Researchers Develop Decision Support System to Help Prevent
Plant Upsets
Whether caused by unauthorized discharge, accidental spill, or
terrorist activity the presence of toxic compounds in the influent to a
wastewater treatment plant is an obvious cause for concern for the plant
manager and staff, as it can cause serious upsets to biological
treatment processes. Research from two WERF projects provides valuable
information that will enable wastewater treatment plants to develop a
decision support system (DSS) that could reduce or potentially eliminate
the impact of suspected toxic compounds on the facility.
The first project, Feasibility Testing of Support Systems to
Prevent Upsets (03CTS7S), was led by Andrew Shaw of Black &
Veatch. Shaw and his team developed a DSS prototype based on
the study of techniques used in other industries, as well as real word
accounts from more than a hundred treatment facilities throughout the
United States. Determining and Assessing Corrective Action
Strategies for Treatment Plants Exposed to Chemical Toxins
(04CTS11S), which was led by Nancy Love, Ph.D., of the University of
Michigan, built on the lessons learned in that earlier project with the
development and testing of a DSS framework that wastewater treatment
plants can use to develop a site-specific system to select and carry out
corrective actions.
Laying the Groundwork
Research for Feasibility Testing of Support Systems to Prevent
Upsets began with the collection of background information on
decision support systems (DSS), including a comprehensive review of
software and data analysis techniques used by various industries to
detect upsets. The review served as the basis for an assessment of
techniques most appropriate to the wastewater industry and suitable for
inclusion in an electronic DSS that can link to a plant’s
supervisory control and data acquisition (SCADA) system to help
operators prevent plant upsets. In the process, the research team
identified four key features for a good DSS designed to prevent plant
upsets:
- The DSS should detect anomalies or unusual activity as early as
possible.
- Various anomaly detection methods are available, and all of them
require a database of historical data, which plants can use as the basis
for detecting unusual conditions.
- A case-based reasoning system (CBRS) is an approach to
computer-based learning from the field of artificial intelligence (AI)
that can provide a useful approach for a DSS. A CBRS uses cases, which
are examples of past problems encountered and solved. Cases are
organized in a database called a case base. A CBRS looks for patterns in
the attributes of the current case and compares them to those for cases
in the case base.
- Many online instruments and analyzers are available that can be used
to provide data for the DSS. The development of these devices is moving
forward rapidly.
In addition to the review, the research team conducted a survey of
more than 100 wastewater treatment facilities, mostly in the United
States, to see how many of them experienced upsets, what they thought
caused the upsets, and the approaches they took to mitigate the impacts.
The most common cause of upsets was toxic organic shock loads (i.e. a
toxic substance in the influent caused a plant upset), which frequently
resulted in ineffective nitrification, deflocculation of the activated
sludge floc, or foaming. The most commonly used methods for upset
mitigation were influent diversion (i.e. diverting influent flows to a
holding tank) and “source determination and termination”
(i.e. finding out where the toxic substance came from and taking action
to prevent it from happening again).
Having established baseline information about plant upsets and
potential DSS elements, the project went a step further by developing a
DSS prototype using online analyzers and a separate array of sensors in
the primary effluent of a wastewater treatment plant (hereinafter Plant
1). These instruments provided data to which data mining techniques were
applied in order to test for anomalies that would indicate the presence
of known toxins added in the test rig. The online analyzers included two
respirometers and a spectrophotometer. Sensors in the separate array
included pH, oxidation-reduction potential (ORP), dissolved oxygen (DO),
temperature and conductivity. Clustering of the data from the pH, ORP,
and conductivity provided very promising results for many of the tested
contaminants. The research team identified several practical issues
through the prototype testing which they believed should be
addressed in a full-scale DSS:
Automated real-time data from sensors located as near to the plant
inlet (or in the sewer system) provide the best means for being able to
initiate action to prevent plant upsets; however, it is difficult to
maintain sensors and analyzers in the high-fouling environment in these
locations.
- Maintenance of primary sensors to provide consistent data is
imperative for anomaly detection.
- Data filtering is required to enable the DSS to distinguish between
anomalies and other routine activities such as sensor maintenance and
sensor faults.
- The DSS needs to collect, analyze, and store large amounts of
data.
- Automation can provide rapid information for a DSS, but it is
important to include humans in the ultimate decision making process for
carrying out corrective actions.
Many of these issues were addressed in the follow-on project
Determining and Assessing Corrective Action Strategies for Treatment
Plants Exposed to Chemical Toxins (04CTS11S), in which attention
was given to the corrective actions that follow the alarm raised by the
DSS.
From Prototype to Proving Ground
Determining and Assessing Corrective Action Strategies for
Treatment Plants Exposed to Chemical Toxins commenced with a
workshop involving operators, managers, consultants, and researchers to
brainstorm ideas for corrective action strategies to mitigate the
potential impact of toxins if they are detected by the DSS and to
identify which strategies they thought would be most effective.
Two very different facilities were used as case sites in order to
give different perspectives for the corrective actions that should be
taken for each. Neither plant had previously established corrective
action strategies for the type of toxic events being considered for the
DSS. Plant 1 is an activated sludge facility with two separate
treatment trains that provide basic primary and secondary treatment with
BOD removal only and seasonal nitrification. A second WWTP (Plant 2)
provides more advanced treatment with full biological nitrogen and
phosphorus removal followed by tertiary filtration. The research team
selected different corrective actions specific to the different
facilities.
After the workshop, researchers set up lab-scale test rigs to
simulate the two different facilities and different corrective action
strategies were tested. Following this initial work, researchers
constructed a pilot-scale facility of Plant 1. This rig was
equipped with multiple online sensors throughout the system, and
researchers simulated toxic events using hypochlorite, which is a
powerful oxidant that stays mainly in the liquid phase, and then
cadmium, which is a toxin that mainly accumulates in sludge. The team
tested these corrective action strategies:
- Sludge diversion and storage while event takes place, in order to
protect the activated sludge during the toxic event and thus enabling it
to return quickly to full treatment following the event.
- Purposely short-circuiting aeration basins by shutting off aerators
and allowing the mixed liquor to settle, while the bulk of the flow
passes over the activated sludge, hopefully preventing it from seeing
significant toxin concentrations.
- Chemically Enhanced Primary Treatment (CEPT) to remove toxins in the
primary tanks, thus preventing them from getting to the secondary
treatment stage where they would cause significant upsets to biological
treatment.
A final output from Determining and Assessing Corrective Action
Strategies for Treatment Plants Exposed to Chemical Toxins
(04CTS11s) is a DSS framework that plants can use to develop a
site-specific system to select and carry out corrective actions. The DSS
framework provides a practical basis for any facility to develop its own
DSS through consideration of the following six stages:
- Stage I: Receive Warning, either through an
automated system linked to online sensors, a pre-warning via a telephone
call, or plant operators noticing some problem on site.
- Stage II: Gather Additional Information to
determine if the contaminant is likely to cause an upset.
- Stage III: Characterize Contaminant to determine
the expected fate and type of impact of the contaminant and decide which
corrective action should be used.
- Stage IV: Take Immediate Corrective Action based on
the information from previous stages.
- Stage V: Monitor to determine the effectiveness of
the corrective action and potentially make adjustments.
- Stage VI: Take Longer Term Action (Recovery) to
bring plant back into normal operation. Also, feedback is given
from the last two stages to enable the DSS to be constantly
improved.
Wastewater treatment plants need proactive corrective action
strategies that minimize the impact of chemical, biological, and
radiological contaminants on wastewater treatment plants. This DSS
framework provides operators with a systematic decision making process
in which to respond to a range of contamination events that will
minimize the degree to which damage is imparted to the treatment process
and environment, while ensuring the utility’s ability to meet
their obligations to regulators. In addition, the pilot study data and a
broad range of background information generated by the two projects are
invaluable to any utility looking to develop their own DSS.
This research shows that it is feasible to develop support systems to
prevent upsets using real-time information collected by
sensors/analyzers subject to advanced data mining techniques to predict
the experimentally simulated toxic shock, leading to more proactive
approaches to minimize impacts and assisting in the execution of
remedial actions following the detection of an upset event.
For more information, please contact Claudio Ternieden, WERF’s
assistant research director, at cternieden@werf.org. More
information on WERF’s security-related research efforts can be
found online at www.werf.org/security.
WERF thanks Andrew Shaw, process specialist for Black &
Veatch, for contributing this story.
December 3, 2009