Protecting water utilities from terrorist attacks and contaminants
a week to take samples, said David Hart, the lead Sandia software developer for CANARY.
Compared to weekly samples, CANARY works at lightning speed.
“From the start of an event — when a contaminant reaches the first sensor — to an event alarm would be 20-40 minutes, depending on how the utility has CANARY configured,” McKenna said.
The challenge for any contamination detection system is reducing the number of false alarms and making data meaningful amidst a “noisy” background of information caused by the environment and the utility infrastructure itself.
CANARY researchers used specially designed numerical algorithms to analyze data coming from multiple sensors and differentiate between natural variability and unusual patterns that indicate a problem. For example, the Multivariate-Nearest Neighbor algorithm groups data into clusters based on time and distance, explained Kate Klise, a numerical analyst at Sandia. When new data is received, CANARY decides whether it is close enough to a known cluster to be considered normal or whether it’s far enough away to be deemed anomalous. In the latter case, CANARY alerts the utility operator, Klise said.
The computer program uses a moving 1.5- to 2-day window of past data to detect abnormal events by comparing predicted water characteristics with current observations. But a single outlier won’t trigger the alarm, which helps to avoid costly and inefficient false alarms. CANARY aggregates information over multiple 2- to 5-minute time steps to build evidence that water quality has undergone a significant change, McKenna said.
“We’ve taken techniques from different fields and put those together in a way they haven’t been put together before; certainly the application of those techniques to water quality monitoring hasn’t been done before,” McKenna said.
CANARY also provides information about gradual changes in the water, McKenna said.
One unintended benefit of the software is that when utility operators better understood the data being sent by their sensors, they could make changes to the management of the water systems to improve its overall quality, McKenna said.
“What we found from utilities we work with is that a better managed system is more secure, and a more secure system is better managed,” McKenna said.
Harry Seah, director of the Technology and Water Quality Office at the Public Utilities Board (PUB), Singapore’s national water authority, wrote in a letter supporting CANARY that the software provided a “quantum leap” in the utility’s practice.
In the past, Seah wrote, the utility depended on preset limits of three water characteristics to determine water quality.
“With the implementation of CANARY, relative changes in the patterns of these three parameters can be used to uncover water quality events, even if each individual parameter lies within the alarm limits,” Seah wrote. “This dramatically improves PUB’s ability to respond to water quality changes, and allows PUB to arrest poor quality water before [it reaches] the consumers.”
As more versions of the software are installed at water utilities, researchers are working on new application areas for CANARY, such as computer network traffic logs and geophysical log analysis used by petroleum drillers to analyze rocks at different depths.
