By Dan Kroll
Multi-parameter monitoring with interpretive algorithms is becoming the method of choice for water security monitoring. These systems have some potential to generate false alarms. Addressing and eliminating possible sources of error in a systematic and focused manor can help to correctly interpret results form such systems.
Warning Systems and Algorithms
The purpose of early warning systems is to alarm if water quality deviations become excessive. In some cases, these software programs are capable of analyzing these deviations to see if they match an agent fingerprint. It is important to understand that these sorts of systems are only capable of a presumptive classification. Any match is only indicating that the water quality sensors readings have changed in the same manner that they would be expected to if the agent in question were present aka in inferential method This is simply a signal to investigate further and can act as a guideline to that investigation.
Trigger Shape and Duration
There is valuable information that can be gleaned from the duration and shape of alarm signals. Spikes of a few minutes duration are of less concern because they affect only a small quantity of water. A change that is continuous and persistent is of more concern.
An actual agent will usually present a characteristic rise time and a plateau of stabilization. Then a drop off will occur when the contaminated water has moved past the sensors. It is possible that this kind of pattern will produce classification of different agents on the signal rise, plateau, and fall. Classifications on the rise and fall are not as reliable as those from the plateau of the response.
Responding to Unknown Alarms
When a trigger occurs there could be any number of causes. Some analysis must be done to understand the problem. The first concern with a validation of an alarm is the sensor set. There are a number of things that should be considered.
• Are there any sensor alarms on the instruments that might be the cause of the trigger alarm?
• Are all sensors functioning normally?
• Have the sensors been properly calibrated? Do the sensors give similar results to those obtained with independent verification techniques such as handheld pH meters etc?
• Have the sensors been properly maintained?
• Have any reagents run out? Running out of reagents on some instruments can dramatically shift readings. Those shifts can trigger an alarm.
• Have flow rates to the instruments changed? If available it is a good idea to check pressure readings to see if there has been any unusual pressure changes that might affect flows.
If all of the sensors are operating without problems, then the sampling systems should be checked.
• Are there any blockages or leaks in the sampling system?
• Has the sampling system been shut off for some reason?
• Are there air bubbles in the sample line?
• Has the system been tampered with or vandalized?
The Algorithm and computer systems should be considered.
• Is there a loss of communications?
• Are there sensor alarms (Hi, Low, Frozen)?
• Low-pressure alarms from any of the instruments?
• Are there any sensor diagnostic messages?
If these systems are all in order, then it is likely that the alarm is caused buy a real change in water quality. A key question then becomes: Is the cause attributable to know operations or is it something unknown? Data needs to be gathered regarding the operations upstream of the sampling point to see if there is any rational explanation for the change.
• Are there unusual weather conditions?
• Has work been done on or near water mains?
• Are there changes in treatment plant operations?
• Has there been a switch in source waters or a known change in these sources' quality?
• Are other monitoring sites responding in a similar manner and, is the response distribution such that it could indicate a change in the source water?
• Are different treatment chemicals being used?
• Is there maintenance occurring at the treatment plant or in the distribution system.
• Are there unusual water demands?
• A water main break? Hydrant flushing?
• Are control and feed systems for pH, chlorine, fluoride, ammonia or other treatment chemicals functioning normally?
These are just some of the possible causes that should be considered when investigating an alarm. Some early warning systems allow the operator to associate a name and an alarm priority with patterns produced by such events so investigations of root cause can be more effective over time.
Analysis of Unknowns
Many event detection systems are designed to be equipped with an automatic sampler that will draw a representative sample of the water when. The utility should have adequate measures in place for analyzing samples that could be toxic or infectious. If the system presents a classification of a likely agent, this classification should be treated as tentative until verified by further testing. Classification of an agent by this type of system offers a valuable first pass at determining a cause and can be used to direct further analysis. Tests should be tailored to address the class of agent being presented by the match.
Just because an alarm is not classified by an event detection system with a match to a known agent does not mean that the alarm is benign. Classification algorithms usually only contain a small subset of likely compounds. Many others are not. That these systems have the ability to trigger on such compounds even if they are not specifically found in the library is one of the great strengths of the multi-parameter inferential method. While no classification is given on these types of alarm, the information presented by the individual parameter sensors can be important in guiding forensic analysis. For example if an alarm occurs that is the result of changes in conductivity with no noticeable changes in Total Organic Carbon levels or chlorine residual, it would make no sense to waste valuable time in doing an analysis focused on organic contaminants
While not all inclusive, the concepts and considerations presented here should make interpreting and responding to alarms form early warning systems a more routine event. If followed with common sense these steps and procedures can result in increased reaction time and proper response to a variety of situations.
About the Author: Dan Kroll is chief scientist at Hach Company's Homeland Security Technologies division, in Loveland, CO. He has been the lead researcher on a variety of method development projects for the physical, chemical and microbiological quality of water and soils for which he holds several patents.
WaterWorld Online, August 2010