously measuring critical water quality parameters and wirelessly uploading these data to nearby ground transceivers for analysis.
The network of fixed and inline mobile water quality monitoring sensors’ (termed the early warning system (EWS)) overall performance is highly sensitive to the locations at which the fixed sensor are placed, and the inline sensors are input. This Ph.D. has strived to develop sensor networks with fixed sensors and mobile sensors placed or input at locations which provide the best performance in the face of a potential contamination event, under the operational conditions expected to be imposed on the system, specifically, uncertainty in consumer demand, and background water quality.
Ultimately, the inline mobile sensor proposed herein requires operation and maintenance costs which are too high to justify its use for continuous monitoring of a water distribution system. However, the methods developed in this Ph.D. to incorporate operational uncertainty in to system design show promise for designing robust early warning systems. For the first time, surrogate water quality parameters have been employed for contamination event detection in the sensor placement task.
Results of this Ph.D. show the importance of considering uncertain operation in sensor network design. In a majority of previous studies, the sensor placement task has been performed using deterministic WDS models. Although deterministic models are simpler to implement, their resultant sensor networks show high failure rates when operating in an uncertain environment.
Placing sensors to detect contamination via surrogate water quality parameters has shown to greatly influence the system’s performance. When evaluating the performance of a sensor network to detect contamination via surrogate water quality parameters (as will be required in the real-world) benchmark sensor networks previously developed without considering surrogate water quality parameters or water quality uncertainty lead to much higher failure rates than a sensor network designed with consideration to uncertain water quality, and detection via surrogate water quality parameters.