Finding the source

released 08.23.07

By J. William Bell

Using real-world data, engineers improve their methods of pinpointing water contamination sources.

Urban water distribution systems cover hundreds of square miles and include thousands of miles of pipe. But in most cases, drinking water is largely unmonitored after it leaves the treatment plant. "It's possible to intentionally contaminate a water supply using very rudimentary equipment," says Jim Uber, an environmental engineering professor at the University of Cincinnati. "Obviously, this is a threat to the health and economy of any urban area."

With this in mind—and more day-to-day goals like reducing customer complaints by pinpointing causes of reduced water quality—Uber and North Carolina State University's Kumar Mahinthakumar, Ranji Ranjithan, and Downey Brill develop new methods of locating the source of contaminants and testing approaches to limiting their impact. Based on data from sensor networks in large metropolitan areas, they use evolutionary computation to simulate various possible sources, assess the results, and launch another set of simulations. Eventually, the hypothetical sensor data from a possible solution matches real-world sensor data, and they can find the source.

This procedure uses hundreds of processors simultaneously on TeraGrid systems at NCSA, the San Diego Supercomputer Center, and the University of Chicago/Argonne National Lab, automatically figuring out how many jobs to send to what site based on the length of the systems' queues. This cyberinfrastructure for source identification in water systems—by team members at North Carolina State, Cincinnati, the University of South Carolina, and the University of Chicago—was presented at the 2007 International Conference on Computational Science.

Using real-world data from the Greater Cincinnati Water Works, they've already run numerous simulations on a "skeletonized" metropolitan distribution system. "Most of our simulations look at a few hundred nodes, and we’re working on one with 11,000," Mahinthakumar says. "But a whole network in a city could have 300,000 nodes." Cincinnati, through the U.S. Environmental Protection Agency's Water Sentinel program, is installing just such a system.

Current simulations are already showing officials how to cope with problem situations. In one case, the team is learning when to engage different or new sensors in cases in which some sensors within the network malfunction. Their simulations have also allowed the team to improve the fidelity of their algorithms. For example, the code handles issues of what they call "nonuniqueness" better than ever before. "Two different contamination sources can present very similar sensor profiles, so methods for identifying distinctions in these cases is critical," Mahinthakumar says. These improvements were presented at the World Environmental and Water Resources Congress 2007.

For further information: http://www.secure-water.org/

This research is supported by the National Science Foundation's Dynamic Data Driven Applications Systems program.

Team members
Downey Brill
Ken Harrison
Jitendra Kumar
Li Liu
Kumar Mahinthakumar
Ranji Ranjithan
Sarat Sreepathi
Jim Uber
Gregor von Laszewski
Emily Zechman

del.icio.us Slashdot