- Users Guide (Unofficial)
- Frequently Asked Questions (FAQ)
- Related Publications
- Mailing Lists
- Nightly Tests (Current Stable))
- Nightly Tests (Current Beta)
- Nightly Tests (Trunk))
This website hosts the bug-tracking, wiki, and downloads for the CANARY software. Use the wiki to search for more detailed information than may be found in the user's manual, including frequently asked questions and other documentation. Search the tickets database for bug reports or feature requests, or submit a new bug report. The downloads section contains the most up-to-date versions of CANARY, in both binary and source form, and has links to additional documentation, such as tutorials and webinars.
CANARY is part of the TEVA Project.
If there are errors on the wiki, please enter a ticket with a link to the error and a description - only registered users can modify the wiki at this point. Thanks!
Introduction (from the CANARY User's Manual)
Contamination warning systems (CWSs) have been proposed as a promising approach for reducing the risks associated with contamination of drinking water. In order to maximize detection likelihoods, CWSs can incorporate multiple detection technologies, such as online continuous water-quality monitoring, public health surveillance, physical security monitoring, and customer complaint surveillance. The goal of a CWS is to detect contamination incidents in drinking water systems rapidly enough to allow for the effective mitigation of adverse public health and economic impacts. In 2006-2010, the U.S. Environmental Protection Agency (EPA) is deploying and evaluating CWSs at a series of drinking water utilities.
With current technology, the online monitoring component of a CWS is based on available water quality sensors that measure, for example, free chlorine, total organic carbon, electrical conductivity, oxidation-reduction potential, and pH. Recent research has shown that many contaminants of concern will cause detectable changes in these water quality parameters (Hall et al., 2007), (US EPA, 2005). However, these parameters are known to vary considerably over time in water distribution systems due to normal changes in the operations of tanks, pumps, and valves, and daily and seasonal changes in the source and finished water quality, as well as fluctuations in demands.
Data analysis tools are needed to distinguish between normal variations in water quality and changes in water quality triggered by the presence of contaminants. Often referred to as event detection systems (EDS), such data analysis tools can read in SCADA data (water quality signals, operations data, etc.), perform an analysis in near real-time, and then return the probability of a water quality event occurring at the current time step. A water quality event is defined as the period in time within which water of unexpected characteristics occurs. The CANARY software described here provides a continuous measure of the probability of an event to a water utility operator.
The goal of CANARY is to take standard water-quality data and use statistical and mathematical algorithms to identify the onset of periods of anomalous water quality, while at the same time, limiting the number of false alarms that occur. The working definition of “anomalous” can be set by the user by selecting the configuration parameters. These parameters may be configured differently from one utility to the next and may even need to vary across monitoring stations within a single utility. CANARY can be set up to receive data from a SCADA database, and return alarms to the SCADA system. In addition, it can be run on historical data to help set the configuration parameters in order to provide the desired balance between event detection sensitivity and false alarm rates.
CANARY is designed to be extensible, allowing outside researchers to develop new algorithms that can added to CANARY. In this version, there are several change detection algorithms within CANARY, including: a linear filter, a multivariate nearest-neighbor algorithm, and a set-point proximity algorithm. These algorithms identify a background water quality signature for each water quality sensor and compare each new water quality measurement to that background to determine if the new measurement is an outlier (anomalous) or not.
Please see the Related Publications page for references and other publications.