Leak detection in the Adelaide CBD water network
By Dr Mark Stephens, Dr Jinzhe Gong, Dr Angela Marchi, Luke Dix, Azean Wilson and Professor Martin Lambert.
First published as an Ozwater'18 Conference Paper.
With more emphasis on efficiency from regulators and increasing accountability expected by our customers, there is a growing expectation for water utilities to proactively address water network failures and gather data to strategically manage asset maintenance and renewal programs. In this context, the implementation by the South Australian Water Corporation (SA Water) of a Smart Water Network (SWN) for the Adelaide CBD is a next generation system for data collection and asset operation and management. A SWN typically consists of a network of permanently installed sensors, a data management system supported by Internet of Things (IoT) technologies and techniques for responsive data analysis and alert generation. SA Water has been commissioning its SWN for the Adelaide CBD since 1 July 2017. As part of this SWN system, 305 accelerometers have been installed for permanent acoustic monitoring and ongoing leak detection.
This paper describes the deployment of the acoustic sensors, the analysis of the acoustic data, the procedure for field response, the outcomes and impacts of the practice and the lessons learnt. The information and insights presented may be useful for water utilities considering permanent acoustic monitoring systems for their pipeline networks.
Year of case study implementation
Case study summary
The primary objective of the permanent acoustic monitoring system in the Adelaide CBD is to detect leaks on pipes (as opposed to joints or fittings), at an early stage of development, so that they can be repaired before more damaging pipe main breaks occur. A total of 305 acoustic accelerometers have been installed on existing, below ground infrastructure. An expert team has been formed to commission this acoustic system, consisting of members from SA Water [and its operating partner Allwater (SUEZ)] and the University of Adelaide. Systematic procedures for alert generation and operational response have been established. A number of leaks have been successfully detected, and issues rectified, including a significant leak that was likely to have developed into a water main break.
The specific issue
Leaks in water distribution systems add to operating treatment and pumping costs. Leaks on pipes, and sometimes at joints, can also develop into main breaks that can damage infrastructure, property and cause public disruption. If a proportion of pipe leaks can be detected and repaired at an early stage of their development, then a proportion of water main breaks, and their more disruptive consequences, can be prevented.
SA Water has taken a proactive approach to managing leaks and main breaks in the Adelaide CBD water pipe network and established a Smart Water Network (SWN). As an important component of the SWN, 305 acoustic accelerometers have been deployed throughout the Adelaide CBD water pipe network (refer to Figure 1). Figure 1 indicatively shows all sensors comprising the SWN as grey icons, including the 305 acoustic accelerometers, as well as the acoustic alerts as blue icons that have occurred in the period from the beginning of July to the end of December 2017. Approximately 50 per cent of the total length of pipe in the Adelaide CBD network can be monitored by the 305 loggers operating at the expected coverage per logger on cast iron pipe. This expected coverage is based on an anticipated average 150 metre noise logging range, and the total length of pipe in the Adelaide CBD network. The acoustic accelerometers have been located to concentrate coverage on cast iron pipes, which comprise 70-75 per cent of the CBD’s water pipes, and in areas where the disruption resulting from pipe failure is greater. New PVC pipe, as well as cathodically protected mild steel pipes, are not being monitored. The acoustic accelerometers have been installed in existing below ground valve chambers (refer to Figure 2).
The acoustic accelerometers were supplied by Von Roll, Switzerland (Ortomat-MTC system) through Detection Services (its Australian supplier) and are magnetically connected (in either a vertical or horizontal orientation) to existing valve and fire plug fittings in the chambers. The logging and the Global System for Mobile (GSM) communications unit is connected to the accelerometer with the antennae placed in a predrilled hole through the chamber wall and into the ground outside the chamber to improve communications. The sensors transmit data via GSM to a cloud-based server and data management systems. A dedicated expert team has been established for alert and event analysis as well as system operation and maintenance.
Summary of activities and implementation
The acoustic accelerometers have been initially set up to measure noise levels at 5 minute intervals between 2:00 am and 4:00 am daily and at 30 minute intervals during other times of the day. A short (9 second) sound file is also recorded by each sensor at 2:05 am each morning. The data is transmitted via the GSM network daily between 6:00 am and 7:30 am to a Von Roll cloud-based server data processor and storage, before being imported by SA Water’s integrated data management platform (ViewTM, Visenti, Singapore). This acoustic data is further processed in the customised SA Water data management platform. User-defined criteria generate meaningful alerts when potential leak noise is identified amongst background environmental noise.
Various data anomaly, trend, baseline shift and other forms of processing can be applied to create alerts including “smart” algorithms that learn to detect and identify patterns. The alert types, thresholds and other characteristics, which are applied in analysing the measured acoustic data and / or generating alerts, have been initially implemented and then iteratively refined between July and December 2017. Different types of alerts and settings have been created and assessed using normal noise activity levels around the Adelaide CBD, both in the water network and the general environment. Controlled acoustic tests with known leak characteristics have also been conducted. Table 1 below shows the alert types that are active in the SA Water customised (Visenti) data management platform as of December 2017.
The alerts are generated by analysing the minimum night noise level (minimum noise level between 2:00 am and 4:00 am daily), night noise level (taken at 5 minute intervals between 2:00 am and 4:00 am daily) and noise level readings at 30 minute intervals over each day, except during the night noise level period. These three acoustic data sets are checked by the data management platform for value anomalies, trends, baseline shifts and / or deviations from “learnt” noise patterns. For example, loggers 12 and 24 near Hindley Street (an entertainment precinct) repeatedly recorded elevated noise levels on Friday and Saturday evenings over two months, as shown in Figure 3. The use of the Smart Value Band (SVB) alert enables this regular behaviour to be quantified. The sound files recorded each day are not currently subject to alert generation processes, but are subject to frequency domain analysis, which is available in the data management platform.
Trained operators within SA Water’s Operational Control Centre (OCC) assess the alerts and associated data, including the sound file data, and escalate requests for field investigations when appropriate. Different noises from water consumption, water meters, irrigation system flows, electrical, mechanical and traffic sources can be identified by trained operators and a library of frequency spectrums associated with each noise source is being created. The use of cross-referenced data from the SWN smart customer meters has helped to confirm water consumption noise versus leak noise.
A leak correlation function within the Von Roll data management platform, which enables the source of leak noise to be located, has been successfully applied to test data. However, due to the logger density, and variable noise propagation characteristics throughout the network, this leak correlation function has not been frequently utilised. Instead, after data analysis by the trained OCC operators, a trained field leak localisation team from Allwater (the field operations alliance partner of SA Water) is deployed. Listening sticks, ground microphones and/or portable acoustic correlators are generally used in the field at night by this team to pinpoint leaks when environmental noise is lower. The team is equipped with Primayer Enigma acoustic accelerometers (see Figure 4 below). Acoustic hydrophones and the Eureka system are able to be calibrated for noise velocity along a pipe by establishing a known noise source to account for the effect of PVC pipe, or other repairs not recorded in SA Water’s assets register. Once a leak is found, repair work is scheduled with an appropriate priority.
Outcomes and measurable impacts
Twenty-seven leaks have been detected by the acoustic monitoring system within the first six months after commissioning. With the system set up for daily transmission of all acoustic data between 6:00 am and 7:30 am, the average time between the start of a leak event and an alert from it arriving in the data management platform was 16.2 hours. The majority of the leaks were on, or downstream of customer water meters, or at fire hydrants and / or stop (isolation) valves.
An example of the data from an ongoing leak, detected on logger 301 in Hobson Place downstream of a customer water meter, is shown in Figure 5 below. The three acoustic data sets: minimum night noise level, night noise level and noise level (30 minute intervals) can be seen in Figure 5 over a two-day period (4 - 5 October 2017). All noise levels decreased from a level of 18 to 5 at approximately 11:30 am on the 4 October 2017 - after the customer was informed of the possible leak and took action within the property. The noise frequency spectrums derived from the sound files recorded at 2:05 am on 4 and 5 October 2017 can also be seen in Figure 5. Distinct differences in the frequency spectrums and power densities between the two days are evident, which relate to the initial leak and then reduced leak (the latter had higher frequencies).
Some of the leaks detected in the six months between July and December 2017 were on pipes, including a leak from a circumferential crack on a 100 mm cast iron pipe main that would likely have developed into a main break. The detected leak was repaired promptly following an alert from the SWN data management platform and this avoided a main break under a busy Adelaide CBD road. Figure 6 below shows the minimum night noise levels recorded at logger 14 in Liverpool Street on the 23 and 24 July 2017, both at 2:05 am. These raised alerts in the data management platform and resulted in Allwater Operations reducing traffic and pressure before repairing the pipe on the night of the 24 July 2017. The pipe failure mechanism was a developing circumferential crack (as shown in the photo inset in Figure 6). The noise frequency spectrums derived from the sound files recorded at 2:05 am on the 22, 23 and 24 July 2017 are shown in Figure 6. Significant differences are evident in energy and frequency between the first night when there was no leak, the second night when a leak was developing and the third night, when the leak was developing further.
Sustainability and cost-efficiency
The efficiency and effectiveness of the acoustic monitoring system are expected to be enhanced over time. The in-house operators in the OCC and field response team are gaining more experience in using the information from the SWN and are developing methods to distinguish noise from potential leaks from background environmental noise. The establishment and maintenance of the OCC alert monitoring function and Allwater Operations field response teams is an ongoing requirement.
There is a specific maintenance effort for the acoustic loggers that has been quantified over the period between July and December 2017. Battery power consumption rates have been found to be higher than anticipated, in part because of GSM communications issues and repeated data transmission. Replacement of all logger batteries is underway and likely to be complete for all 305 loggers in early 2018. Changes to the use of the loggers, described later in this paper, are likely to impact battery drain rates. In addition to battery issues, logger chambers need to be periodically dewatered and the equipment generally maintained.
Lessons learnt and critical success factors
Accelerometer type acoustic loggers are externally mounted on existing fittings and are sensitive to environmental noise such as traffic, pedestrians, traffic and pedestrian lights, mechanical plant and electrical equipment. Good contact between the accelerometer and the pipe wall (via a pipe fitting) is important to avoid microphonic effects from environmental noise sources. The sensitivity of the Von Roll loggers was explicitly tested to confirm the relationship between the reported noise levels and real accelerations (see Figure 7 below). The filters in the device were also tested to check they were appropriate for the leak noise frequencies of interest.
Water usage (consumption) by customers is another source of environmental noise that is similar to leak noise. Similar physical mechanisms apply in the creation of consumption and leaks resulting in similar sounds and sustained noise. To reduce false alarms, it is important to distinguish leak noise from the other environmental noise and to reduce the effect of environmental noise when trying to identify leak noise. This requires customised alert generation algorithms and operational experience, including experience in the interpretation of the sound file information. Environmental noise varies with location, in type, magnitude and frequency; therefore subsequent data analysis requires consideration of site specific issues. Twenty-three hydrophones have been installed to focus on noise from leaks, transmitted in the water within the pipes, across parts of the Adelaide CBD with higher main break rates. These hydrophones are under initial testing.
While an evolving main break was prevented on Liverpool Street, other pipe failures, which occurred rapidly, over a period of a few minutes or less, have been observed, and were not detected early enough to enable intervention. The time over which some pipe wall blow-out and longitudinal crack failures occur has been able to be objectively determined using data collected thus far from the SWN. Data for a main break which occurred on Grenfell Street in the Adelaide CBD at 3:11 am on 21 July 2017 is shown in Figures 8 and 9 below as an example. The proportion of pipe failures which occur over a time period of days, hours or minutes will be objectively determined over the next 12 months for the Adelaide CBD using data from the SWN. Furthermore, the speed of occurrence of blow-out, longitudinal crack and circumferential crack failures will be objectively determined through examination of failed pipe sections and the data from the SWN.
Figure 8 (inset) shows the minimum night acoustic noise levels from the nearest four loggers to the Grenfell Street main break location over seven days before and one day after the main break. None of the minimum night acoustic noise levels are greater than 2 (a low relative level) and no trends in the data are apparent. The main break occurred during the two hour recording of night noise levels at 5 minute intervals. Figure 9 shows that low noise levels were recorded up to and including 3:10 am and that an increase to relative noise levels between 100 and 4000, across the nearest four loggers, occurred by 3:15 am, 4 minutes after the main break at 3:11 am. These persisted until an operator attended and shut off the flow. Figure 9 (inset) shows the blow-out failure patch on the failed section of 200 mm cast iron pipe main.
To detect more rapidly occurring pipe failures, firmware upgrades and configuration changes to the Von Roll loggers were investigated and implemented. These upgrades enable continuous monitoring for exceedance of a threshold noise level for a fixed time period and ongoing recording of noise data at 10 minute intervals, rather than the 30 minute intervals previously used. The threshold and time period implemented were based on an assessment of the typical noise patterns observed in the data from the loggers gathered over approximately three months (at 30 minute intervals every day). This assessment was important to avoid implementing a threshold that is either too insensitive to detect an event, or too sensitive and would result in many false alerts. Identifying an appropriate time base for the exceedance of the threshold was also important in avoiding too many false alerts.
The first iteration for the continuous monitoring alert has been set for a noise level above 20 that is sustained for 60 minutes (6 by 10 minute intervals). The typical night leak detection level advised by Von Roll for minimum night noise conditions is 10 and a level of 20 was considered likely to be close to the environmental noise levels at a significant number of locations. The physical accelerations associated with these noise levels can be gauged using Figure 7 above. The continuous monitoring alert has been deployed to 34 of the loggers and tested by creating noise from a 1 L/s discharge between two Gilbert Street (within the Adelaide CBD) loggers (see Figure 10 below).
The test confirmed that data is transferred and an alert is raised approximately 1 hour after the end of the 60 minute period with noise above 20. The time to generate an alert is significantly faster than the previous average alert arrival time of 16.2 hours after the event. The 34 loggers with firmware and alert upgrades are currently under test for any communications and / or battery usage issues as well as their effectiveness in alerting leaks, before all loggers are upgraded. The threshold noise level and time frame may be reduced further if false alerts are not generated, or increased if too many false alerts are generated.
Field investigation is important because it is the way in which leak locations are pinpointed. When no surface leak is present, pinpointing the location of leaks is challenging, especially in a city environment. Complex pipe topology and dampening mechanisms can rapidly attenuate acoustic signals and/or environmental noise levels can be relatively high and obscure leak noise over short time frames. Leak noise data from the SWN relating to real leaks and/or simulated leaks is being used to objectively determine noise dampening rates.
The use of a combination of leak detection techniques, for example listening sticks, ground microphones and/or acoustic correlators and asset information such as pipe connectivity and repair history, are essential to achieve high success rates. The use of portable correlators with velocity calibration functionality is particularly important in locating noise sources, where sections of unknown pipe materials have been used to complete repairs and/or used as part of new connections and have not otherwise been recorded.
Ozwater is the Australian Water Association's annual international water conference and exhibition which takes place in alternating cities each May. To find out more, visit the Ozwater website.