Urban Water Security Research Alliance
A Real-time Event Detection System
for Wastewater Source Control
a collaborative project between CSIRO and
Griffith University
Roger O'Halloran, CSIRO Australia
SEQ water supply security
•Water grid
•Desalination
•Potable recycled water (PRW) scheme
SEQ water quality research alliance:
•Supports PRW scheme
•Qld state govt, CSIRO, UQ, Griffith Uni
Background
7-barrier SEQ PRW scheme
Inter-barrier
monitoring points
CC monitoring
points within
a barrier
Our project:
•Water quality information management
Project team:
•Prof Huijun Zhao (Griffith Uni)
•Dr Roger O‟Halloran (CSIRO)
Objectives:
•Early warning system
•Treatment plant protection
•Treated water quality
Project Overview
Stage 1
•Review existing online sensing
technologies
Stage 2
•Build online WQ monitoring system
•Develop event detection maths
•Demonstrate event detection system
Stage 3
•Real-time prototype
Project Objectives
System Configuration
• Simple and robust
• Low power and small (if portable)
Analytical aspects
• Consumes little or no reagent
• Little on-going calibration or
maintenance
• Stable and reliable sensors
Requirements for online WQ analysis
WQ analysis system must
• provide long-term, continuous, real-time
water quality information
• enable operators/managers to manage
potential risks at the earliest possible
barrier/control point
• improve level of control, better safeguard
plant operation
Desired Outcomes for End-users
Typical online water quality
information collection systems
Current online WQ systems
• Perform well in the laboratory
• Few have good field performance
• None satisfactory for raw sewage
Why?
• Most use analytical methods developed
for laboratory use
• Require strictly controlled
measurement conditions
• Difficult or impossible in the field
Current online WQ systems
Traditional analysis…
• Determines amounts of a particular
substance
• Allows decisions about the system
• Does it meet regulatory requirements?
• Can it be discharged?
• Ultimate purpose of analysis is to
characterise the system
• Accurate parameter measurement not
essential, so long as precise system
characterisation is achieved
Research context: 7-barrier scheme
Inter-barrier
monitoring points
CC monitoring
points within
a barrier
Technical Challenges
• As the first control point, Barrier 1 is
critical for the PRW system
• Wastewater sources highly diversified
• Compositions/matrix are complex
• Raw sewage represents most difficult
analytical measurement environment
• Real-time quantitative detection is
practically impossible
Monitoring system for Barrier 1
Utilise sensor signals differently
• Use a number of robust sensors for
different matrix characteristics
• Measure data continuously – get time
change information
• Analyse data collectively
• Correlations between sensor signals
• Look for patterns, trends, events
Develop a real-time event detection
system
Online WQ systems - what can we do?
Our Approach
• Matrix change recognition using
multiple sensors
• Sensors must tolerate sample
matrix, be self-contained
• Each responds to one or more
physical/chemical aspects of matrix
change
• Sensor signals must be acquired
continuously and simultaneously
Our Approach
• Sample characterisation collectively
represented by the sensor signals
• “Blind man‟s buff”!
• Event recognition from time plots
• determine if system is „normal‟ or
„abnormal‟
• A significant event detected by simply
determining changes in sensing signal
• no need to accurately determine
analytical value of each parameter
Our Approach
Event detection maths involves two key
elements:
• Real-time sensor data is clustered as
„normal water‟ matrix and used as the
reference-baseline.
• A set of criteria are used to discriminate
anomalous water matrix changes (events)
from normal water matrix changes
(reference-baseline)
Multivariate Approach
Ala
rm/R
eport
ing
Sam
plin
g
•••
S1
Establishing
Baseline
Identifying
Anomalous
No
S2
Establishing
Baseline
Identifying
Anomalous
No
Sn
Establishing
Baseline
Identifying
Anomalous
No
Yes
Yes
YesCross-checking Abnormality Of
Other Sensors
Colle
ctive
Abno
rmalit
y
Analy
sis
No
No
Yes
Re-establishing
Baseline
Re-establishing
Baseline
Ala
rm/R
eport
ing
Sam
plin
g
•••
S1
Establishing
Baseline
Identifying
Anomalous
No
S2
Establishing
Baseline
Identifying
Anomalous
No
Sn
Establishing
Baseline
Identifying
Anomalous
No
Yes
Yes
YesCross-checking Abnormality Of
Other Sensors
Colle
ctive
Abno
rmalit
y
Analy
sis
No
No
Yes
Re-establishing
Baseline
Re-establishing
Baseline
• A Robust Multivariate Filter used for real-time
establishment of reference-baseline
• Multivariate Statistical Test is used to determine if the
new input data are anomalous relative to the historical
reference-baseline
Univariate Approach
Mimics process used by trained person
1.Reference baseline estimated for each
variable separately using a modified
running median
• eliminates short-term noise
2.Univariate statistical test applied to each
data point
• anomalous relative to the baseline?
Current System Configuration
6 self-contained sensing probes (temp, pH, EC, DO, ORP, turbidity)
Temporal Patterns
Sudden matrix change
Industrial Discharge Event
Event Detection ValidationT
em
p
20.0
20.5
21.0
21.5
22.0
21 Jun 2008 23 Jun 2008
12:006:00 18:00
pH
56
78
21 Jun 2008 23 Jun 2008
12:006:00 18:00
EC
05
10
15
20
21 Jun 2008 23 Jun 2008
12:006:00 18:00
Turb
050
100
150
200
250
21 Jun 2008 23 Jun 2008
12:006:00 18:00
SentinelOne-JUNE2008.Rdata
Black dots: original sensing signals; Green lines: reference-baseline from robust filter; Orange
and red symbols: alerts and event warnings, respectively
Industrial discharge event
Effect of event on effluent quality at Barrier 2
Multivariate – acid discharge eventT
em
p
24.0
24.5
25.0
30 Nov 2008 1 Dec 2008
12:00 12:006:0018:00
pH
23
45
67
89
30 Nov 2008 1 Dec 2008
12:00 12:006:0018:00
EC
0.0
0.5
1.0
1.5
2.0
2.5
3.0
30 Nov 2008 1 Dec 2008
12:00 12:006:0018:00
Turb
050
100
150
30 Nov 2008 1 Dec 2008
12:00 12:006:0018:00
SentinelOne-OCT_DEC2008.Rdata
Univariate – acid discharge event
Storm Overflow Events
Multivariate – storm overflowT
em
p
21.5
22.5
23.5
19 Nov 2008 20 Nov 2008
12:00 12:006:0018:00
pH
5.0
6.0
7.0
8.0
19 Nov 2008 20 Nov 2008
12:00 12:006:0018:00
EC
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
19 Nov 2008 20 Nov 2008
12:00 12:006:0018:00
Turb
050
100
150
200
250
300
19 Nov 2008 20 Nov 2008
12:00 12:006:0018:00
SentinelOne-OCT_DEC2008.Rdata
Univariate – storm overflow
Will enable
• Improved event response
• Safeguard barrier operation
• Better coordination between barriers
• Better source control
• Identify problem discharges
• Characterise plant capacity and
operational protocols/conditions
Real time event detection system
Conclusions
We have devised and experimentally
validated a rapid event detection principle
We have demonstrated the feasibility of
applying the event detection principle for
wastewater source control
The event detection methodology is
effective even in very noisy data
The technique has potential applications in
other water types
Ipswich Water
Research Team
SEQ Alliance and Queensland Government
Acknowledgements