Permanent acoustic logging: a technology whose time has come?

Permanent acoustic logging has matured as a leakage detection technology over the last decade and, as such, has been rolled out by a number of water companies during the last AMP cycle. While different hardware manufacturers have taken markedly different approaches to the data being collected, the basics of the technology itself are the same. The essential property of all acoustic loggers is that they ‘listen’ to the sounds being produced on a network with the aim of providing useful localisations of possible leaks. They are generally designed to make recordings at set intervals, most often at night when network activity is lowest and pressure is highest, with analysis of the sounds ‘heard’ providing insight into what might be happening on a network. This blog will discuss some notable aspects of the available hardware as well as the methods of data collection and processing currently seen in the market and what might be expected in the future.

Hardware

In terms of what hardware is available, an early distinction to draw is the difference between hydrophones and accelerometer-type devices. Both have their pros and cons and the optimal use of each very much depends on the characteristics of the network they are being used to cover. The key difference between the two is that, where accelerometer type devices can be attached magnetically to a valve or fitting, hydrophones must be in contact with the water inside the pipe. The result of this is that hydrophones may be slightly better at picking up the sounds being carried inside a pipe and that more accelerometer units are necessary to cover the same area of network. However, this is to be balanced against the relative ease of attaching an accelerometer.

The first observation then, among the deciding factors for potential buyers of the technology, is that the number of fittings available on a network matters. In areas where fittings are not abundant, hydrophones might present an advantage in terms of network coverage, where, in other areas, the convenience of an easily attached and non-intrusive device might take precedence.

In either case, the collected data is transmitted and collated in different ways too. In recent years, there has been a steady move away from using ‘repeaters’ as a method of data transmission as technology advances. Repeaters have been recognised as a potential point of failure in the chain of data transmission and so manufacturers are increasingly opting for more reliable methods, utilising technologies including 3g/4g, LTE-M and NBIoT. Most notably in areas that have been prone to comms difficulties in the past, these newer technologies have the potential to significantly reduce the run-time of leaks.

Accelerometers have been the most commonly used devices by the clients with whom SME Water has been working of late and so this blog will focus primarily on these from here on out.

The newest generation of loggers are compact, integrated devices that are very capable of reducing both awareness time and time-to-find for those carrying out leakage detection activities. Traditionally, these loggers offer a battery life in the region of three to five years, depending on how they are configured and how much data is required of them. As battery life continues to improve thanks to a range of technological advances, we are starting to see notably longer lifespans being achieved. Significantly impacted by pipe materials and potential ambient noise sources, it is important for those intending to use permanent acoustic loggers to consider their deployment strategy carefully. For example, plastic pipes do not transmit sound waves as effectively as metallic pipes and so loggers on plastic networks must be deployed closer together to achieve the same level of network coverage. Similarly, nearby electrical installations such as streetlights will produce distinctive interference in acoustic data at 50Hz intervals, another factor worth considering when deploying loggers. An example of electrical interference is presented below:

Figure 1. A Fourier transform showing electrical noise

Data processing

In SME Water’s experience there are two methods of processing the data that is generated by acoustic loggers. These can be broadly characterised as utilising either edge-analytics or centralised data processing. As an emerging term, “edge analytics” refers to the analysis of data from some non-central point in a system. For example, loggers capable of running an analysis of the sound data they sample each night and assigning themselves a status based on the result. This reduces the amount of data being sent to a central point as relatively cumbersome sound files are only transmitted when the logger indicates there could be a leak within its range. In addition to improving the efficiency of a system that could be receiving data from tens of thousands of loggers at a time, it also has the notable benefit of extending the battery life of the logger itself.

The other approach is for the loggers to provide their data to a centralised point for processing which, as noted above, can be quite a feat. One potential advantage of this approach is that the data from loggers in close proximity can be automatically processed with reference to each other by a central system, while a downside might be that sending sound files to a central point on a nightly basis can put the batteries of such devices at a disadvantage when compared to those that need only transmit sound files when they have entered ‘alarm’.

Without the necessary data to take a view on which approach offers superior benefits it’s worth stating that, over the coming years, it is reasonable to suggest that further advances in both battery life and processing power could potentially swing the pendulum in favour of a centralised data processing model.

Network sampling

We’ve put the cart before the horse a little by discussing processing of the data before we discuss the various approaches to actually collecting the data but the distinction between edge analytics and central processing is useful to understand before discussing why various manufacturers might choose to sample differently.

For example: one widely used model will wake up several times during the night when the network is quiet. During the first scheduled sampling, a logger will measure the sounds transmitted by the pipe to pick out characteristics that have been associated with possible leaks. If a logger indicates that a leak might be present, it powers down and then wakes up again a little later in the same night to check if it can still detect the same characteristics. If on both occasions it determines the sound qualifies as a possible leak it will collect and send four bits of data for further analysis, two sound metrics and an alarm status (which are sent every night) as well as the sound file that triggered the alarm. This is a good example of edge analytics being carried out to minimise the amount of processing being carried out in a central location. An interesting case study in the proper use of edge analytics is provided by Liverpool City Centre, where the timings at which loggers were configured to take their measurements had to be changed to take account of the city’s busy nightlife. Anecdotally, the acoustic loggers were sensitive enough to pick up the conversations being held at loud decibels in the street above and so, to prevent excess interference, had their determination time moved back to a quieter point in the night.

In contrast, another manufacturer’s devices will ‘listen’ to the sound being transmitted by a pipe five times over the course of a day. The data is fed back to a central point for analysis by an AI that has been trained to detect the characteristic qualities of a leak noise and to determine if the same tell-tale sound is present in all five recordings. Similarly, another group of devices listen for persistent noise indicative of a leak over the course of several days. The data is fed back daily to a central point for processing and analysis, with similarities between the sounds being recorded by adjacent loggers being noted as part of the analysis. Using this method the device maker builds up confidence that the sound that is being detected is likely to be leakage.

Conclusions

Presented above are a few of the key areas of consideration in the field of permanent acoustic logging. Though far from an exhaustive examination of the subject we hope that it has been informative.

As discussed, in the future we believe we can expect to see the large-scale deployments of acoustic logging equipment providing a huge amount of data for analysis and interpretation, with the potential for wide-ranging benefits in the water industry. Thanks to the economies of scale, we can also expect to see unit costs drop and, as communication technologies continue to advance, cheaper data transmission making for an increasingly effective solution for water companies striving to meet their 2050 leakage target.

SME Water has been assisting in the creation of processes to support the effective use of acoustic loggers in recent months as well as investigating some innovative uses of the data being collected. In conjunction with our Paradigm product we have been looking at how best to pick areas for leakage campaigns so as to get the best return on investment and avoid chasing customer demand. As such, investigating what can be learned about customer demand from acoustic data has been a recent area of focus, as has developing an understanding of ways in which acoustic data can best be interpreted.