This paper presents some aspects of the time propagation of underground water leakage in controlled laboratory conditions using a drilled polyvinyl chloride (PVC) pipe and interpreting ground penetrating radar (GPR) images. GPR pre-processed images are interpreted for easy identification and extraction of surfaces and volumes of water leakage. Finally, the temporal evolution of a water leak is shown using 3D models based on interpretation of GPR images. Water volumes obtained using this approach can be easily observed by personnel who lack highly specialized training in the analysis of raw data. The results of this study are promising and can help develop techniques to validate non-destructive models for the identification, distribution, and prediction of water leaks in water supply systems using GPR.

INTRODUCTION

Water leakage detection and control is one of the greatest challenges in the management of water supply systems (WSSs). This paper addresses the problem of leakage in WSSs from the perspective of non-destructive methods.

The most popular non-destructive methods currently used to locate leaks in WSSs are: acoustic methods; infrared thermography; and ground penetrating radar (GPR) (Demirci et al. 2012). GPR has been shown to be an effective non-destructive tool by showing in radargram images the contrast between leaked water and the surrounding area – the contrasts being caused by dielectric differences (Charlton & Mulligan 2001; Crocco et al. 2010). We performed laboratory tests to extract features of water leakage from GPR images. Temporal evolution of water leakage was the initial feature evaluated in this work, supported by leak evolution tracing via 3D modeling.

The main idea was to produce reliable knowledge and robust data that facilitated an understanding of the phenomenon through 3D representation, thus favoring the prediction and/or detection of water leakage in the system. The aim was to increase the approximation abilities of the model, so as to improve operational and maintenance activities and favor decision-making in WSS management. This is possible because GPR image interpretation enables identifiable surfaces (with variable grade difficulty) to be obtained and this then enables objects or phenomena underground to be rebuilt in 3D models. This process facilitates the successful use of GPR data by personnel who are not highly qualified.

In this paper, water leak propagation from a buried polyvinyl chloride (PVC) pipe is addressed. This is performed using GPR images obtained under controlled laboratory conditions. Plastic material was selected because of the difficulty experienced in identifying plastic pipes when using GPR (Ayala-Cabrera et al. 2011) and the decreased visibility of surrounding objects caused by plastic materials (Ayala-Cabrera et al. 2013b).

Several authors have addressed leak detection, including Eyuboglu et al. (2003), Hasan (2012), Demirci et al. (2012), and Cataldo et al. (2014). These works are promising with respect to GPR usage in WSS leak detection. However, all of these works are based on the location and interpretation of the hyperbolae generated either in raw or pre-processed images. In this sense, we evaluate the viability of applying a pre-processing GPR image methodology that the authors term ‘agent race’ (Ayala-Cabrera et al. 2013a). With this approach, we aim to eliminate image contrast requirements, as used in Ayala-Cabrera et al. (2014), and follow the evolution of a water leak from GPR images – so that the contrast can be revealed using the information from events registered in GPR images based on the same data set. If this approach is possible, then the application of GPR in uncontrolled soil conditions and without contrast could give significantly improved images. Other interesting work shows support for this line of research: with non-destructive methods for plume dispersion of contaminants (Moradi et al. 2011) in the case of geophysical methods for delineating plumes caused by oil contamination through soil and groundwater. Pollution plume evaluation in an urban environment is also addressed in Vaudelet et al. (2011). Knowledge obtained with the methodology described in this paper may provide basic know-how for other engineering fields in such areas as fluid dispersion through subsoil, infrastructure risks associated with water leaks, and wet bulb distribution for irrigated areas.

This document has been organized as follows: the first section contains a brief introduction of the tasks performed. The general approach along with the difficulties and benefits is presented in the ‘Overall approach’ section. The third section presents a case study and survey characteristics. The approach used to interpret and classify the objects obtained by the GPR images is presented in the fourth section. The fifth section gathers the resulting interpretation and classification analysis. The paper ends with conclusions.

OVERALL APPROACH

A 3D model representation of leaks in WSSs is of great interest. For example, a 3D representation of a laboratory-simulated water leak can be observed in Tavera (2008). In this work, Kirchhoff migration and the Hilbert transform were used as pre-processing methods. Although this work is promising, it only presents the leak in a partial manner and based on obtaining the first hyperbolae. In contrast, we assess the feasibility of extracting contours, providing information beyond the first hyperbolae, and enabling more robust 3D representations that show the development of the phenomenon over time.

The most important benefits of 3D representation include a clear visualization of the studied phenomenon; support for feasibility studies of hydraulic behavior in WSSs; and, finally, support for technical WSS management – such as maintenance, rehabilitation, and general decision-making.

The difficulties associated with this study are related to the GPR images from which data were extracted. The first drawback was the absence of clear protocols for performing image interpretations when there is excessive noise in the images – such as radio-waves, mobiles, and electrical sources (Brito-Schimmel & Carreras 2010) – which cause deformation in the response from the buried objects and made the task of identification harder. This means that interpretation becomes more subjective. Consequently, algorithms were developed that enabled automatic 3D leak modeling and so avoided subjectivity.

CASE STUDY: THE LAYOUT FOR THE LABORATORY TESTS

We describe in this section the layout of the laboratory tests (Figure 1). For the set of tests performed, a plastic pipe commonly used in WSSs was buried in dry soil in a tank.
Figure 1

Survey configuration: (a) pipe burying process in the laboratory tank; (b) polypropylene plate – antenna sliding surface; (c) CS schematic configuration.

Figure 1

Survey configuration: (a) pipe burying process in the laboratory tank; (b) polypropylene plate – antenna sliding surface; (c) CS schematic configuration.

The characteristics of the buried pipe were: (a) PVC; (b) diameter of 100 mm; (c) length of 0.95 m; (d) hole drilled to simulate the leak in the central part of the pipe; and (e) two points for water input and output with connections at the ends (see Figure 1(a)). A wooden tank measuring 1.0 × 1.0 × 0.60 m was used. After the pipe was positioned, its supports were removed, and it was then covered with dry soil (before water was introduced into the system). The surface of the tank was covered with a polypropylene plate (see Figure 1(b)). Eleven paths parallel to the X-axis and another 11 paths parallel to the Y-axis were marked on this plate. These 22 paths were spaced 0.10 m apart to form the sampling grid. Three of these 22 paths were parallel to the pipe, and one was positioned in a transversal direction. We term this layout the control set (CS); each CS was composed of four slices, s, with domain {A, B, C, D}. The three parallel slices (sA–sC) were 400 mm long and the transversal slice (sD) was 700 mm long (see Figure 1(c)). The selection of these profiles crossed the pipe in longitudinal (CSsC) and transversal (CSsD) directions, both profiles intersecting azimuthally above the leak point (see Figure 1(c)). The other two profiles were placed close to the pipe and intended to keep track of the leak. A GPR antenna was placed on each line of this mesh.

The GPR equipment used in each survey was a commercial monostatic antenna with a central frequency of 1.5 GHz. The parameters of the equipment were 120 trace/s, 512 samples/trace and 20 ns/512 samples.

The total duration of the tests, corresponding to a timeframe of 35 minutes, was divided into three stages. Stage 1 started at [(t0, 00:00:00)] with the initial condition that the test was performed under the established configuration with no water in the system. Stage 2 [(t1, 00:02:10), (t2, 00:06:02)] involved filling the pipe. Times t1 and t2 correspond to a volume of water less than that required to fill the pipe. Stage 3 [(t3, 00:11:42), (t4, 00:14:02), (t5, 00:17:03), (t6, 00:21:54), (t7, 00:28:05), (t8, 00:33:06), (t9, 00:35:06)], in turn, corresponds to the water added to the system once the pipe has been filled.

INTERPRETATION AND CLASSIFICATION OF THE GPR PROFILES

The proposed approach to extract contours of the GPR images is presented in this section. The process follows these steps: (a) raw image pre-processing; (b) pre-processed image abnormality analysis and location; (c) interesting abnormality color range identification; (d) contour extraction; and (e) 3D model representation. This process is described below.

The data obtained from the GPR laboratory surveys were the main input and were visualized as radargrams consisting of the signals received by the equipment stored in a matrix (MB, raw image matrix, Figure 2(a)). This matrix was composed of m-vectors bk, (traces), representing the change in depth of the electromagnetic inspected soil properties. Arranging this matrix by columns, we obtained . Vector bk of length m corresponds to the registered data volume in each trace, depending on the characteristics of the equipment (in our case 512 data per trace). Considering the difficulties in raw data analysis caused by the poor visibility of some objects (based on their electromagnetic characteristics), a raw image pre-processing methodology is proposed in this paper that helps improve visualization of abnormalities contained in the survey images.
Figure 2

An example of the applied process: (a) raw image – MB matrix; (b) pre-processed image – Output 1; (c) normalized pre-processed image – Output 2; (d) image after range selection – Output 3; (e) contouring, group selection; and (f) contour extraction and retrieval.

Figure 2

An example of the applied process: (a) raw image – MB matrix; (b) pre-processed image – Output 1; (c) normalized pre-processed image – Output 2; (d) image after range selection – Output 3; (e) contouring, group selection; and (f) contour extraction and retrieval.

The pre-processing algorithm used was proposed by Ayala-Cabrera et al. (2013a). The algorithm is based on game theory and uses the so-called multi-agent paradigm. Specifically, the process is simulated by a race between agents, as explained below. The algorithm input in our case is the radargram resulting from the GPR prospection, consisting of the m × n matrix, MB. The GPR prospection generated consists of n traces, which were employed as parallel tracks to be walked by the n agents. The race is an endurance test for the competing agents, the prize for each agent being a movement step for each effort performed. Those efforts are based on wave amplitude value changes in each column of the matrix (Ayala-Cabrera et al. 2013a). The race takes a total time (tt), tt = tw + tr = m, tw being the warm-up time and tr the competition time. The displacements of the agents during time tr are conditioned by the trend change of the wave amplitude on the trace that is being run. The race ends when time tt has elapsed, and the race winner is the agent that has obtained the largest displacement during this time. Output 1 (Figure 2(b)) of this process consists of an m1 × n matrix, m1 being the maximum number of displacements. Columns in this matrix describe the movement of the agents related to the competition. In this work, the movements obtained by the agents are called time lines. On each time line, the time obtained in the competition by each agent is sorted by increasing values and indexed from 1 – giving equal index to equal times. These time lines are later normalized, obtaining Output 2 (Figure 2(c)) which is the matrix used in the sequel.

Output 2 results are subjected to detailed visual analysis, and compared to detect characteristics that give information regarding the water development in the CS. The zone affected by the water input into the system is identified by comparing the current time of each slice with the previous time, thus observing the appearance of new forms for the current time. In cases in which the only variation in the system is the input of water from the leak, the presence of a new form in the evaluated slices is associated with the trace left by the water input. Once the affected zone is identified in each image of Output 2, an iterative segmentation is made, based on one or more ranges ([min max]). Figure 2(d) contains the identified zone as the zone affected by the water input. As a result of this range selection, a matrix (Output 3) is obtained that has the same dimensions as Output 2: {if v ≥ max and v ≤ min, then v1 = 1, otherwise v1 = 0}, where v is the data value in Output 2, and v1 is its corresponding binary value in Output 3.

From Output 3 a contour detection was performed (Figure 2(e)) using the Moore-Neighbor tracing algorithm modified by Jacob's stopping criteria (Gonzalez et al. 2004), implemented in Matlab's bwboundaries function. From those generated contours a selection that shows spatial continuity when compared with previous time step contours was made. The selected contours were transferred to the original raw image space. Finally, to improve comprehension and visualization of the results, a fusion of the different contours (in the original space) was performed for each capture set. Fusion of the shapes obtained from GPR images after the iterative interpretation was performed using 3D models. This fusion was performed by positioning the points at their respective spatial coordinates and then plotting the end points for each group. These extreme points were connected by a Delaunay triangulation, which enabled surface visualization.

The contours obtained in this process will serve in future research to train intelligent data classification systems that can automatically detect contours. This can be achieved by using shape characteristics (such as eccentricity and orientation) and system parameters (such as pressure, water volume, and soil humidity) and the final objective is to validate or generate models that facilitate an improved understanding of the phenomenon.

The aim is to improve the visualization in subsequent extraction and classification of contours.

ANALYSIS OF RESULTS

In this section, the obtained results are presented after the GPR image interpretation and classification process. It must be mentioned that although the selected contours do not cover the total leakage area, a considerable part of this area is effectively selected. This means that the obtained results show some deformation and are not fully accurate. Nevertheless, these results give relevant information about the represented phenomenon. We also claim that, with some improvements, complete information could be automatically extracted. The obtained results of the process of interpretation and extraction of contours are shown in Figure 3.
Figure 3

Leakage area contour slices A through D.

Figure 3

Leakage area contour slices A through D.

Figure 3 shows that times corresponding to Stage 2 (t1 and t2) in the slices from A through C do not present a significant difference, which is coherent with the survey plan when considering that this is the pipe-filling stage. From Stage 3 (t3 through t9), the pipe has reached its maximum volume and starts leaking; it is just after t3 that the slices start to demark the zone being filled with leaked water. It is observed that from t3, in slices A and B, the extracted contours present a similar form. For slice C the first demarcating contour appears at t4. The difference between the extracted contours lies in the difference in zone growth and in ascending contour displacement, caused by the volume increase and rising water.

In slice D, the results corresponding to the transversal slice for the pipe over the leak spot are shown. For Stage 2, in t1 and t2, it is observed that it is possible to extract the pipe zone in the filling process, in contrast to the previous slices (slices A through C). The pipe is demarked by a hyperbola and this is observed in all the images; the contours are fully (t1 and t2) or partially (t3 through t9) enhancing the pipe (zone within the hyperbola) and leaked water (remaining zone). In Stage 3, the selected zone set grows and takes the main role in the image, enabling the water action to be seen once it is out of the pipe.

The variation of the forms in the various images shows that it is possible to obtain (to some extent) the path of the water as it leaks from the pipe. While the results are not accurate and reveal some deformation, it is also true that they can be calibrated to establish a model of leaked water for each type of breakage, and relate the leak to the pressure and flow values in the system. Similarly, a more complete study with different soil types could serve to identify the type of propagation and model the plume dispersion.

With the objective of better understanding the presented phenomenon in Figure 3, Figure 4 shows the results obtained in 3D representations to improve visualization.
Figure 4

3D model leak evolution; (a)–(i) represent times t1–t9, respectively.

Figure 4

3D model leak evolution; (a)–(i) represent times t1–t9, respectively.

Figure 4 shows the volume form caused by the leaked water through the different survey stages, making evident the difference between Stages 2 and 3. The corresponding volumes show the evolution of the leak over time in the area of interest. Form and volume variation is observed as the leakage grows, along with its development. In addition, it is possible to appreciate how the water leak tends to grow upwards. This enables determination of the tank volume that is affected by the simulated water leak, thus generating the basis to create theoretical models of dispersion under less controlled conditions. The 3D images obtained by using this approach may enable personnel who are not highly qualified in GPR to understand the phenomenon of a buried water leak.

CONCLUSIONS

From this study we can conclude that it is feasible to use non-destructive methods (GPR in our case) in water leak detection from buried pipes to reconstruct the stages of the leak and show how water is dissipated underground. Since this work shows that it is possible to obtain the wet area contours, from which more complex studies can be made, it is also possible to determine the volume of wet soil, and the volume change over time. From this perspective, we conclude that the obtained contours of this study contribute information beyond the initial hyperbolae study that some previous works rely on. This approach favors the generation of more accurate models in terms of reproduction or detection of leaks and is also a step towards a possible automatic process in the future.

In addition, the application of pre-processing to the GPR images enables the image data to reveal more information based exclusively on these data (that is to say, with no reference or contrast images) and so improves visualization. This brings us closer to an automatic detection and location of these abnormalities. The work performed has shown the feasibility of generating 3D models that are easily interpreted by personnel who are not highly qualified in the use of GPR, enhancing the understanding of the phenomenon.

ACKNOWLEDGEMENTS

Part of this work has been developed under the support of an FPI (Formación de Personal Investigador)-UPV (Universitat Politècnica de València) scholarship granted to the second author by the Programa de Ayudas de Investigación y Desarrollo (PAID) of the Universitat Politècnica de València, and the support of Fundación Carolina PhD, within its short stances scholarship program for the first author. The use of English in this paper has been revised by John Rawlins.

REFERENCES

REFERENCES
Ayala-Cabrera
D.
Herrera
M.
Izquierdo
J.
Pérez-García
R.
2011
Location of buried plastic pipes using multi-agent support based on GPR images
.
J. Appl. Geophys.
75
,
679
686
.
Ayala-Cabrera
D.
Izquierdo
J.
Montalvo
I.
Pérez-García
R.
2013a
Water supply system component evaluation from GPR radargrams using a multi-agent approach
.
Math. Comput. Model.
57
,
1927
1932
.
Ayala-Cabrera
D.
Herrera
M.
Izquierdo
J.
Ocaña-Levario
S. J.
Pérez-García
R.
2013b
GPR-based water leaks models in water distribution systems
.
Sensors
13
(
2013
),
15912
15936
.
Ayala-Cabrera
D.
Campbell
E.
Carreño-Alvarado
E. P.
Izquierdo
J.
Pérez-García
R.
2014
Water leakage evolution based on GPR interpretations
.
Procedia Engineering
89
,
304
310
.
Brito-Schimmel
P.
Carreras
C.
2010
Metodología para la prospección geofísica en arqueología: apuntes a partir de los trabajos de Iesso, Can Tacó, Molins Nous y el Goleró
.
Lucentum
29
,
9
22
.
Charlton
M. B.
Mulligan
M.
2001
Efficient detection of mains water leaks using ground penetrating radar / Subsurface and Sensing Technologies and Applications III
.
In: Proceedings of SPIE – The International Society for Optical Engineering, Volume 4491, SPIE, USA,
pp.
375
386
.
Crocco
L.
Soldovieri
F.
Millington
T.
Cassidy
N. J.
2010
Bistatic tomographic GPR imaging for incipient pipeline leakage evaluation
.
Prog. Electromagn. Res.
101
,
307
321
.
Eyuboglu
S.
Mahdi
H.
Al-Shukri
H.
Rock
L.
2003
Detection of water leaks using ground penetrating radar
. In:
Proc. of the 3rd International Conference on Applied Geophysics
,
Environmental and Engineering Geophysical Society
,
USA
, pp.
8
12
.
Gonzalez
R. C.
Woods
R. E.
Eddins
S. L.
2004
Digital Image Processing Using MATLAB
.
Pearson Prentice Hall, Upper Saddle River, NJ, USA
.
Hasan
A. E.
2012
The use of ground penetrating radar with a frequency 1 GHz to detect water leaks from pipelines
. In:
Proc. of the 16th International Water Technology Conference, Turkey
,
International Water Technology Association (IWTA)
, pp.
7
10
.
Moradi
M.
Hafizi
M. K.
Taheri
B.
Kamal
H. S.
2011
Application of geophysical methods to delineation of LNAPL contaminated plume
. In:
Proc. of the 7th IASME/WSEAS International Conference on Energy, Environment, Ecosystems and Sustainable Development, France
,
WSEAS Press, Athens, Greece
, pp.
135
139
.
Tavera
M.
2008
Aplicación de georadar para la mejora del rendimiento de una red Hídrica
.
Master thesis
,
Polytechnic University of Valencia
,
Valencia
,
Spain
.
Vaudelet
P.
Schmutz
M.
Pessel
M.
Franceschi
M.
Guérin
R.
Atteia
O.
Blondel
A.
Ngomseu
C.
Galaup
S.
Rejiba
F.
Bégassat
P.
2011
Mapping of contaminant plumes with geoelectrical methods. A case study in urban context
.
J. Appl. Geophys.
75
,
738
751
.