| Product(s): | WaterSight |
| Version(s): | 10.00. |
| Area: | Documentation |
Background
WaterSight has the ability to automatically generate alerts for anomalous situations that occur in the network such as bursts, leaks, fires, flushing, changing zones boundaries, tank overflows, PRV malfunction, amongst others. The main goal is to support the water utility in the early identification of anomalous events, namely by improving awareness times and by supporting the decrease of response times, and therefore contributing to savings in terms of water that is being lost and improving the quality of service provided to the customer.
Some examples of events that can contribute to the non revenue water levels, affect the quality of service provided to the customer as well as the useful life of the assets are the leaks and bursts within the network. Some are visible, other can be invisible (infiltrate under the soil). Some correspond to big flows, other to small flows. In order to be able to adequately configure triggers for these types of occurrences, it is important to first describe the various types of bursts and leaks that can occur within the network.
Types of Bursts
Leakage can come from transmission and distribution mains – pipes, junctions or valves (usually have medium to high flow rates and short to medium runtimes) and from service connections (sometimes referred to as the weak points of water supply networks, because their joints and fittings exhibit high failure rates). Leaks on service connections are difficult to detect due to their comparatively low flow rates and thus often have long runtimes.
In terms of leak flow and run time, generally bursts can be grouped in three different types (Farley, M., Leakage Management and Control. WHO, 2001):
- Reported bursts: Visible and usually quickly reported by the public or observed by water utility staff. They have a short awareness time. Usually correspond to large bursts, with big flow rate and short durations.
- Unreported bursts: Commonly occur underground and are not visible at the surface. They are usually discovered during leak detection surveys and often have a long awareness time. Can correspond to small to medium bursts, from low to medium flow rates and from medium to long durations.
- Background leakage: An accumulation of very small leaks that are difficult and not cost effective to detect and repair individually. Therefore it is assumed that many background leaks are never detected and repaired but leak until the defective part is eventually replaced. Background leaks often cause a major share of real water losses due to their great number and their long runtimes.
The Figure below presents the different types of bursts and the typical leakage rate and runtime associated.
Figure 1. Relation between the different types of bursts and their leakage rate and runtime (from Guidelines for water loss reduction)
Configuring Bursts Alerts (based on a flow increase)
Alerts can be configured by the admin user in the Alerts Administration page and once generated by the system they become available in the Alert list page.
Alerts can be generated based on two methods: i) pattern based alerts, where SCADA or other telemetry data (processed) is compared with the expected behavior for that sensor or zone and ii) absolute based alerts which are evaluated by directly comparing real time data (processed) with the user defined threshold. For more details about this, please go to the Alerts administration help page.
The aim of this article is to support the user in the definition of the most adequate triggers in order to "catch" most of the system bursts and simultaneously trying to minimize the number of false positives. In order to achieve this, the user should try to replicate in the alerts trigger configurations the type of bursts represented in Figure above.
The examples explained below use the sensor or zone pattern to generate the alarm. This means that the alarm only gets generated if the real values go outside the pattern confidence bands (above percentile 95), and also based on a minimum duration and threshold value defined by the user. These type of alarms (based on the pattern) should only be assigned to flow or zone sensors that really have a typical daily pattern.
Large Burst Alerts
Large bursts usually correspond to big flow rates and short durations, as they are quickly reported by the public or the utility staff.
WaterSight can help the utility further reducing awareness times for large bursts. Alerts for these events should be configured according with the following:
- Type: Flow
- High/Low: High
- Absolute/Pattern: Pattern
- Time Series: 15 minutes
- Pattern History: Any (shorter periods work better for zones that have frequent zone operational changes or that have significant influence on seasonality; longer periods work better for zones that are stable, with few changes in the boundaries configuration)
- Value/Threshold: To decrease generation of false alerts (false positives) this value should depend on the typical sensor or zone flow. Some examples below that can serve as guidance only if there is no previous experience or other information available.
|
Average Flow (P50) |
Value/Threshold |
|
> 120 l/s |
> 30 l/s |
|
30 – 120 l/s |
20 - 30 l/s |
|
10 – 30 l/s |
10 - 20 l/s |
|
< 10 l/s |
< 10 l/s |
Note: These values are based on previous experience and should be redefined as the user starts to validate the alerts generated by the system.
- Min duration: 0.5 hours or 1 hour (recommended)
Note: Configuring a large burst alert with bigger durations may increase awareness times for these type of occurrences. Configuring a large burst with smaller threshold values can generate significant false positives.
Figure 2. Above - Large burst event represented in the sensor detail graph (15 minutes time series). Below - Large burst configuration example (values in m3/h).
Medium Burst Alerts
Medium bursts commonly occur underground and are not visible at the surface and have medium flow rates and durations.
WaterSight can help the utility further reducing awareness times for medium bursts. Alerts for these events should be configured according with the following:
- Type: Flow
- High/Low: High
- Absolute/Pattern: Pattern
- Time Series: 15 minutes
- Pattern History: Any (shorter periods work better for zones that have frequent zone operational changes or that have significant influence on seasonality; longer periods work better for zones that are stable, with few changes in the boundaries configuration)
- Value/Threshold: To decrease generation of false alerts (false positives) this value should depend on the typical sensor or zone flow. Some examples below that can serve as guidance only if there is no previous experience or other information available.
|
Average Flow (P50) |
Value/Threshold |
|
> 120 l/s |
> 20 l/s |
|
30 – 120 l/s |
10 – 20 l/s |
|
10 – 30 l/s |
5 – 10 l/s |
|
< 10 l/s |
< 5 l/s |
Note: These values are based on previous experience and should be redefined as the user starts to validate the alerts generated by the system.
- Min duration: from 2 hours to 12 hours (recommended)
Note: Configuring a medium burst alert with shorter durations or smaller threshold values can generate significant false positives.
Figure 3. Above - Medium burst event represented in the sensor detail graph (15 minutes time series). Below - Medium burst configuration example (values in m3/h).
Small Burst and Leak Alerts
Small bursts and leaks commonly occur underground and are not visible at the surface and have low flow rates and long durations. Usually are associated with small bursts or leaks, which increase in flow is usually difficult to detect by just looking at the instantaneous or 15 minutes values. For these cases, the minimum nightly flow (MNF) should be monitored and alerts should be triggered when there is an increase on these values.
WaterSight can help the utility further reducing awareness for small bursts and leaks. Alerts for these events should be configured according with the following:
- Type: Flow
- High/Low: High
- Absolute/Pattern: Pattern
- Time Series: Daily Minimum
- Pattern History: Any (shorter periods work better for zones that have frequent zone operational changes or that have significant influence on seasonality; longer periods work better for zones that are stable, with few changes in the boundaries configuration)
- Value/Threshold: To decrease generation of false alerts (false positives) this value should depend on the typical sensor or zone flow. Some examples below that can serve as guidance only if there is no previous experience or other information available.
|
Average Flow (P50) |
Value/Threshold |
|
> 120 l/s |
> 6 l/s |
|
30 – 120 l/s |
3 – 6 l/s |
|
10 - 30 l/s |
1.5 - 3 l/s |
|
< 10 l/s |
< 1.5 l/s |
Note: These values are based on previous experience and should be redefined as the user starts to validate the alerts generated by the system.
- Min duration: from 2 days to 6 days (recommended)
Note: Configuring a small burst alert with shorter durations or smaller threshold values can generate significant false positives.
Figure 4. Above - Small burst event represented in the sensor detail graph (daily minimum time series). Below - Small burst configuration example (values in m3/h)
Although large bursts are usually associated with big flows, the small bursts and leaks are usually responsible for most of the volume that is lost due to the big awareness, location and repair times.
Configuring Bursts Alerts (based on a pressure decrease)
Bursts alerts can also be configured and triggered based on a decrease in pressure (in relation to a pattern) and assigned to the individual sensor. However in some cases it is more likely that a burst will cause a more significant increase in the flow than a significant decrease in the pressure, as several systems are running with oversized pipes (pressure change/head loss also depends on pipes diameters).
Zone Alerts vs Sensor alerts
When configuring bursts alerts based on a flow increase, it is usually more accurate to assign the alert to a zone than to the individual sensor.
Zones correspond to areas of the system surrounded by closed valves and where all inflows and outflows (if exist) must be measured. Zones can include for example DMA (District Metered Areas), pressure zones or operational zones. In WaterSight the zone flow is already the final balance between all inflows, outflows and zone storage (configured in the Zones configuration page). Therefore the calculated zone flow represents all the consumption occurring inside that zone plus the water losses. If an alert is triggered and is assigned to a zone, this means that a potential anomaly is occurring inside the zone and can be an indication of a burst or leak.
Alerts assigned to individual sensors can be less reliable, as summarized below:
- If there is an anomalous flow increase associated with the sensor, this means that a potential anomaly may be occurring any zone downstream that sensor but it is not possible to identify the exact zone;
- If the zone is supplied by more than one inflow you might get an alert of significant flow increase in one inflow and a flow decrease in the other. In this case the total inflow to the zone remains the same, there is no anomaly, however you might still get an alert if it is assigned to individual sensors and not to the final zones.
- Zones have usually a much more reliable and consistent patterns than the individual sensors. Therefore if alerts are assigned to sensors and triggered based on the patterns, there is a bigger chance for generating false positives or even to not trigger any alarms for real bursts or leaks happening in the network;
Example below
- If there is an anomalous flow increase in sensor number 1, it is not possible to know if the increase is at the DMA A level or in other DMA or even due to some tanks overflows.
- Pattern for individual sensor number 1 includes the distribution to DMA A plus pumping to the two downstream tanks, so that is possible that sensor 1 will not have a very clear and well defined pattern (spikes are expected to occur at different times of the day). We can have a much more reliable pattern where we exclude the interferences of the tanks and pumps, if we rather consider the zone flow of DMA A, as it will be already the final balance between all inflows and outflows (in this case 1 – 2 – 4)
Assigning alerts to individual sensors are usually useful to track sensors failures, flat readings or malfunction. Or to trigger alerts based on tank levels or inadequate pressures (more information here about configuring absolute based alerts).
Why not using Absolute alerts?
Defining absolute alerts based on an increase in flow or a decrease in pressure to notify about burst and leaks, may only be viable in some conditions when a specific zone does not have a reliable pattern. Whenever patterns exist, the pattern should be used to trigger these alerts. Disadvantages of using absolute alerts to trigger bursts and leaks events are related with the generation of false positives (alerts that are generated and shouldn't have been) or related with missing important real events (alerts not triggered for real events). Below some explanations:
- Flow and pressure significantly changes during the night and the day. A specific absolute defined value may trigger an alarm during the night (when flow is usually lower) but not during the day (when flow is usually higher), or the other way around (may not trigger an alarm during the night because flow is too low and the absolute value is not reached) - Figure 5 and 6;
- Absolut alerts tend to generate several repeated alerts for the same event. While pattern based alerts only (automatically) finish when values return to normal (to the pattern), for absolute based alerts the event automatically finishes whenever the absolute threshold value defined by the user is not met anymore. Then, if the real value goes again below or above the user defined absolute threshold, another alert is generated (whenever there are bursts or leaks, the real flow and pressure values change along the day together with the consumption and pressure in the network) - Figure 7 and 8.
Figure 5. Example of an anomaly not detected because its maximum value was below the absolute value defined in the alert rule.
Figure 6 - Gradually increase in the time series values along time that would not be detectable using and absolute alert type.
Figure 7. 11 events were generated for the absolute alert (assuming an absolute threshold of 200)
Figure 8. Only one event was generated for the pattern alert. The event starts when values start to go outside the pattern bands and it automatically finishes once returns to normal (to the patterns again).
For more information about when/in which conditions absolute alerts should be used, please see this article.
Configuring Alerts Based on Volume
Based on an absolute volume
The user can also configure alerts based on a volume that passes through the sensor or zone. For example if the user wants to trigger an alarm whenever the volume per day passing in that sensor is above 5 000 m3 , the configurations should be the following:
- Type: Flow
- High/Low: High
- Absolute/Pattern: Absolute
- Time Series: Daily Average
- Value/Threshold: 58 l/s (= 5000 m3 divided by 24h and converted to l/s)
- Min. Duration: 1 day
In this case the alert will only be generated if the average daily flow (real) is more than 58 l/s.
Note: The Value/Threshold units are those defined in the Digital Twin settings, under Administration.
Based on a volume lost
The user can also configure alerts based on a volume lost. For example if the user wants to trigger an alarm whenever the volume lost per day is above 5 000 m3, the configurations should be the following:
- Type: Flow
- High/Low: High
- Absolute/Pattern: Pattern
- Time Series: Daily Average
- Value/Threshold: 58 l/s (= 5000 m3 divided by 24h and converted to l/s).
- Min. Duration: 1 day
In this case the alert will only be generated if the average daily flow (real) is 58 l/s above the normal expected daily flow (P50). The difference between the real daily flow and the expected daily flow (58 l/s) corresponds to the burst flow
Note: The Value/Threshold units are those defined in the Digital Twin settings, under Administration.
Geolocating leaks
n case the system is well sectorized in small DMAs (district metered areas), flow measurements exist at all zones inputs and outputs and there are several pressure measurements available inside the DMA, then all those measurements can be used to better geolocate the potential leak inside the zone. In order to achieve this please:
- Configure flow increase alerts for each zone, based on pattern type alarms (already described above);
- Configure pressure decrease alerts for each pressure sensor, based on pattern type alarms (already described above);
Whenever a flow alert is triggered for a zone, this can also generate a pressure decrease somewhere inside the zone. So if you receive a flow alert followed by a pressure decrease alert for the same zone, then it means that the leak/burst is potentially located near that specific pressure sensor. Please note that leaks geolocation effectiveness based on pressure sensors monitoring can be significantly affected in the the following cases:
- oversized networks (in those cases pressure decreases are very small and therefore very difficult to notice/detect);
- lack of pressure sensors inside the zone;
Also please note that this methodology can be used to detect new bursts or leaks that appear in the system, but not to geolocate existent leaks that already exist in the system for a long period (as WaterSigth generates alerts based on differences between real values and patterns, being those estimated based on historical trends that can include leakage).
See Also
OpenFlows WaterSight TechNotes and FAQ's
WaterSight Learning Resources Guide