Determining nightline leakage and usage
Every water company has a top hit list of problem DMAs the ones with huge nightlines which elude all efforts to track down leakage, writes Charles Harris and Roger Ironmonger of Primayer. These invariably include unmeasured trade users the number one headache for leakage practitioners.Socrates from Primayer is a new, patented tool for analysing DMA nightlines, specifically designed for discriminating between areas of high use and high leakage.
Socrates is much more than a conventional data logger. Its statistical processing capability holds the key to teasing out the elusive boundary between legitimate night use and leakage on a site by site basis. It has applications in DMA demand analysis, consumption monitoring, control system fault diagnosis and in leakage reporting.
Nightlines water flows into DMAs during periods of low use by consumers are monitored extensively by the water industry to estimate leakage levels. This is carried out using data loggers or telemetry devices attached to flow meters.
Conventionally, average values (usually via accumulated pulse counts) are recorded over 15 minute intervals. This period has been chosen for convenience; it appears to give a reasonably detailed view of network operation without overloading supervisors with data. Unfortunately, new demands on network efficiency mean much more precise diagnostic tools are required.
The problem is that conventional averaging hides important details. Figure 1 shows an example on a real DMA. The minimum 15 minute average value is 5.34 litres/sec over the period 02:30 to 02:45. The actual flow pattern can be seen varying above and below the average valves; it is markedly different and, crucially, the minimum valve (an indicator of leakage) is considerable lower. The example is taken from a mixed trade/domestic DMA and it illustrates problems endemic to such sites. Varying trading and production patterns will almost always defeat conventional sampling methods. When accounting for leakage, practitioners also need to understand how domestic customers use water at night. Traditionally, it is assumed all domestic properties use the same average quantity of water every night regardless of social or climatic circumstances. A common global estimate is 1.7 litres per property per hour. This is perhaps the biggest single flaw in the system by which the industry now estimates leakage. Tests with consumption monitors show the assumption is invalid. As an example, Figure 2 shows one week¹s data derived from such a site the night to night variation approaches 50 per cent.
Socrates tackles these problems by exploiting the maximum resolution available from the flow measuring system. It collects thousands of data values per night which it processes into a smaller set of micro-values. This set (typically 500 values) exposes the crucial details of night lines which are masked by conventional averaging. This has enormous value in diagnosing discrepancies in trade use. It is also possible to analyse nightly distributions of micro-values to find estimates of domestic consumption which reflect prevailing site conditions.
On pulse generating flow meters, Socrates uses an advanced form of Pulse Interval Timing (PIT). On analogue meters it carries out rapid sampling. These methods potentially enable it to probe nightlines more thoroughly than conventional pulse counting or averaging. However, simple PIT measurements will give artificially low values. Before presenting results, Socrates processes the raw data to remove extraneous artefacts associated with the measurement system, pressure surges and hydraulic oscillations.
On pulse generating flow meters, there can be substantial variation in output pulse values unrelated to variation in the flow being measured. On some types of meter, the variation is systematic. On others it is random. Socrates uses proprietary filtering algorithms to remove these errors. Sudden hydraulic disturbances, such as pump starts, can generate surges in networks causing flows to oscillate strongly abut mean levels. The troughs of these oscillations are difficult to distinguish from genuine short term reductions in user demand. PIT measurements which do not discount such troughs will always give wrong estimates of baseline flows. The errors will be serious. To avoid errors, Socrates checks for sudden changes in pressure. Pressure change thresholds can be set by the user and if these are exceeded, Socrates assumes a transient has occurred and ignores flow data around this section of the nightline.
When searching for minimum valves, Socrates also sets restrictions on the rate at which flows are allowed to change around minimum points. This catches spikes and and other short transients which are not evident in the pressure trace.
In addition to transient hydraulic artefact some systems exhibit continuous low level hydraulic oscillation even after removing measurement artefact. If this Œcyclic¹ hydraulic artefact is not filtered, errors will be caused by confusing trough values in natural hydraulic cycles with periods of low user activity. Socrates uses autocorrelation to determine the optimum averaging time to smooth out such oscillations. This averaging time is calculated for each and every nightline; it is the minimum time over which measurements must be averaged to minimise errors. The averaging time is called the micro-period and average values calculated over such periods are called micro-values. Figure 3 shows an example of cyclic artefact.
For each nightline, Socrates searches for a minimum flow valve which satisfies pressure and flow transient restrictions and which is free of measurement and cyclic hydraulic artefact. Because of the measurement speed, it is able to probe deep into nightlines to reveal the lowest baseline flows.
Figure 4 shows results from a rural DMA. Socrates reported a 25 per cent lower nightline than 15 minute averaging on the night shown. The selected minimum point at approximately 04:35 shows a stable dip in the nightline some 30 seconds long. Socrates rejects the minima around 04:26 as transients.
In another example, a data logger on a city centre site had detected high Œleakage¹ the logger was reporting a minimum flow of 11 litres/sec. Socrates, however, revealed a true minimum of 6.3 litres/sec with an additional varying flow component superimposed. This extra use explained the high nightline and saved time consuming investigation.
On many conventional logging systems, Œfast' may not be fast enough. High speed measurements are required to detect surges. No conventional logger automatically detects transients. Additionally, no conventional logger detects and compensates for measurements system error nor for micro-periodicity.
Fast logging on pulse counting systems is subject to large quantisation errors. For example, a 3.5 litre/sec flow logged every second from a pulse-head generating one pulse/litre will store the sequence of flow values as 3, 4, 3, 4, 3. There will be no intermediate values. This is not precise enough.
Even if fast logging was suitable, the shear volume of raw data would overload network supervisors. Socrates avoids this by processing raw measurements before presenting a limited set of key facts describing the state of the distribution system. Socrates can be used in cul-de-sac type consumption monitors where flows into statistically chosen small groups of houses are measured to calculate per capita consumption.
Figure 5 shows a Socrates nightline plot from a 111 property ACORN F50 site. Clearly evident is a Œcore' flow on which intermittent Œevents' such as toilet flushes and washing machine fills are superimposed. Because these sites are small, events rarely coalesce. The Œcore¹ flow will thus be very close to, if not actually at, the leakage level of the site. Figure 6 compares the reporting of the lowest nightline flows using conventional averaging and Socrates over one week.
Conventional averaging over-estimates the minimum flow, and hence the leakage, on this site by, on average, 37 per cent. This improved accuracy clearly has a positive implications for Per Capita Consumption calculations used in integrated flow analyses of leakage. Compared with conventional monitoring, on sites of up to approximately 500 properties, Socrates routinely records minimum flows undistorted by intermittent night use. This opens the possibility of using larger consumption monitor areas. Core flow levels would still be exposed, but the larger property count would improve the statistical analysis of night use. An average night flow value, say over one hour, from a consumption monitor comprises two components: intermittent use and core flow. Socrates always penetrates to the core flow. Core flow itself comprises three parts: continuous use, plumbing losses and leakage. On a consumption monitor, core flow is arguably all waste. An example of continuous use is a tap dripping because it hasn¹t been turned fully off. A plumbing loss might be a tap dripping because of a faulty washer. These two components vary randomly and are the principal reason why consumption monitor minimum micro-values vary about mean levels. A statistical estimate of the actual leakage boundary may be derived from the distribution of the minimum micro-values over a series of nights. An example is shown in Figure 7.
The concept of core flow can be extended to DMAs. Larger DMAs, up to 2,000 properties, have been simulated by adding together random sequences of real Socrates nightlines. These studies indicate that Socrates will typically penetrate the intermittent use on a 500 property DMA to within four per cent of the actual core flow. On a 2,000 property DMA it penetrates to within 16 per cent.
Core flow on a DMA will comprise the same components as on a consumption monitor but with two extra random elements: continuous trade use and the penetration error described above.
These four components vary randomly and cause minimum nightly micro-values to vary about mean levels. A statistical estimate of the actual leakage boundary may be derived from the distribution of these minimum values over a series of nights. Figure 8 shows a typical analysis applied to a DMA comprising 1,700 properties and a harbour. It is worth noting in this example, and it is by no means unique, that conventional analysis yields ridiculous leakage values, even before trade allowances are subtracted. In contrast to these dubious speculations, the Socrates method makes no assumptions. It needs neither domestic night use, or trade night use values. The estimate is based solely on the evidence of the nightline.
New demands being made on network efficiency require much more precise tools
to monitor performance. Conventional monitoring, based on limited average
data-sets, masks important details of night flows, so making troubleshooting
difficult. In this article, we have seen practical cases where Socrates'
high resolution data has helped the leakage manager determine what is or
isn't leakage. We have also seen how simplistic measurements based solely on
fast or PIT sampled absolute minimum flow rates will be dangerously
misleading. We have also seen how tangible benefits are gained by using
Socrates on consumption meters. The errors caused by counting in
intermittent night use are eliminated. The concept of core flow has been
introduced and with it the potential for a completely default-free analysis