See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/288063712

Thermal deformation of concrete dams:

Justification, clarification and improvement of
statistical analysis

Article · January 2012

DOI: 10.1680/dams.57999.145

CITATIONS READS

0 75

6 authors, including:

Matthieu Briffaut Frédéric Dufour

University Joseph Fourier - Grenoble 1 Grenoble Institute of Technology
40 PUBLICATIONS 155 CITATIONS 109 PUBLICATIONS 999 CITATIONS

SEE PROFILE SEE PROFILE

Alexandre Simon Jean-Paul Fabre

Électricité de France (EDF) Électricité de France (EDF)
5 PUBLICATIONS 21 CITATIONS 9 PUBLICATIONS 17 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

COST TU1404 WG2 benchmark on numerical modelling of CBM at early age - stage 2 View project

Arch Dams : Spatial variability of the seismic ground motions View project

All content following this page was uploaded by Jean-Paul Fabre on 08 April 2017.

The user has requested enhancement of the downloaded file.

Thermal Deformation of Concrete Dams: Justification,
Clarification and Improvement of Statistical Analysis

M. TATIN, EDF-DTG / Grenoble-INP / University Joseph Fourier / CNRS

UMR 5521, 3SR Laboratory, Grenoble, France.
M. BRIFFAUT, Grenoble-INP / University Joseph Fourier / CNRS UMR
5521, 3SR Laboratory, Grenoble, France.
F. DUFOUR, Grenoble-INP / University Joseph Fourier / CNRS UMR
5521, 3SR Laboratory, Grenoble, France.
A. SIMON, EDF-DTG, Grenoble, France.
J.P. FABRE, EDF-DTG, Toulouse, France.
B. ROUSSET, EDF-DTG, Grenoble, France.

SYNOPSIS. Concrete dam displacements measured by pendulum need to

be interpreted to evaluate the dam behaviour. This article presents different
statistical models used to evaluate the thermal displacements in concrete
dams. The difficulties to assess these displacements are highlighted and
possibilities of improvements are presented. A new model based on both
water and air temperatures is then detailed and results obtained for a French
dam are presented.

INTRODUCTION
Dam safety is an important issue for dam management. Although the
probability of dam failure is very low, such an event would lead to very
significant loss. The associated risk is thus very high. Moreover, as the
structural vulnerability increases with dam ageing, it is essential to monitor
them to ensure their safety. The structural health diagnosis of large concrete
dams is based on monitoring which aims at detecting and quantifying, as
soon as possible, the slightest change in dam behaviour.
Monitoring consists in collecting data from instruments and to interpret
these measures. The main part of dam surveillance is to analyse gathered
data to ensure that the dam is functioning as intended, to detect any possible
anomalies, and to warn of any change which could endanger its safety. Data
analysis is also a mean to better understand the long term behaviour of
dams. Since structural responses of dams are influenced by several factors,
engineers use different analysis tools to evaluate dam behaviour from
collected data.

Dams: Engineering in a Social & Environmental Context. Thomas Telford, London, 2012
2 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT
The analysis of displacement measurements from direct or inverted
pendulums represents an important part of concrete dam surveillance. These
displacements are influenced by various factors such as hydrostatic load,
thermal effect and time-dependent irreversible phenomena (creep, alkali-
aggregate reaction, adaptation, consolidation, damage...). The simplest
analysis consists in plotting the measured displacements as a function of
time. Nevertheless, this type of graphs is difficult to analyse because of the
scattering due to external reversible influences (thermal and filling
conditions). Consequently, to provide surveillance of its dams, EDF uses
statistical models to separate the influences of the different explicative
factors. It is then possible to observe anomalies or irreversible evolutions.
Moreover, the understanding of reversible influences, gives precious
information on the behaviour of the structure.

MODELS IN USE AND POSSIBILITY OF IMPROVEMENT

Displacements measurements are widely influenced by the evolution of
thermal conditions and filling conditions. Consequently, direct surveillance
of displacements can only be done by comparison of similar situations
(thermal state and retention level). In order to prevent this problem and to
generalise comparison to all situations encountered, EDF uses statistical
analysis methods to correct measurements from both thermal and
hydrostatic influences (postulated as reversible influences) and thus to
highlight irreversible behaviour (figure 1).

Figure 1. Measured and corrected displacements (top left), modelled thermal

displacements (top right) and modelled hydrostatic displacements (bottom)
 TATIN BRIFFAUT DUFOUR SIMON FABRE ROUSSET 3

These statistical models are called “surveillance at reconstitute constant

conditions”. Two of these models are mainly used to analyse displacements
of concrete dams: HST (Hydrostatic, Seasonal, Time) and HSTT (Thermal
HST).

HST (Hydrostatic, Seasonal, Time)

The so-called HST statistical analysis method has been developed at EDF
by Willm and Beaujoint [1]. In this model, the measured displacements are
assumed to be the sum of three influences:
- The thermal influence which is modelled as a sinusoidal function of
the season only. The seasonal evolution depends on an angle S
which is equal to 0° the 1st of January and to 360° the 31st of
December.
- The hydrostatic influence, modelled as a polynomial function of
degree 4 of the retention level Z.
- The irreversible influences, if any, modelled by a combination of a
polynomial function (creep, swelling and quick evolutions) and an
exponential function of the time t (consolidation, adaptation ...)

Recorded dam displacements Y0 are thus modelled by the following

expression:

𝑌0 = 𝑎1 + 𝑎2 ⋅ 𝑡 + 𝑎3 ⋅ 𝑡 2 + 𝑎4 ⋅ 𝑡 3 + 𝑎5 ⋅ 𝑡 4 + 𝑎6 ⋅ 𝑒 −𝑡 𝜏
(1) +𝑎7 ⋅ 𝑍 + 𝑎8 ⋅ 𝑍 2 + 𝑎9 ⋅ 𝑍 3 + 𝑎10 ⋅ 𝑍 4 + 𝑎11 ⋅ cos 𝑆
+𝑎12 ⋅ sin 𝑆 + 𝑎13 ⋅ cos 2S + 𝑎14 ⋅ sin 2S + 𝜖

The coefficient a1 to a14 and the parameter η are adjusted on measurements

by the least square method. ε represents the scattering of the model which
contains the uncertainties of both the experimental measurements and the
model.

The results obtained with HST over several decades of displacement

analysis of concrete dams have confirmed the relevance and robustness of
this method. The main advantage of this model is its simplicity and the fact
that it does not need temperature measurements to account for the thermal
influence.
However, the main limitation of this model is that the thermally induced
displacements are modelled as an annual periodic response. Approximating
the thermal effect by an average seasonal distribution gives relevant results
in the vast majority of cases. Nevertheless, since the real temperature
evolution is not accounted for in Eq. (1), the performance of HST is not
always sufficient, in particular for time periods colder or warmer than
seasonal average. Moreover, HST is not able to capture a drift in the thermal
4 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT
state of the structure (global warming). Consequently, if such a drift occurs
it could be interpreted as an irreversible effect.

HSTT (Thermal HST)

Since 2004, an improvement of the HST model, called HSTT (Thermal
HST) [2] is in use for arch dams. This model completes the seasonal
function by a corrective term which takes into account the delayed effect of
the structure to daily temperature difference between the one recorded in the
air and the average seasonal one for this time of the year. As for hydrostatic
and time-dependent effects, they stay unchanged compared to the HST
model. In this model, the major factor in thermally induced displacement is
assumed to be the air temperature. For each day j, the air temperature Θ(j) is
decomposed as the sum of the average seasonal air temperature N(j)
(modelled by a sinusoidal function with a period of one year) and the
difference ΔΘ(j) between the daily recorded temperature and this seasonal
temperature.

(2) 𝛩(𝑗) = 𝑁(𝑗) + 𝛥𝛩(𝑗)

The principle of HSTT is described in table 1.

Table 1. Principle of HSTT
Air = Average seasonal + Deviation at day j
temperature temperature at day j N(j) ΔΘ(j)
Thermal inertia of the structure
(delay and attenuation)
Thermal = Seasonal function of period 1 + Thermal corrective
displacement year with shifted phase function delayed
same for H.S.T. & H.S.T.T. specific H.S.T.T.

The delayed elevation ΔΘR of the mean temperature of the structure induced
by the air temperature deviation ΔΘ is calculated by convolving the
deviation ΔΘ with the impulse response of the structure. This convolution
product can be expressed as a recurrence formula:
𝑇 𝑇
(3) 𝛥𝛩𝑅0 (𝑡 + 𝑑𝑡) = 𝛥𝛩(𝑡 + 𝑑𝑡)(1 − 𝑒 𝑑𝑡 𝑇0
) + 𝛥𝛩𝑅0 (𝑡)𝑒 −𝑑𝑡 𝑇0

In Eq. (3), the parameter T0 is the characteristic time of the structure

representing its thermal inertia which depends on the geometrical and
thermal properties of the structure.
The thermally induced displacements due to the deviation ΔΘ are then
assumed to be proportional to the delayed elevation ΔΘR of the mean
temperature with a factor of thermal sensitivity K.
 TATIN BRIFFAUT DUFOUR SIMON FABRE ROUSSET 5

In the HSTT model, the recorded displacements Y0 are modelled by:

𝑌0 = 𝑎1 + 𝑎2 ⋅ 𝑡 + 𝑎3 ⋅ 𝑡 2 + 𝑎4 ⋅ 𝑡 3 + 𝑎5 ⋅ 𝑡 4 + 𝑎6 ⋅ 𝑒 −𝑡 𝜏
(4) +𝑎7 ⋅ 𝑍 + 𝑎8 ⋅ 𝑍 2 + 𝑎9 ⋅ 𝑍 3 + 𝑎10 ⋅ 𝑍 4 + 𝑎11 ⋅ cos 𝑆
𝑇
+𝑎12 ⋅ sin 𝑆 + 𝑎13 ⋅ cos 2S + 𝑎14 ⋅ sin 2S + 𝐾𝛥𝛩𝑅0 + 𝜖

In Eq. (4), the coefficient a1 to a14 , K and T0 are adjusted on measurements

by the least square method. ε represents the scattering of the model which
contains the uncertainties of both the experimental measurements and the
model.

This improvement results in a better correction of the thermal influence for

colder or warmer time periods than usual. The HSTT model can explain in
the vast majority of cases the displacements at the crest of thin arch dams.
For example, the heat wave of 2003 in France has generated important
displacements which were not explained by the HST model and are better
explained by the HSTT model. Besides, compared to HST, HSTT reduces
the scattering of corrected data and has a better explanatory quality. This
reduced scatter allows an earlier detection of anomalies and then the
structure behaviour can be better diagnosed.

Possibilities of improvement
It is well known that temperature variations are one of the most important
influences that affect the recorded displacements. Thus, this thermal effect
has to be modelled as accurately as possible. There are different thermal
influences which act on dam displacements: air temperature, water
temperature, heat transfers from foundations, solar radiation...
The HST model takes into account all these different phenomena in only
one seasonal function. In the case of HSTT, the deviation on air temperature
compared to the average seasonal value is separated from the other
influences. Nevertheless, in both models, the different thermal influences
are not explicitly taken into account.
An important possibility of improvement for the model is to separate all
these influences which are probably not well modelled by a unique seasonal
function.
Moreover, in the HST and HSTT models the seasonal and the hydrostatic
influences are considered to act separately, whereas they are in reality
coupled. Indeed, the influence of water temperature is dependent on the
retention level. Besides, the retention level follows often a cyclic evolution
due to exploitation conditions. If this evolution of the retention level is in
phase with the seasonal temperature variation, hydrostatic and thermal
effects are correlated and it is then difficult to separate them.
6 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT
USE OF TEMPERATURE MEASUREMENTS
As detailed before, the HST model does not need any temperature
measurements to evaluate thermal effect. The use of temperature
measurements in the models allows to better take into account the thermal
state of the structure and then to evaluate more precisely the thermally
induced displacements. Several methods are possible to use temperature
measurements in statistical models.

In the HSTT model, an improvement has been done by integrating air

temperature measurements. The elevation of the mean temperature within
the structure is determined by solving a direct one-dimensional heat transfer
problem with recorded air temperature as boundary conditions.

It is also possible to exploit concrete internal temperatures if representative

internal thermometric data are available. One of the advantages of this
method is to account for all external solicitations (radiation, air temperature,
water temperature) because internal thermometers directly measure the
effect of these solicitations. However, one issue is the loss of information
when using embedded thermometers. Indeed, the high frequencies of
thermal signal do not penetrate deeply in structures and are not captured by
thermometers if they are too far from the surfaces. For example, a signal
with a time period of one day will have a length of penetration of
approximately 50 centimetres. If the thermometer is located at more than 50
centimetres from the boundary it will not measure the daily variations of the
solicitations. Nevertheless, as these high frequency signals impact a few part
of the structure, their effect on global displacements should be lower.
The first approach proposed is to use directly the measured temperature in
the statistical formulation. It is the case of the HTdT model (hydrostatic,
direct temperature, time) proposed by Weber [3]. In this model temperature
measurements inside the dam are new explicative variables that replace the
seasonal function. The general formulation of this type of model is given by
Eq. (5) where nT is the number of thermometers used and Ti is the
temperature measurements from the thermometer i. It is interesting to use
the measurements for several elevations near both the upstream and the
downstream surfaces and in the middle of the cross section in order to catch
as well as possible the different thermal influences.

(5) 𝑌0 = 𝑎1 + 𝑎2 ⋅ 𝑡 + 𝑎3 ⋅ 𝑡 2 + 𝑎4 ⋅ 𝑡 3 + 𝑎5 ⋅ 𝑡 4 + 𝑎6 ⋅ 𝑒 −𝑡 𝜏
+𝑎7 ⋅ 𝑍 + 𝑎8 ⋅ 𝑍 2 + 𝑎9 ⋅ 𝑍 3 + 𝑎10 ⋅ 𝑍 4 + 𝑖=𝑛𝑇𝑖=1 𝑏𝑖 ⋅ 𝑇𝑖 + 𝜖

A better way to exploit concrete temperatures is to use them to reconstitute

the thermal field inside the structure and to calculate displacements from
this thermal field. In this case a one-dimensional inverse problem is solved
to obtain the temperature at the surface from two thermometers inside the
 TATIN BRIFFAUT DUFOUR SIMON FABRE ROUSSET 7

structure and then the thermal field is rebuilt from the surface temperatures.
This method has been treated by Léger and Leclerc [4] in the case of
periodic signal and by extension for any transient signal by adding trailing
temperatures at the end of the signal. The reconstituted thermal field is
decomposed along different sections in average and linear temperature
difference which are used in the so-called HTT (Hydrostatic, Temperature,
Time) model. The formulation of this model is given by Eq. (6) where nTsec
is the number of sections where the one dimensional thermal field T(x) is
computed from thermometers located in the section, Tm,i and Tg,i are
respectively the mean and linear difference temperatures of the section i and
Tref is the reference temperature (long term average concrete temperature).
In this model the thermally induced displacements are separated in two
parts. One is proportional to the elevation of the mean temperature from the
reference temperature, and the other is proportional to the linear difference
temperature. As for the HTdT model, the sections have to be properly chosen
to account for all the thermal influences. However, in practice, the chosen
sections are the ones which contain the thermometers.

𝑌0 = 𝑎1 + 𝑎2 ⋅ 𝑡 + 𝑎3 ⋅ 𝑡 2 + 𝑎4 ⋅ 𝑡 3 + 𝑎5 ⋅ 𝑡 4 + 𝑎6 ⋅ 𝑒 −𝑡 𝜏

(6) +𝑎7 ⋅ 𝑍 + 𝑎8 ⋅ 𝑍 2 + 𝑎9 ⋅ 𝑍 3 + 𝑎10 ⋅ 𝑍 4

+ 𝑛𝑇𝑠𝑒𝑐
𝑖=1 𝑏𝑖 ⋅ 𝑇𝑚 ,𝑖 − 𝑇𝑟𝑒𝑓 + 𝑐𝑖 ⋅ 𝑇𝑔,𝑖 + 𝜖

A more rigorous approach to calculate thermal displacements from

embedded thermometers has been proposed by Weber, Perner and
Obernhuber [5]. In this method, the temperatures at the boundaries are
calculated from internal thermometers with a one-dimensional inverse heat
transfer problem. These temperatures are then extrapolated to the entire
upstream and downstream surface and convolved with impulse responses to
compute the mean and the gradient of the temperature field across the
structure. The thermal displacements are not calculated statistically as
before but employing the thermo-elastic reciprocal theorem. To calculate the
displacement at a given point in a given direction, one needs to know the
stress first invariant field Θ = ζxx + ζyy + ζzz in the structure due to a unit
force applied at this point and in this direction. This stress field can be
obtained by a finite element simulation. The thermal displacements δ due to
the thermal field T can be calculated by:

(7) 𝛿(𝑡) = 𝛼 ⋅ 𝑉
𝛩 ⋅ 𝑇(𝑡) ⋅ 𝑑𝑥

In this expression α is the thermal expansion coefficient. To simplify Eq. (7)

it is assumed that the stresses vary linearly over the thickness for the applied
unit force. This approximation is reasonable for a thick enough structure and
far enough from the foundations. With this approximation the integral can
be written in the radial direction as Eq. (8) where ΘM, ΘD, TM and TD are the
8 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT
mean and the gradient of the first invariant and the mean and the gradient of
the temperature over the thickness respectively.
𝐿 𝐿
(8) 0
𝛩 ⋅ 𝑇(𝑡) ⋅ 𝑑𝑉 = 𝐿 ⋅ 𝛩𝑀 ⋅ 𝑇𝑀 (𝑡) + 3 ⋅ 𝛩𝐷 ⋅ 𝑇𝐷 (𝑡)

Eq. (8) justifies that only the mean and the gradient of the temperature field
are necessary to compute thermal displacements if the structure is thick
enough. The mean and the gradient of the temperature are then considered to
be constant over horizontal arches to compute the integral along an arc of
length s (Eq. 9).
𝑠 𝐿 𝑠
𝛼⋅ 0 0
𝛩 ⋅ 𝑇 𝑡 ⋅ 𝑑𝑆 = 𝑇𝑀 𝑡 ⋅ 𝐿 ⋅ 𝛼 ⋅ 0 𝑀
𝛩 ⋅ 𝑑s
𝐿 𝑠
+𝑇𝐷 𝑡 ⋅ 3 ⋅ 𝛼 ⋅ 0 𝛩𝐷 ⋅ 𝑑𝑠
(9) = 𝑇𝑀 𝑡, 𝑕 ⋅ 𝑀 𝑕 + 𝑇𝐷 𝑡, 𝑕 ⋅ 𝐷 𝑕
𝑠 𝐿 𝑠
𝑤𝑖𝑡𝑕 𝑀 𝑕 = 𝐿 ⋅ 𝛼 ⋅ 0 𝑀
𝛩 ⋅ 𝑑𝑠 𝑎𝑛𝑑 𝐷 𝑕 = 3 ⋅ 𝛼 ⋅ 0 𝐷
𝛩 ⋅ 𝑑𝑠

The variables M(h) and D(h) give respectively the influences of the mean
and the gradient of the temperature at the dam height h on the thermal
displacements where the unit force has been applied. The final thermal
displacements can be obtained by integrating Eq. (9) over the height.

The difficulties in the use of embedded thermometers are the necessity to

use several measurements to reconstitute correctly the thermal field. Indeed,
water temperature depends on the depth, solar radiation is not homogeneous
on the impacted surface (shade effect, and orientation of the surface) and
near the foundations the measurements are highly influenced by the
conductive heat transfer from the rock.

NEW MODEL AND APPLICATION

Based on the different ideas exposed previously, a new model has been
developed at EDF. This model accounts for air and water temperature by
solving a one-dimensional heat transfer problem at several elevations of the
dam with water and air temperature signals at the boundaries. A seasonal
function is also used to take into account the radiative effect.

Impulse responses
For a one dimensional semi-infinite medium starting from x=0, the
structural response to an impulsion (Dirac) of weight Tup is given by Eq.
(10), where t is the time, and a is the diffusivity of the medium.

𝑇𝑢𝑝 ⋅𝑥 −𝑥 2
(10) 𝑇𝑠𝑒𝑚𝑖 −𝑖𝑛𝑓𝑖𝑛𝑖𝑡𝑒 𝑥, 𝑡 = 2⋅𝑡⋅ ⋅ 𝑒 4⋅𝑎 ⋅𝑡
𝑎⋅𝜋⋅𝑡
 TATIN BRIFFAUT DUFOUR SIMON FABRE ROUSSET 9

To obtain the solution for a finite medium of length L, one needs to

superpose an infinite number of semi-infinite solutions in order to satisfy the
boundary conditions at x=0 and x=L. The mean temperature for the finite
medium can then be expressed as:

1 𝑛 +1 𝐿
∞ 𝑛
(11) 𝑇𝑚𝑒𝑎𝑛 𝑥, 𝑡 = 𝐿 ⋅ 𝑛=0 −1 𝑇𝑠𝑒𝑚𝑖 −𝑖𝑛𝑓𝑖𝑛𝑖𝑡𝑒 𝑥, 𝑡 𝑑𝑥
𝑛𝐿
𝑇𝑢𝑝 ⋅ 𝑎 −𝑛 2 ⋅𝐿 2
∞ 𝑛
= ⋅ 1+2⋅ 𝑛=1 −1 ⋅𝑒 4⋅𝑎 ⋅𝑡
𝐿⋅ 𝜋𝑡

The expression of Eq. (11) is the mean temperature response in a one

dimensional wall of length L to an impulse solicitation of weight Tup on one
side of the wall. By extension, Eq. (12) provides the mean temperature
response in the wall for an impulse solicitation of weight Tup on one side
and Tdo on the other side of the wall:

(𝑇𝑢𝑝 +𝑇𝑑𝑜 )⋅ 𝑎 ∞ −𝑛 2 ⋅𝐿 2
(12) 𝑇𝑚𝑒𝑎𝑛 (𝑥, 𝑡) = ⋅ (1 + 2 ⋅ 𝑛 =1
(−1)𝑛 ⋅ 𝑒 4⋅𝑎 ⋅𝑡 )
𝐿⋅ 𝜋𝑡

Consequently, to obtain the mean temperature in the one-dimensional

medium of length L for any signals on the two sides of the medium, one
needs to convolve the mean of the two signals (Tup + Tdo)/2 by the impulse
response given by:

−𝑛 2 ⋅𝐿 2
2⋅ 𝑎 ∞ 𝑛
(13) 𝑇𝑚𝑒𝑎𝑛 𝑥, 𝑡 = 𝐿⋅ ⋅ 1+2⋅ 𝑛=1 −1 ⋅𝑒 4⋅𝑎 ⋅𝑡
𝜋𝑡

It is worth noting that the impulse response for the mean temperature (Eq.
(13)) is the same as the one used and demonstrated by Weber, Perner and
Obernhuber [5].
Concerning the gradient temperature in the one-dimensional medium of
length L, it can be calculated for any signal on the two sides of the medium
by convolving the signal (Tup – Tdo)/2 by the impulse response [5] given by:

−𝑛 2 ⋅𝐿 2
6⋅ 𝑎 ∞ 12⋅𝑎
(14) 𝑇𝑔𝑟𝑎𝑑 𝑥, 𝑡 = 𝐿⋅ ⋅ 1+2⋅ 𝑛=1 𝑒 4⋅𝑎 ⋅𝑡 −
𝜋𝑡 𝐿2

Air and water temperatures

In this model the dam is sliced up into 10 layers (Fig. 2). Each layer is
considered as a one-dimensional medium and the mean and the gradient of
the temperature field are individually computed from air and water
temperatures using convolving products with the previously seen impulse
responses (Eq. (13) and (14)).
10 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT

Figure 2. Dam slicing into 10 parts along the height.

For each layer, the signal Tdo is the daily air temperature signal Tair and,
depending whether the layer is above or below the water level, the signal Tup
is the daily air temperature signal Tair or the daily water temperature signal
Twater(d) at depth d, respectively.
The daily air temperature signal Tair is determined by air temperature
measurements. As water temperatures are not monitored along the reservoir
depth, a model is used to predict the signal Twater(d). The water temperature
model is an empirical model based on two measurements campaigns, one in
summer and one in winter. The temperature at the water surface is modelled
by Eq. (15) where the angle Sw varies linearly between 0° (the 25th of
January) and to 360° (the 24th of January the following year).

(15) 𝑇𝑠𝑢𝑟𝑓 (𝑡) = 13 − 8 ∗ cos(𝑆𝑤 )°𝐶

Temperatures at different depths under the water surface are calculated by

means of Eq. (16) where d is the depth under the water surface (in meter).

(16) 𝑇𝑤𝑎𝑡𝑒𝑟 (𝑑, 𝑡) = 5 + (𝑇𝑠𝑢𝑟𝑓 (𝑡) − 5) ⋅ 𝑒 −𝑑 13

°𝐶

Statistical model and results

As shown in Eq. (9), with some assumptions, at a given elevation, the
contribution to thermal displacements is proportional to the mean and the
gradient of the temperature field calculated for this elevation. The
coefficients of influence M(h) and D(h) are adjusted statistically for each
layer. To account for solar radiation, a seasonal function has been added to
the model. The formulation of the model is given by Eq. (17) where nL is
the number of layers, Tmean,i and Tgrad,i are respectively the mean and the
gradient of the temperature field calculated for the layer i.

𝑌0 = 𝑎1 + 𝑎2 ⋅ 𝑡 + 𝑎3 ⋅ 𝑡 2 + 𝑎4 ⋅ 𝑡 3 + 𝑎5 ⋅ 𝑡 4 + 𝑎6 ⋅ 𝑍 + 𝑎7 ⋅ 𝑍 2
(17) +𝑎8 ⋅ 𝑍 3 + 𝑎9 ⋅ 𝑍 4 + 𝑎10 ⋅ cos 𝑆 + 𝑎11 ⋅ sin 𝑆 + 𝑎12 ⋅ cos 2S
+𝑎13 ⋅ sin(2S) + 𝑛𝐿 𝑖=1 𝑏𝑖 ⋅ 𝑇𝑚𝑒𝑎𝑛 ,𝑖 + 𝑐𝑖 ⋅ 𝑇𝑔𝑟𝑎𝑑 ,𝑖 + 𝜖
 TATIN BRIFFAUT DUFOUR SIMON FABRE ROUSSET 11

Figure 3. Comparison between displacement measurements with pendulum

and displacements calculated by the model.

It is worth noting that the modelled displacements are able to well reproduce
the observed displacements (figure 3) and particularly when these
displacements are more important than usual (2003 heat wave for example).
These unusual displacements are due to unusual thermal conditions and can
be observed in figure 4 which compares thermal displacements modelled by
HST, HSTT and the new model. When the thermal displacements of HST
cannot take into account these unusual thermal conditions, those of HSTT
and of the current model are more representative of the real thermal
conditions.

Figure 4. Comparison between thermal displacements calculated by HST,

HSTT and the new model
 12 DAMS: ENGINEERING IN A SOCIAL &
ENVIRONMENTAL CONTEXT
CONCLUSIONS AND PERSPECTIVES
Different methods are used to estimate thermally induced displacements in
concrete dams. The use of temperature measurements seems to add precious
information for statistical models but it is not easy to exploit them correctly.
The new model presented in this contribution, although it is more complex
than HSTT (it adds water temperature measurements, separates the
influences of several layers and take into account the gradient effect), does
not improve significantly HSTT. The scattering is slightly reduced and the
correlation coefficient is equal to 0.9825 instead of 0.9803 for HSTT.
Nevertheless, different improvements could be considered:
- The air temperature signal posses high frequencies and could be
insert in a smoother way.
- Concrete temperature signal could be used in order to have a better
assessment of the thermal field inside the structure.
- Radiative effects and water temperatures effect could be added in a
better way (measurements)
- The hypothesis of uni-axial conduction is not confirmed, in
particular near the foundations.

To improve statistical model, it is also possible to use finite element

simulations to evaluate and quantify all thermal effects (solar radiation,
convection, transfer from foundation, etc.) and to validate the hypotheses
and better assess the thermo-mechanical fields of the statistical approaches.
At the end, it is possible to recapture statistical models, to clarify and to
improve them in order to estimate more accurately the behaviour of arch
dams.

REFERENCES
[1] Willm, G., Beaujoint, N. (1967). Les méthodes de surveillance des
barrages au service de la production hydraulique d’Électricité De France.
Problèmes anciens et solutions nouvelles. Neuvième congrès des grands
barrages, Istanbul.
[2] Penot, I., Daumas, B., Fabre, J.P. (2005). Monitoring behaviour.
International Water Power & Dam Construction, December 2005.
[3] Weber, B. (2001). Linear regression models for dam monitoring based
on statistical analysis. Proceedings, 6th International Benchmark Workshop
on Numerical Analysis of Dams, ICOLD, Austria.
[4] Léger, P., Leclerc, M. (2007). Hydrostatic, Temperature, Time-
Displacement Model for Concrete Dams. Journal of Engineering Mechanics,
March 2007.
[5] Weber, B., Perner, F., Obernhuber, P. (2010). Displacements of concrete
dams determined from recorded temperatures. 8th ICOLD European Club
Symposium, Innsbruck, 2010.

View publication stats