Acoustic Leak Detection (ALD)

Posted on Thursday, October 29, 2020 by Sune Dupont

How Acoustic Measurements are used to Locate Water Leaks

We have all heard it, and the UN Sustainable Development Goal No. 6 highlights it — clean (drinking) water is a scarce source that we need to protect. But how are leaks in pipe networks actually discovered if the leak does not manifest itself as a new water fountain in Mrs Smith’s backyard?
Multiple techniques ranging from radioactive isotopes over interferometry to satellite based radars are used to find leaks in pipe networks. However, acoustics is one of the most applied methods owing to its simplicity.

The Physics of Leaks

When a fluid exits a pressurized pipe, acoustic noise is generated. This fact has been known for more than a hundred years and been used extensively for detecting leaks in fluid systems containing pressurized medias such as oil, gas and water. We now know, that the source of the noise is mostly due to turbulence and cavitation bobbles generated, where the fluid exits.
However, many parameters influence the noise generation such as pipe diameter, leak size and fluid pressure. Furthermore, the pipe material and its surroundings determine how the noise is propagated in the grid. The main part of the acoustic spectrum of the generated noise has frequencies in the range from 0–2 kHz, which facilitate multiple resonances like pipe circumference modes and longitudinal pipe bending modes. To sum up; this is a highly complex system, where a full mathematical description is unpractical for even a small distribution grid having a 1 km mainline and 50–100 service lines to consumers. Fortunately, a full understanding is not necessary to utilizethe generated noise as a tool for finding leaks in the grid. Knowing that pipes work like low pass filters and plastic pipes (PE and PVC) introduce remarkable more damping compared to metal-based pipes will get you far.

Leak on metal service connection just before the water meter.

Basic Leak Detection using Acoustics

In many cases, a simple listening device placed on a curb stop (valve at beginning of service line) close to the leak is enough to hear and recognize the fizzling sound of a leak. Historically, a simple stick was actually used, much like the stethoscope used by a midwife. Present day instruments are more advanced and often implement either accelerometers or hydrophones (underwater microphones). Furthermore, many systems exist, where the listening devices are permanently installed in the grid and offer constant monitoring. This makes it possible to discover evolving leaks before they become severe. Here a good coupling between the pipe and the sensor is required. Therefore, the sensor is often either equipped with a strong magnet (accelerometers) or integrated directly into the pipe (hydrophones).

Leak found by correlation method. The times t1 and t2 are results of cross-correlation between simultaneously captured noise data at sensor 1 and sensor 2.

Getting More Out of Data

To get most out of data, some systems also introduce advanced data treatment like frequency analysis and cross-correlation techniques. The former enables filtering away unwanted (known) ambient noise sources by comparing spectral components. Crosscorrelation, in principle, facilitates leak pinpointing, i.e. finding the exact location of the leak. This requires two sensors, where the leak must be located between the sensors (See figure 2). Performing a cross-correlation of simultaneously captured measurements will give the time delay of the leak noise between the sensors. If the speed of sound in the pipe is known, this delay can be converted into a distance, i.e. the location of the leak. Especially noise correlation has proven its worth during the last two decades. However, several pitfalls exist using this kind of analysis.

As stated earlier, very local conditions can affect both the noise generation and its propagation, making it hard either to make exact frequency filters or calculate the required speed of sound. Furthermore, the massive introduction of PE based pipes require existing sensors to be located closer as the generated leak noise will be damped more. There are many ways of targeting these issues; adaptive frequency filters, active speed of sound measurements, cloud-based artificial intelligence and sensor sensitivity improvement to mention a few. Often these methods require massive calculations (= higher current consumption) or better equipment (= more expensive). However, another simple approach could be to flood the grid with sensors. If sensors are placed closely throughout the grid, leak noise will not have to propagate far before detected by a sensor.

A Simple Approach to a Complicated Problem

Integrating an acoustic noise sensor into a water meter is one way of solving this task. Meters are often placed at the end of every service connection which facilitates a very good coverage of the grid. The sensor will get an inherent optimal coupling to the grid, since a water meter is an integrated part of the pipe. Furthermore, modern smart meters all have a wireless transmission of data, which the acoustic sensor can piggyback on. However, to succeed with this tactic, the acoustic sensor must be cheap such that the combined price of the water meter is acceptable to the customers. Furthermore, the acoustic sensor should have a low current consumption since water meters often need to run on a battery for more than 15 years. Our newest water meter developed here in Denmark by Kamstrup actually introduced this novel technique without compromising any of the existing features of the meter. To reduce data transmission, a single measure of the acoustic noise level at the meter is returned once a day to either a central or local system.

Testing in Real Life

Figure 3 shows a map, where every dot marks an acoustic sensor integrated into a water meter. Within a radius of 250 meters, this amounts to more than 250 sensors monitoring the grid for leaks. During a test trial period (half a year) six previously unknown leaks were found in the shown area. These were all on the service lines and none could be heard on the curb stops, where you normally would listen for leaks, i.e. the leaks were only found owing to the acoustic sensor integrated into the water meter. Meter-based noise measurements also create new challenges. The main one being that meters are often installed inside houses, where ambient noise sources like pumps and district or central heating are more likely to cause acoustic interference. This is, fortunately, a local challenge and as seen in the top panel in figure 4, the noise pattern from a pump is often very different from a leak, making it possible to distinguish most leaks from pumps based on post data analysis.

Map showing the locations of water meters equipped with an acoustic noise sensor. A red dot means that the noise level is high, and the reason could be a leak. Data is by curtesy of VandCenter Syd.

Our initial test showed that many leaks can only be detected by the nearest meter. However, in some cases, multiple meters detect the leak. This is often seen as very similar noise graphs (Figure 4 bottom panel). The similarity rises as the sensors register the same noise source and are mostly seen in metal-based pipes, where the noise can travel far, or for leaks on the mainline where the noise source is close to multiple sensors.

To sum up, acoustic measurements are a great and simple tool for leak detection. Several technologies exist and have been used with success for the last decades. The newest step forward is water meters equipped with acoustic noise sensors. This is a simple approach to a complicated problem and even though simple, it is striking how far you get simply by the power of many.

Top panel shows noise graphs from a metal pipe leak (orange), a PE pipe leak (green) and pump noise (purple). Bottom panel shows a large correlation between three installations meaning that the register the same noise source. This is mostly seen for noisy leaks and leaks on mains.

This blog post was originally submitted to Destination AARhus on 18 September 2020.


Related blog posts