Distributed fiber optic temperature sensors for leakage

detection hydraulic structures

Amir Ali Khan, Pierre Cunat, Yves Laurent Beck, Jerome I. Mars, Valeriu
Vrabie, J.P. Fabre

To cite this version:

Amir Ali Khan, Pierre Cunat, Yves Laurent Beck, Jerome I. Mars, Valeriu Vrabie, et al.. Distributed
fiber optic temperature sensors for leakage detection hydraulic structures. WCSCM 2010 - 5th World
Conference on Structural Control and Monitoring, Jul 2010, Tokyo, Japan. pp.10071. �hal-00539714�

HAL Id: hal-00539714

https://hal.science/hal-00539714
Submitted on 24 Nov 2010

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

DISTRIBUTED FIBER OPTIC TEMPERATURE SENSORS FOR LEAKAGE DETECTION IN

HYDRAULIC STRUCTURES

A. A. Khan*1, 2
P. Cunat1
Y. L. Beck1
J. I. Mars2
V. Vrabie 3
J-P. Fabre1
1
EDF, 21 avenue de l’Europe, 38000 Grenoble, France
2
GIPSA-Lab, Department Images Signal, Grenoble INP, France
3
CRESTIC, University of Reims, France
*[email protected]

Abstract
Structural health monitoring of civil engineering infrastructures like dams and dikes is a vital issue because of their harsh working
environments. It is inevitable to detect all kind of anomalous behavior in order to take preventive measures well in advance. Most of
the failures in the embankment dams are attributed to internal erosion, the movement of soil particles under the influence of
significant water flow thorough the structures. This significant flow in turn brings about a change in temperature of the structure
thus making temperature a very useful measurand, containing information about leakages. The conventional techniques for
surveillance of dikes are comprised of visual inspections and local measurements carried out punctually. However, subject to the
constraint that these structures are spread over large distances (several tens of kilometers), semi-autonomous methods have to be
employed. Distributed temperature sensors based on optical fibers present an economically viable solution in this regard. The
low-cost telecommunication fiber, serving as an intrinsic sensor, is probed to obtain temperature profiles with high spatial (1 m) and
temperature resolutions (0.01 to 0.1 °C). The raw temperature data being not immediately interpretable in terms of leakage
detection, requires subsequent intervention of analysis techniques. In this paper, we would present the analysis methods based on
advanced signal processing techniques such as source separation as well as several case studies on real experimental sites of EDF
(major dam owner in France) which have successfully allowed water leakage detection.

Introduction
Hydraulic structures like dams and dikes which continuously surround water are always at risk because of
severe environment, internal erosion, ageing, ground movements, etc. For dam owners like Electricité de
France (EDF), the surveillance of these structures assumes a very high priority. Efficient surveillance
solutions have been developed over the years for taking pre-emptive actions to avoid a possible rupture.
About 50% of the total dam failures reported are attributed to internal erosion, movement of soil particles
under the influence of a significant flow through the structure (Fell et al., 2007). Internal erosion is usually
identified by the presence of water leakages. Although, soil parameters like resistivity and permeability are
implicitly linked to internal erosion, temperature offers a more explicit link (Johansson, 1991). The
sensitivity and reliability of temperature measurements for identification of significant flows in the
embankments is well established in literature (Kappelmeyer, 1957, Armbruster et al., 1989, Fry, 2004).

The fact that dikes constitute structures spread over several kilometers together with continuous monitoring
requirements imposes the constraint of semi-autonomous systems. The distributed temperature sensors
(DTS) based on fiber optics present an efficient solution in this regard. The idea is to develop monitoring
systems capable of generating automatic alerts with centralized control in order to complement the existing
auscultation methods like visual inspections. The possibility of multiplexing a large number of sensors on a

Khan, Cunat, Beck, Mars, Vrabie and Fabre

1
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

single optical fiber and the flexibility of sensor placement are some of the unrivaled advantages of DTSs.
Moreover, their long-range measurement capability (up to 20 kms) and efficient spatial (1 m) and
temperature (0.01 to 0.1 °C) resolutions give them distinct advantage over conventional sensors.

The temperature measurement method using DTS, considered in this paper, is the so called passive method,
a natural measure of temperature. In the absence of anomalies, the measurements are driven by
phenomenon of conduction: the heat transfer is attributed to the interaction between air temperature and
temperature of water naturally present in the ground. Leakage resulting from water flow through the dike
brings an additional heat component driven by the phenomenon of convection. Thus, when convection
superposes conduction, the temperature measurements can help in leakage detection. The fiber optic, buried
at a strategic location inside the dike (e.g., at the downstream toe), can therefore intercept the temperature
changes. The change in the local thermal conditions due to leakage is captured by the fiber optic based DTS.
However, for real on site installations, the leakage signature is not the only factor influencing the acquired
temperature. Other environmental factors, like existing structures (drains), self-response of the ground in
which the fiber is buried, seasonal temperature variations, ground heterogeneities, precipitation, etc, can
equally influence the acquired data. In brief, the recorded temperature data are not immediately exploitable
in terms of leakage identification and thus require intervention of analysis methods.

This paper would focus on the methods based on advanced signal processing techniques exploiting only the
temperature acquired by the fiber optic without any a priori whatsoever. Two approaches were developed in
this regards, a medium-term analysis approach based on source separation techniques (Khan et al., 2008a)
and a daily alarm approach based on singularity detection (Khan et al., 2010). This paper aims at presenting
some case studies of application of these approaches on real instrumented sites in France. The analysis
methodologies would therefore only be briefly recalled and the interested readers can refer to the cited
references. The paper starts with a description of temperature data acquisition system using fiber optic
sensors followed by a brief exposition of the proposed leakage detection methodologies. Finally, the results
of their application on different experimental sites are demonstrated.

System Description and Data Model

Figure 1 illustrates a typical installation of fiber optic at the downstream toe of the canal, being interrogated
by a distributed temperature sensing device placed in instrumentation room. The commercial DTSs are
based on Raman backscattering principle using the optical time domain reflectometry (OTDR) (Hartog,
2000). The major components of the sensing system are: a pulsed laser source, a directional coupler and an
optical fiber cable serving as an intrinsic temperature sensing element. The interrogation principal is very
similar to that of radar. The light photons emitted by the laser source interact with the molecules of the fiber
material. Some of the photons are backscattered by the thermally induced molecular vibrations, attributed
to Raman scattering. The Raman spectrum consists of two backscattered components, at equidistant
frequencies around the incident frequency. The intensity ratio between the temperature dependent
Anti-Stokes and temperature independent Stokes components gives the temperature as a function of arrival
time at fiber input. Knowing the speed of light inside the fiber, the temperature along the entire fiber length
can be obtained with a pre-defined resolution.

The principal advantages of DTS over conventional sensors include their commercial viability, long-range
measurements, efficient spatial and temperature resolutions, ruggedness and immunity to electromagnetic
interference, etc. There exists a trade off between different parameters of DTS device like spatial and
temperature resolutions, speed of measurement, range, signal to noise ratio, etc. (Nikles, 2007). The initial
installation also requires proper calibration with respect to some reference temperature. For the current

Khan, Cunat, Beck, Mars, Vrabie and Fabre

2
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

paper, temperature data were recorded with a device offering a temperature resolution of 0.05°C, a spatial
resolution of 1 m and a range of 5 km.

A water leakage across the dike generates a thermal anomaly recordable by the fiber optic. Several
measurements over time allow following the temporal evolution of this anomaly. The acquired raw
temperature data can thus be formulated as spatio-temporal data. However, the identification of leakage
signatures on the raw temperature data is not easy due to the influence of several environmental factors.
Consequently, subsequent processing of these data is required in order to extract useful information
concerning the leakages.
canal dike structure
S
DT
tic
r op downstream toe
e
fib

leakage
)
°C
e ( x
ur e
ra t nc
Temperature signal sta
e
mp di
measured by the fiber optic : Te

Figure 1 : Schematic of thermal dike monitoring for leakage detection using distributed fiber optic sensors.

The data recorded by the fiber optic can be written as a two-dimensional signal, a function of time t and
{ }
distance x : Y = y(t, x), 1 ≤ t ≤ N t ,1 ≤ x ≤ N x , where Nx and Nt represent the fiber length and the total
number of acquisitions, respectively. In principal, significant flow of water due to leakages could be
identified by measuring the difference of temperature between that of the canal and the soil in which fiber is
buried. However, this difference could equally be a result of other factors like existing structures (drains,
tunnels, and manholes), seasonal temperature variations, solar radiation, precipitations, soil
heterogeneities, etc. Moreover, the acquired signal is strongly influenced by response of the near surface in
which fiber is buried, masking the leakage related useful information. Anomaly detection, particularly the
leakage detection requires isolation of all these factors from the raw temperature data.

The recorded signal can be modeled as a mixture of above factors. Even though, these factors can be a
non-linear function of temperature, the temperature values induced by these factors are considered without
trying to quantify the factors which produced these values. A source as function of distance along the fiber
is attributed to different factors allowing to express the recorded temperature as a weighted linear mixture
of these factors. In addition, considering the physical independence of these factors, the associated
temperature sources can be assumed independent. The temperature matrix can thus be expressed as:

Y = MF + B (1)

N ×p
where, M ∈ ℜ t represents the mixing matrix, F ∈ ℜ
p× N x
the independent sources induced by the
N ×N
factors, and B ∈ ℜ t x the noise.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

3
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

Different pre-processing techniques such as data deseasoning and normalization could be employed for
attenuating the influence of certain factors. Likewise, the temporal periods of precipitations must be
identified as they could introduce false alarms in the detection algorithms. In this regard, a criterion for
identifying such periods, based on higher order statistics, was proposed (Khan et al., 2008b).

Leakage Detection Methodologies

Two methodologies based on advanced signal processing techniques were developed by the authors (Khan
et al., 2008a and Khan et al., 2010). A brief review of these methodologies would be presented in this
section.

Medium Term Analysis : Source Separation Approach

The fiber optic allows monitoring of temperature profiles along the dike with metric resolution. An
approach based on the separation of different thermal contributions (sources) could be used. The idea would
be to find the matrices M and F in Eq. (1) given only the observation matrix Y, the recorded temperature
data. The most commonly used techniques for source separation are singular value decomposition (SVD)
(Klema et al., 1980) and independent component analysis (ICA) (Hyvärinen et al., 2000).

In the first step, SVD, a matrix filtering technique, allows decomposition of original data space into
orthogonal subspaces. The output of SVD is then treated using the independent component analysis, based
on the assumption that the sources are mutually independent. The goal is to estimate the sources and their
contributions to the mixture by maximizing a criterion of statistical independence. The algorithmic details
do not fit into the scope of this paper and the interested readers can refer to (Hyvärinen, 2000). The final
output is a leakage detection parameter, Yleakage, a function of distance and time. The algorithm is
summarized by the schematic of Figure 2.

Y Yleakages
Preprocessing Source separation

Figure 2 : Source separation based approach for leakage detection.

Daily Analysis : Dissimilarity Based Approach

The source separation approach is useful for the case of monitoring. However, for real time early warning
system, another approach was proposed by authors (Khan et al., 2010). The method is based on the
assumption that singularities such as leakages, existing structures, ground singularities, etc., show daily
temperature variations that are different from the non-singular zones. The main hypothesis is that a large
part of the dike is constituted of non-singularity zones and the method would be applicable to the sites
satisfying this condition.

The different stages of the proposed technique are summarized in Figure 3. The system analyzes data for a
given day k at all the sensing distances. The meteorological conditions are first verified to ensure
precipitation free days. A reference vector representing the 24-hour temperature variation for non-singular
zones is estimated from the data itself.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

4
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

The deviation with respect to this reference allows formulating a measure of dissimilarity with the help of
L2 norm. The resultant dissimilarity measure, dk(x), is a function of distance. A thresholding scheme based
on probabilistic modeling of this measure and a constant false alarm rate is then applied. This formulation
allows to obtain an adaptable threshold for each analysis day with a fixed probability of false alarm. The
output is thus a detector of singularities as a function of distance along the fiber for each day. Subsequent
adaptation of the system can allow the sub-daily analysis as well. The proposed detection system has the
capability to serve as an early warning system, identifying the anomalies well before they are observable
through visual inspections.

Yi Dissimilarity
Data selection HOS Measure

d i(x)
dth i(x) PFa
Thresholding
i = day

Figure 3 : Daily alarm system based approach for leakage detection.

Results and Discussion

Over the last 10 years, EDF has developed expertise from optical fiber sensor installation to data
interpretation for structural health monitoring of dikes. The application of the leakage detection
methodologies on the real temperature data acquired at two experimental sites of EDF would be presented
in this paper.

Experimental Site # 1
The first experimental site is located in the south of France with a Mediterranean climate. The fiber optic
cable is buried at the downstream toe of the dike; bordering the feed canal at a depth of 1 m. Figure 4 shows
the installation configuration, where 2.2 km of fiber circumvents two drains, D1 and D2. The geological
conditions at the site imply that the fiber is at two different elevation levels. The raw temperature data
shown in Figure 5 as a spatio-temporal function depict seasonal temperature variations and the two drains.

Canal
Instrumentation
room Zone 1 Zone 2
DTS from 0.1km to 1.25km from 1.25km to 2.2km

Optical fiber cable

0.
1
km

Drain D1 Drain D2

Figure 4 : Illustration of fiber optic installation at experimental site # 1.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

5
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

Figure 5 : Raw temperature data acquired at experimental site # 1 over 2 kms and 6-months in 2005.

In order to test the leakage detection system, artificial percolation type leakages were used at this site, their
generation not fitting into the scope of this paper. These leakages, controlled in terms of their flow rate,
localization and duration, are described in Table 1.
Table 1. Structures and artificial leakages at experimental site # 1.

In the first instance, the source separation approach results would be presented. The preprocessing consists
of normalization and precipitation identification using the higher order statistics criterion. The output of the
detection scheme, Yleakage, is shown in Figure 6. A zoom in the vicinity of the artificial leakages, L1, L2 and
L3, shows an efficient detection and localization of the artificial leakages. The temporal evolution of these
leakages can also be followed with this result. L1 is detected as being the most energetic, followed by L2,
whereas L3 is detected with the least energy due to its low flow rate as well as its initiation during the night
when the water and air temperature difference is not significant. In order to localize the leakages, the result
can be projected on the distance axis, yproj(x). One such projection during october, 2005 (Figure 7) allows
detection of real leakages, localized over several meters, just before and after the second drain, D2. The
periodic analysis done for the data acquired in october 2007 for the same site allowed detection of these real
leakages as well.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

6
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

Figure 6 : Source separation based detection results for experimental site # 1 for year 2005.

Figure 7 : Projection of source separation based detection results for experimental site # 1 with real leakages in
october, 2005.
The daily analysis approach also allows the detection of the artificial as well the real leakages. The results
on the artificial and real leakages of 2005 are not presented here in order to avoid repetition. However, in
order to highlight the significance of daily alarm system, the detection results for 7 days in the year 2007 are
shown in Figure 8, whereby the drains and the real leakages are identified as anomalies. The detection
parameter was normalized here with respect to the maximum for each day. This kind of graphic can be very
useful to follow the evolution of different anomalies.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

7
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

Figure 8 : Detection results for daily alarm based analysis system for experimental site # 1 over 7 days in 2007.

Experimental Site # 2
The second experimental site considered here is located in the north of France, a continental climate. The
fiber optic cable is buried at the intersection of the embankment slope and surface of drainage canal.
Periodic visual inspections have shown the presence of leakages over a large part of the dike. The detection
results using the source separation approach are shown in Figure 9 in terms of the projection yproj(x) of the
detection parameter during the months of april and july, 2007. Different anomalies were observed for this
experimental site, which corroborate quite well with the visual inspection results. Moreover, the leakages
show evolution over the course of time. This can be observed by the positions marked with arrows where
the anomalies present in april (Figure 9a) disappear in july (Figure 9b). This is a peculiarity of the site,
where the main canal exhibits lots of seasonal variations in terms of its flow rate, resulting in the fact that
certain leakages appear and disappear over the course of time.

(a) April, 2007 (b) July, 2007

Figure 9 : Projection of source separation based detection results for experimental site # 2.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

8
5th World Conference on Structural Control and Monitoring 5WCSCM-10071

Conclusion
Health monitoring of dams and dikes is primordial in order to avoid the failure of these structures. Internal
erosion, being a major cause of ruptures in dams and dikes, alters the temperature field of the structure. The
fiber optic based distributed temperature sensors provide a semi-autonomous surveillance solution,
monitoring while ensuring quasi-visual inspection along the entire dike length. The raw temperature data
acquired through these sensors must be analyzed for identification of leakage related information. Two
signal processing based approaches, a medium term approach and a daily alert system based approach, were
shown to be very efficient in detection of leakages. The proposed methods were validated on two different
sites of EDF, equipped with fiber optic temperature sensors along the dike, whereby real leakages were
detected. The distributed fiber optic temperature sensor based detection is an early warning system owing to
the fact that it is automatic and allows measuring the anomalies well before they are physically visible.

References
Armbruster, H., Brauns, J., Mazur, W., and Merkler, G. P. (1989). “Effect of leaks in dams and trials to detect leakages by
geophysical means.” Lecture Notes in Earth Sciences: Detection of subsurface flow phenomena, M. G. etal., ed.,
Springer-Verlag, Berlin Heidelberg.
Fell, R. and Fry, J. J. (2007). Internal Erosion Of Dams And Their Foundations. Taylor and Francis.
Fry, J. J. (2004). “Détection de fuite sur les digues par acquisition de profils de température le long d’une fibre optique.” Sécurité
des digues fluviales et de navigation, Actes de colloque du CFGB, Orleans, France. (in French).
Hartog, A. H. (2000), “Distributed fiber-optic temperature sensors: principles and applications,” Optical Fiber Sensor Technology,
K. T. Grattan and B. T. Meggitt, eds., Kluwer, 241–301.
Hyvärinen, A., Karhunen, J., and Oja, E. (2001). Independent Component Analysis. Wiley, New York.
Johansson, S. (1991). “Localization and quantification of water leakage in ageing embankment dams by regular temperature
measurements.” Proc. ICOLD, 17th Congress, Vienna.
Kappelmeyer, O. (1957). “The use of near surface temperature measurements for discovering anomalies due to causes at depths.”
Geophysical Prospecting, 3, 239–258.
Khan, A. A., Vrabie, V., Mars, J., Girard, A., and d’Urso, G. (2008a). “A source separation technique for processing of
thermometric data from fiber-optic DTS measurements for water leakage identification in dikes.” IEEE Sensors Journal, 8,
1118–1129.
Khan, A. A., Vrabie, V., Mars, J. I., and Girard, A. (2008b). “A least square approach for bidimensional source separation using
higher order statistics criteria.” EUSIPCO, Lausanne, Switzerland.
Khan, A. A., Vrabie, V., Mars, J., Girard, A., and d’Urso, G. (2010). “Automatic monitoring system for singularities detection in
dikes by dts data measurement.” IEEE Transactions on Instrumentation and Measurement.
Klema, V. C. and Laub, A. J. (1980). “The singular value decomposition: its computation and some applications.” IEEE. Trans on
Auto. Control, 25(2), 164–176.
Nikles, M. (2007). “Fibre optic distributed scattering sensing system: Perspectives and challenges for high performance
applications.” Proc. Third European Workshop on Optical Fibre Sensors, Napoli, Italy.

Khan, Cunat, Beck, Mars, Vrabie and Fabre

9