Berry, J., Fleisher, L, Hart, W. Phillips, C. and Watson, J.-P. (2005) “Sensor Placement in Municipal Water Networks”, Journal of Water Resources Planning and Management, 131(3) pp. 237-243.
In the article, the authors discuss their research to find the best method for placing sensors in a water distribution system network. As the threat posed by terrorism has become more evident over the last decade, the potential vulnerability of water distribution systems has become a point of concern. Sensors to test for a variety of potential contaminants are under development, but it is important to also investigate the placement of sensors to ensure that a maximum number of people are protected while minimizing cost.
The authors used integer programming to decide where to place the sensors. They called their particular method Mixed Integer Programming, but I was unable to see how it was different from normal integer programming due to their descriptionl. Their method minimized the number of unprotected people, with DV's of whether or not to place a sensor at a location.
They used their MIP method on three pipe networks: two example networks from EPANet and one real network. For each network, they found the flow patterns every six hours in a 24 hour period (4 patterns per network). Then they estimated population density served by the nodes and the risk probability for each node, and introduced noise to account for the uncertainty of the exact values. Then they ran their MIP for each model to calculate the population not covered by varying numbers of sensors. They found that, as the number of sensors increases, the number of people not protected by the sensors decreases.
For me, this article was confusing at times. I couldn't really understand how their programming method is different than integer programming, and their explanation of how they were using noise wasn't clear. The discussion on sensor placement sensibility, while interesting, didn't seem to add much to their final conclusions.
I feel like, if sensors were to be placed in a real network using a similar network, it would be important not to neglect time as the authors did in this paper, as it is vital when developing emergency management procedures for once the contaminant is detected. Also, the placement of sensors (industrial vs commercial vs residential) could become a contentious issue for a real network, whereas this paper glosses over that.