Application of acoustic intelligent leak detection in an urban water supply pipe network

Urban water supply and distribution systems are important infrastructures of any city. It has been reported that the percentage of water leakage from distribution systems was even higher than 50% in some cities around the world, such as Jakarta, Delhi, Adana, etc. (McIntosh 2003; Sheng & Jinghui 2014; Porwall et al. 2018; Boztaş et al. 2019). According to the ‘2016 Statistical Yearbook of Urban Construction’ of the Ministry of Housing & Urban-Rural Development of the People's Republic of China (2017), the maximum percentage of leakage in some Chinese cities were as high as 78%. Many Chinese cities have underestimated the leakage of water distribution systems. Therefore, it is still difficult for water companies to control leakage.

The nonrevenue water (NRW) attributed to pipeline leaks, which is known as physical leakage, is estimated as 30% of the total flow (Jinghui et al. 2019). Water loss in the distribution system is more than 80%, compared to the water mains (Jinghui et al. 2019). Physical leakage prevention has become a mandatory research agenda of integrated water management systems. Generally, leakage detection methods of the system were classified into two different categories, namely regional leakage identification and pinpoint location. The online hydraulic model is used for leakage detection and district metering area (DMA) is applied for leakage control (Guo et al. 2013; Mamadea et al. 2014; Wu et al. 2016). Due to the limitations of the quantity of pressure meters and the ability to find small pressure differences, the detection precision identified with the online hydraulic model is not very applicable. The DMA method usually uses clustering algorithms to analyze minimum night flow (MNF) that read from the flow meters to determine the abnormal flow. The effectiveness depends on the scale, hierarchy of partitions and accuracy of flow meters (Aumanand & Konnur 1996; Gao et al. 2005a, 2005b; Hunaidi et al. 2005; Giustolisi et al. 2008; Loureiro et al. 2014; Chen et al. 2018. Assuming that there is approximately 5% of flow error among the flow meters, there could be an error of 100 m³/h for a district with a water consumption of 2,000 m³/h. This kind of leakage-control program is often used to survey leakage of an area, which can be quickly identified by the pressure and flux meters (Zheng & Kleiner 2012; Li et al. 2014; Sadeghioon et al. 2014; Tao et al. 2016; Yu 2017). A pinpoint location method can further narrow the leakage position from a small area which was identified by the regional method. Generally, two types of listening methods are used in leakage detection systems, namely active and passive detection methods. Manual ground listening with a sound stick has been widely used for a long time with relatively high accuracy. However, this method has large weaknesses due to the attenuation of sound in the soil, pavement medium and other factors, as well as worker fatigue. Therefore, manual listening is mostly used as a passive leakage finding method when there is some evidence of possible leakage. In contrast, active leakage detection and control is gaining in popularity among water companies.

Currently, useful and feasible technology-based approaches such as easy installation, low cost, less power consumption and non-destructibility methods are needed to monitor and detect the running pipeline. Many researchers proposed acoustic sensors to locate the leakage in water storage and distribution systems. Techniques based on the monitoring of leak-related noise and vibrations are by far the most commonly adopted (Hunaidi et al. 2000; Gao et al. 2005a, 2005b; Martini et al. 2017, 2018; Yazdekhasti et al. 2018). The hydrophone sensors installed inside the pipeline and acoustic loggers placed on the valves or hydrants along the pipeline have been increasingly used to locate leakages (Muggleton & Brennan 2003; Gongtian et al. 2013; Michael et al. 2014; Alicja & Andrzej 2016; Fatih et al. 2017; Haibat & Jae-ho 2019). Detecting the range and accuracy of hydrophone sensors is better than acoustic loggers, but pipelines can be easily damaged due to the installation of hydrophone sensors. Applying an acoustic logger does not need drilling of pipelines. It is a low-cost, compact-size and non-destructible sensor (Gongtian et al. 2013; Haibat & Jae-ho 2019). Furthermore, these techniques are non-invasive and can be applied without altering the operational conditions of the network (Martini et al. 2017). An acoustic logger based on measuring noise signal may not achieve satisfactory results because of the interference of background noise (Martini et al. 2017, 2018). The correct threshold values characterizing the background noise associated with the non-leaking state is essential for their effectiveness (Martini et al. 2017, 2018; Bach et al. 2020). Measurements of the ratio between background and leaking noise is important to check initial and pre-existent leaks (Martini et al. 2017, 2018; Bach et al. 2020). The Enhance Constant Faults Alarm Rate (ECFAR) algorithm has been proposed for isolating the burst waveform (Bach et al. 2020). Due to the difficulty of detection, the strategy with multiple detection techniques has been applied to investigate a complex underground water distribution system in one community in China (Shifan et al. 2020). The absence of reliable data and mature computation methods has restricted the large application of acoustic loggers in China. Some experimental campaigns have been conducted to assess the effectiveness of leak detection and locating. However, few records can be found to verify the success of these methods in practice.

In this paper, we tested the performance of acoustic sensors to accurately locate leakages in different modes of fixed installation, mobile exploration or in a combined way. In particular, characteristics of the identification and detection accuracy were tested in different frequency and installation distances. In addition, acoustic signals related to the active leaks and the non-leaking condition were measured to identify the relation between the intensity ratio of leaking state noise (LSN) to background noise and the detection precision. Finally, two detection datasets are used to compare the detection performance of acoustic detectors and the manual listening method in the field experiment.

A schematic diagram of acoustic detectors is presented in Figure 1. The system consists of a pipeline monitoring sensor, on-site inspection platform and online monitoring operation platform. There are two types of classical work mode for acoustic detectors. When leakage noise is detected by the acoustic logger, it can be received by repeaters and sent to servers for analysis. The noise read by the logger can also be transmitted to the on-site inspection platform for remote monitoring by the inspector. In the experiment, every signal from acoustic loggers was received by the on-site inspection platform for field staff.

Trial experiments were conducted at Suzhou Water Supply Company (SZWSC) in Suzhou, China. The experiments consist of two parts: a practical experiment in Xiang Cheng waterworks and a field experiment in the residential economic zone. The experiment design is shown in Figures 2 and 3.

A schematic diagram of the acoustic detection test is presented in Figure 2. The experiments were conducted at Xiang Cheng Waterworks, Suzhou, China. The diameter and material of tested pipes is 300 mm and ductile iron respectively. Overburden depth of the pipes is about 50 cm. The pressure of the measured pipe was kept at approximately 0.30 MPa. All the acoustic sensors were installed on the pipes with valve chambers from Number Z1 to number Z8 in order to test the sensing range of the acoustic loggers. A simulated leakage in the pipe was at the point marked ‘0 m’. A simulated leak flow rate was controlled by the tap buried underground. All the simulated leakages of the above tests were set to 5 m³/h. The intensity of background noise was tested when the simulating taps were closing. At the opening status of the tap, sounds of simulated leakages were captured by every acoustic logger. Measurements were performed for both the non-leakage and the leakage conditions to monitor the noise of the environmental and active leaks, respectively. Using Alberto's methods (Martini et al. 2017, 2018), the following procedure was applied:

• Every 30 s an acquisition was started and recorded, collecting a total of six measurements in a certain condition as a data set.

• Six measurements in every point were carried out with acoustic loggers after tap closing (one acquisition every 30 s), which was regarded as non-leakage state or background noise. Average values of every data set were calculated to evaluate the background noise of every different point.

• Six measurements from every point were recorded after the tap was in the opening status. These points were regarded as a leaking state (one acquisition every 30 s). Average values of every measurement were calculated to evaluate the leaking state noise at different points.

A comparison test between acoustic loggers and manual detection is presented in Figure 3. The reddish part in Figure 3 is the acoustic loggers tested area (ALTA), where leakages of pipeline were detected by loggers. Acoustic loggers were installed on the pipes in the valve chamber of ALTA. Leakages of the pipeline shown in the blue part in Figure 3 were traced manually, which is called the manual hunting area (MHA).

The technical parameters of the experimental field location are shown in Table 1, in which the water distribution systems were constructed in January 2000. The diameter of pipeline in this district was 40–400 mm, of which pressure was kept roughly at 0.30 MPa. The overburden depth of the pipes was 20–100 cm. The material of small-bore pipelines with diameters less than 100 m was steel-plastic compound pipe (SPCP) and the others are ductile iron. The residential area and pipeline length of MHA were slightly larger and longer than ALTA.

Different parameters were applied to evaluate test performance, which is classified as follows:

• Economical benefit (EB): cost of leakage detection (CLD) divided by the detected leakage number (DLN), generally written in vector form:

  

  (1)
• Detection efficiency (DE): detected leakage number (DLN) divided by pipeline length (PL), generally written in vector form:

  

  (2)
• Detection speed (DS): the detected leakage number (DLN) divided by detection time (DT), generally written in vector form:

  

  (3)

The results of the fixed sensors from the pilot experiment are presented in Figure 4. Comparative analysis of the noise intensity before and after leakage showed that the results indicated an attenuation trend with the distance increase from the leakage point. The noise intensity captured by sensors Z4 from the simulating tap was 10% higher than the background noise, but the noise intensity by sensors Z5, Z6, Z7 and Z8 were similar to background noise. Therefore, the noise made by leakage can be captured by the acoustic logger set at 380 m away from simulated leakage (i.e. sensors Z4). The acoustic detectors should be installed on the pipes at intervals of less than 380 m apart. Depending on the calculations, the 0.6 dB/m transmission loss confirmed in a previous study (Jinghui et al. 2019) is similar to the acoustic logger's detected value. The experimental pipe fixed on the bracket in Jinghui et al. (2019) study was different to the tube buried underground in this paper. In this work, the noise attenuation due to pipe leakage and solids together reduced by about 83% in 380 m and was different to previous research where the result was 90% was in 50 m (Gongtian et al. 2013).

The alarm status of the sensor in different positions from the pilot experiment is shown in Figure 5. The red signal read from acoustic loggers represents a real leak near this sensor. The yellow and green colors illustrate the suspected signal and no signal status respectively. Based on site testing, less than 300 m from the leakage, loggers could send alerts and the intensity ratio of leaking state noise (LSN) to background noise (BN) was from 1.9 to 2.8. The logger in the valve chamber 380 m away from the leakage emitted a suspected signal, of which the intensity ratio of LSN to BN was 1.3. With the distance of 440–628 m away from the simulated leaking point, a green signal was shown on loggers, which presented the intensity ratio of LSN to BN of about 1. These results indicated that the distance of precise alarm was probably 300 m and the intensity ratio threshold was 1.9, respectively. The intensity ratio is similar to previous research at 1.82–1.86 (Miller et al. 1999). Previous study results reveal that 37–39 dB transients occurring at about 15 signals per second from leakage could be traced by acoustic sensors (Miller et al. 1999). The Acoustic Emission (AE) amplitude of our experimental environment in Xiang Cheng water treatment plant was 45 dB. Results from the current study also resembled the location of the leakages based on Average Signal Level (ASL) values in a previous study (Didem & James 2012). Moreover, in order to obtain more precise location information on leaking points, the distance for logger installation should not exceed 300 m, and the intensity of surrounding noise should not be too high.

The test results in practical applications are shown in Table 2. The pipeline length in MHA was 1.3 times longer than ALTA, but DLN in MHA was 20% lower than ALTA. Furthermore, CLD in MHA is 48,000 Chinese yuan (RMB), converted into about 6,000 yuan per unit. The expenditure of the acoustic loggers is less than the manual detection method. With the consideration of reusing acoustic loggers, the cost of leakage detection would be even lower. With the consideration of different pipeline lengths in MHA and ALTA, the detected leakage number (DLN) of the manual method was 2.7 units per kilometer, which is lower than the acoustic loggers. The DS of the manual method is worse than acoustic loggers. Therefore, acoustic loggers have an advantage in economic and speed performance compared to manual loggers. Researchers had detected 40 leakages in Xiamen University, which saved more than 2 million Chinese yuan in two years (Jiu et al. 2018). The results from the study at Xiamen University are similar to the current study.

The test results detected by two methods in real application are shown in Figure 6. The leakage detected using manual methods was mainly concentrated in small-sized pipes with a diameter less than DN150, especially those less than DN100. By contrast, leakages in various pipelines with different diameters could be detected by applying acoustic loggers. For manual detection, the signal caused by pipe leakage is captured from soil above the pipe, which reduces the sound intensity. For acoustic logger's detection, the signal from leakage is transmitted by the pipe itself, which can be detected with negligible loss by the acoustic loggers fixed in the valve chamber. With the increase of pipe diameter, the buried depth of pipes increases gradually in soil. Therefore, leakage detection by the manual methods becomes more difficult as the depth increases, this is why leakage detected by manual detection is mainly concentrated in small-sized pipes.

Leakage detection in a water distribution system is a comprehensive task that involves many technologies. Acoustic detection devices have the advantage of locating the exact position of leakage. Compared to the manual listening method, acoustic detecting devices have no fatigue like workers. It can be operated 24 hours per day in different modes of fixed installation, mobile exploration or in a combined way. According to the sensitivity of the acoustic logger, we determined that the maximum distance between the fixed loggers is 380 m. Moreover, it is necessary for the water company to construct an integrated water management system that is composed of district metering, model detection, online acoustic detection, repair order distribution, and hotline data. By sharing these data to central management systems, leakage management can be enforced.

This work was supported by the ‘Major Science and Technology Program for Water Pollution Control and Treatment’, China (Grant No. 2012ZX07403-001 and Grant No. 2017ZX07201001).

All relevant data are included in the paper or its Supplementary Information.