Analysis of earthdam monitoring data

Stéphane BONELLI, Rémy TOURMENT, Huguette FELIX

Cemagref, Le Tholonet, 13612 Aix-en-Provence, France

Tel: (33) 442 669 910 E-mail: stephane.bonelli@cemagref.fr

English summary
To understand the long-term behaviour of a dam, it is essential to carefully interpret the
measured data, in order to be able to distinguish between the effects of factors such as drift,
irreversible events and the ageing of the dam and the effects of other factors not involving
ageing processes.
The Hydrostatic-Season-Time (HST) model is presented in this paper. The main
components of the basic model are: the effect of the reservoir level; the seasonal effect; and
time drift. The rainfall must be taken into account when analysing hydraulic related
measurements.
Piezometric levels of a homogeneous embankment dam (15.5 m high) are analysed, as well
as variations of the measurements of three 3D crackmeters, installed in a spillway. These
case studies are interpreted.

Résumé français : Analyse de mesures d'auscultation de barrages en terre

Pour comprendre le comportement à long terme d'un barrage, il est essentiel d'interpréter
soigneusement les mesures d'auscultation, afin de pouvoir distinguer les effets des facteurs
tels que la dérive, les événements irréversibles et le vieillissement du barrage, des effets
d'autres facteurs n'impliquant pas des processus vieillissant.
Le modèle Hydrostatique-Saison-Temps (HST) est présenté dans cet article. Ses
composants principaux de base sont: l'effet du niveau de réservoir; l'effet saisonnier; et la
dérive dans le temps. Les précipitations doivent être prises en considération pour l'analyse
de mesures liées à l'hydraulique.
Les niveaux piézométriques d'un barrage en terre homogène de 15,5 m de hauteur sont
analysés, ainsi que les variations des mesures de trois vinchons, installés dans un coursier
de déversoir de crues. Le résultat de ces études de cas sont interprétés.

Streszczenie polskie: Analiza danych pomiarowych dla zapór ziemnych

Aby zrozumieć zachowanie się zapór w długim okresie czasu niezbędne jest
przeprowadzenie dokładnej interpretacji danych pomiarowych w celu stwierdzenia różnic
między efektami spowodowanymi takimi czynnikami jak zmiany w czasie, zjawiska
nieodwracalne, starzenie się zapór, a wpływem innych czynników nie związanych ze
starzeniem się.
W pracy przedstawiony jest model HST (Hydrostatyka-Sezon-Czas) . Głównymi składnikami
podstawowego modelu są: wpływ piętrzenia, zjawiska okresowe, zmiany w czasie. Przy
analizie danych pomiarów hydraulicznych należy uwzględnić opady.
W pracy przeanalizowane są odczyty piezometrów umieszczonych w zaporze ziemnej
wysokości 15,5m, jak też dane pomiarowe z trzech 3D szczelinomierzy umieszczonych w
sekcji przelewowej. Wyniki tego przykładu są zinterpretowane.

133
1. Introduction
Dam monitoring comprises two essential methods:
- visual monitoring; this is a qualitative method which is fundamental because it
integrates the complexity of the behaviour of the structure;
- measurements of data; this is a quantitative method which uses instrumentation
and analysis of measurements specific to each structure.

Dam monitoring is essential for the follow-up of the dam, from its design to its
putting out of service; it is a component of its structural behaviour and of the control
of its safety. It is also invaluable to make progresses in the knowledge on the
behaviour and the ageing of the dam, and make it possible to improve the studies
and expertises in their various technical and economic aspects (Poupart et al.,
2000). From this point of view, monitoring is an essential component of progress. It
makes possible to tell the owner about, before it is too late, necessary
reinforcement and repair works and, in extreme cases, emergency measures
ensuring protection of the downstream populations. Statistics show that there is a
strong correlation between the most serious accidents of dams and the absence of
organized monitoring (Londe, 1990).

The follow-up of ageing is essential. The acceleration - always alarming - of any

phenomenon must be detected as soon as possible. A true monitoring of the work
supposes that one can follow the evolution of its behaviour in time, substraction of
the other variations previously done. This is exactly what stipulates French
regulation.

Statistical methods of Hydrostatic-Season-Time (or HST) type were proposed in

the years 1960 to analyze measurements of displacements resulting from
pendulums in arch dams (Ferry and Willm, 1958; Willm and Beaujoint, 1967;
Lugiez and Al, 1970). These methods are currently used in several countries, to
analyze other standard measurements (Guedes and Coehlo, 1985; Silva Gomes
and Silva Matos, 1985; Seersucker and Lino, 1999; Carrère and Al, 2000).
Experience gained for a few decades on several hundreds of dams has confirmed
the excellence of this approach as a powerful tool for interpretation of dam
monitoring measurements (sea ICOLD reports: Mary, 1948; Yoshida, 1958; Oberti,
1964; Post, 1985; Marazio, 1989; Dibiagio, 2000).

In this document, application of the HST method for analysing hydraulic

measurements of earthdam is presented. The essential components of monitoring
and analysis of measurements are recalled in section 2 and 3. The HST model is
exposed in section 4. This is the traditional model, intended for concrete dams
displacements measurements analysis, supplemented by a rain component model.
Application to analysis of piezometric measurements is detailed in section 5. An
example of the results of its application to analysis of the measurements of cracks
in the reinforced concrete walls of a spillway is also presented. Section 6 is
devoted to a critical analysis of this method.

134
2. Monitoring
The types of measurements and the most widespread apparatuses for earthdams
monitoring are briefly described below. These measurements relate on the
loadings and the response of the work. The two main loadings are:

- water level, measured by visual staff gauge or byv water level recorder (precision:
centimeter);
- rainfall, measured by pluviometer installed on the dam site, with daily
measurements (precision: millimeter/day).

It can sometimes be necessary to take into account other loadings: the

downstream water level, the side ground water, the snow melt. The response of
the work is apprehended by measurements of displacements and hydraulic
measurements.

Displacement measurements can be of three types:

- surface displacements, measured by topographic measurements -levelling and

planimetry- settlements, upstream/downstream and left bank/right bank
movements; these measurements need specialized aptitudes, once or twice each
year;
- in-depth displacements, measured by instruments which are generally installed
during construction (pendulum, inclinometer, extensometer, settlements gages);
- relative displacements, along a joint or of a crack, quantified by instruments
generally installed when needed (crack meter, 3D crack meter).
Hydraulic measurements can be:
- drains discharge, measured with a graduated container or by flow gauging weir
(weir with liquid level recorder, and use of a weir formula to estimate the flow);
they relate to the drains, the relief wells, the zones of resurgence;
- hydraulic loads, measured by piezometers; the open tube piezometer is the
simplest system and also the most robust; it can be established at any time in the
dam body, but also in foundation, banks, abutments and downstream platform; it
is a low diameter drilling, equipped with a casing of few centimeters;
measurement is done using a probe giving the difference in dimension between
the water level and the head of the stand pipe piezometer; the precision is of the
order of the centimeter;
- pore water pressures, measured by pore water pressure cells or vibrating wire
piezometers, which are low dimension apparatuses, installed during construction
in the dam body or in the foundation; the principle of measurement depends on
the type of apparatus (backpressure or vibrating wire); the precision is from 0.1 to
1 kPa (or 1 to 10 cm in water load).

In the dam body, a complete hydraulic monitoring system generally includes

several upstream/downstream profiles making it possible to evaluate the saturation
of the body of the stopping, and one or more flow measurement points whose role
is to detect the appearance and the evolution of leakages zones. The foundation,

135
the abutments, the banks and the downstream platform must also be monitored.
Figure 1 is an example of monitoring the dam body upstream from the chimmey
drain and base/foundation interface by vibrating wire cells. Figure 2 shows a
piezometer system for monitoring the surroundings of this dam.

Figure 1. Dam cross section and location Figure 2. Map sight of the dam and
of the pressure cells (Chamboux dam). location of the piezometers
(Chamboux dam).

For a dam in regular use, periodicity of measurements is known as normal. In

particular situations (construction, first filling, full emptying or filling, important repair
works), periodicity of measurements is lowered. It is always specific to the
structure. It is essential to carry out measurements in a continuous way during
time. More than their absolute value, it is indeed the evolution of the phenomena in
time which is essential to know. The acceleration of a displacement or the brutal
rise of a water pressure are examples of behaviours which are capital to detect in
due time. It is also essential to carry out measurements in a homogeneous way
during time. The compared analysis of the evolution of several sizes is sometimes
essential to conclude. It is in particular the case of the coupling pore pressure/leak
flow. The lowering in a leak-flow is an example of behaviour which requires the
knowledge of piezometry in the concerned zone: if it decreases, one can show a
clogging of materials inside the dam body (which is a normal phenomenon); if it
increases, one can suspect a clogging of the drainage device (which is a very
worrying phenomenon).

3. Monitoring measurements analysis

The monitoring objective is to sufficiently early detect discontinuities, significant
evolutions, irreversibilities, and all show existence of disorders. In the end, these
are always great displacements of masses or interfaces, announced by precursory
phenomena: rise of pore water pressure, cracks, opening of joints, great
elastoplastic deformations. Here are some examples of evolutions which are likely
to lead to disorders, even possibly to failures, and whose harbingers are sought by
dam monitoring measurements analysis.
Displacements:
- an excessive settlement of the crest, leading to a reduction in freeboard
(difference between the altitude of the crest and the one of projected highest

136
 water level), which diminishes the dam safety with respect to the risk of overflow;
this settlement can reflect a strong compressibility of the foundation, or a
defective compaction of the dam body (insufficient compaction, water content of
the soil material too low during compaction);
- sliding starts, in the downstream slope of the dam, banks slopes or water
reservoir slopes, calling into question the stability of these slopes; the installation
of inclinometers can make it possible to locate the in-depth zone of slipping and
to follow the evolution of the slipping.
Piezometry and pore water pressure:
- a high or increasing piezometry in the dam body; such an evolution is always
worrying; in this case, the the drain can be circumvented or it can be the sign of
slope water input; in all these cases, the dam stability is called in question; if the
piezometric line nears the downstream slope, seepages can evolve to a
regressive internal erosion (piping);
- a high or increasing piezometry in the foundation or the surroundings, harmful to
the structure stability; it can come from a failure in the waterproofing systems of
the foundation and abutments;
- a decreasing piezometry; this evolution in general beneficial can come from the
dissipation of pore water pressures induced by construction, or a reduction in the
permeability of materials by self-clogging.
Leak-flows:
- a decreasing of flows; this decreasing can come from a natural improvement of
the dam and foundation water tightness caused by upstream clogging, evolution
which is beneficial; this reduction can also reflect a filling of the drains, which are
slowly circumvented and so do not control the leak-flows any more; this filling can
be the consequence of an internal phenomenon of erosion (suffusion); in this
very worrying case, upstream piezometry will increase and the uncontrolled flows
can be at the origin of internal erosion (piping) or slips on the downstream slope;
only the measurement of the couple piezometry/flow makes it possible to decide;
- an abnormal increase in flows; it can come from a slope ground water , but also
from a process of internal erosion (piping); in this last case, collected water can
be charged with fine elements coming from the dam body.
A real monitoring of the structure implies to follow the evolution of its behaviour in
time, any other cause from variation being suppressed: this is the analysis under
"constant conditions". It is precisely what is stipulating by the French regulation
relating to the inspection and the monitoring of the dams interesting public safety
(i.e. whose possible failure would have serious repercussions on people, whatever
their height).
The analysis under constant conditions supposes that one is able to quantify the
influence of the two main loadings: water level in the reservoir and precipitations.
This is the analysis of dam monitoring measurements. For a dam in use, the
loadings cause reversible strain since dimensioning was carried out consequently:
the limits of stability or resistance are not reached. The analysis of dam monitoring
measurements then makes it possible to change measurements to constant
conditions in order to quantify the evolutionary and irreversible phenomena. The
speed of evolution can be positive (increase with time), or negative (reduction with
time), and can be accelerated or slowed down. This evolution has then to be

137
interpreted in the light of the knowledge of the structure that one has in addition, or
which one has a presentiment of: dissipation of the pore water pressures of
construction, filling of the drainage curtain, evolution of permeabilities. The
periodicity is annual for the synthesis of rough measurements, and biennial for
their analysis.

4. The HST model

The HST model is based on three effects. The first one is the hydrostatic effect,
which accounts for the variations of measurements in the level of the reservoir. It is
represented by a polynomial function - often of the fourth order - of the reservoir
level Z(t ) at time t :
Z (t n )− Z
Hn = a1 zn + a2 z 2n + a3 zn3 + a4 z 4n , zn = σ (1)

where Z and σ are the mean average and standard deviation over the analyzed
period. The second effect is the date in the year, which accounts for the seasonal
variations S of measurement, periods twelve months and six months. It is
represented by the first two terms of a development in Fourier series:
Sn = b1 sin(ωa tn ) + b2 cos(ω a tn ) + b3 sin 2 (ωa tn ) + b4 sin(ω a tn )cos(ωa tn ) (2)
where ω a = 2π / ∆ta is the annual pulsation ( ∆ ta = 365, 25 years corresponds to one
year). The third effect gives an account of the influence of time, including ageing.
Its expression is variable, and depends on the studied phenomenon. This effect is
often called "irreversible effect". It can for example be represented by the sum of a
linear term, a positive exponential and a negative exponential of reduced time τ n
during the analyzed period [ t0 , t N ]:
t n −t 0
Tn = c1τ n + c2 e τ n + c 3e −τ n , τ n = tN −t 0 (3)

In order to allow analysis of hydraulic measurements, taking into account the

rainfall is necessary. This pluviometric effect P requires to know daily rainfall Q. A
simple method is to take the cumulated rain over the last ten days (Crepon and
Lino, 1999). The taking into account of cumulated rain over several former periods
can lead to good results. One can represent it by a linear combination of the
cumulated rain over seven days during the four previous weeks, and cumulated
rain over fourteen days from the previous weeks (Plancke, 1986; Bonelli et al.,
1998):
4 7
Pn = ∑ d k pnk + d5( p5n + p6n ) + d6 ( pn7 + p8n ) , pkn = ∑ Qn − j − 7(k −1) (4)
k =1 j =1

The statistical model gives an estimate Yn of measurement yn with an error ε n :

Yn = Y0 + H n + Sn + Tn + Pn , yn = Yn + ε n (5)
The 18 parameters are estimated by multiple linear regression which minimizes the
average standard deviation between the centered reduced values, which conducts

138
to maximize the determination coefficient R2. In order for the analysis to be
statistically significant for a risk of 5%, R2 must be higher than 0.45. To minimize
the number of explanatory variables while maximizing R2, a step by step process of
Stepwise type of selection or elimination of the explanatory variables is used. The
explanatory variables are introduced gradually according to a criterion based on
the test of Fisher-Snedecor (F) which measures the increase in R2, and their
selection is called into question after introduction of a new variable. This method is
traditional, and it is advisable to refer to the works of statistics and data analysis
specialists for more details (cf for example Diday, 1982).
The number of parameters p of the final model is in practice always lower than 18,
since variables that are considered not to be very explanatory are not part of it.
Values of R2 and F, then the detailed analysis of each effect and the residues
make it possible to judge quality of an adjustment. Several calculations are
sometimes necessary to lead to an exploitable analysis, by adapting in particular
the period of the analysis to the studied phenomenon: it is in particular the case at
the time of exceptional circumstances (emptying of the reservoir), or of
discontinuities due to interventions on the monitoring device (cleaning of the
drains, replacement of a monitoring device). In order to allow a synthesis of the
behaviour of the structure, we defined some indicators (table 1).

Effect Indicator Meaning

H ( Zmax )− H (Z ) evaluated between the maximum level and the
H α= Z max − Z mean level in order not to take into account
measurements at low level, too very few
(exceptional emptying)
maximum amplitude which would be reached if the
S B = b12 + b22 + annual and semi-annual effects would be in phase
1
2
b32 + b24

∆ta ( TN − T0 ) speed of annual evolution evaluated from the initial

T v= tN −t 0 and final values over the analysis period

P d1 influence of the cumulated rain from the seven

days before measurement

Table 1. Indicators for a HST analysis

5. Application of HST model

5.1 Application to internal hydraulic measurements (Chamboux

dam)
Surface displacements (levelling and planimetry) of an earthdam are generally not
analyzed by statistical method, because the number of measurements is too low
(one to two measurements each year). When settlings are very significant, the
method of Asaoka (1978), based on Terzaghi’s equation of consolidation, allows to

139
estimate the final compressing from the end of primary consolidation. Deep
displacements, as well as the leak-flows can be analyzed by a statistical method of
type HST when there are measurements enough. The rest of this part of this
document relates to piezometric measurements and pore water pressure
measurements.
Piezometric levels of an homogeneous earthdam 15.5 m high are analyzed. The
foundation consists of granitic arenas (k=10-5 to 10-6 m/s). An grout curtain seals it.
The embankment comprises arenas of k=10-8 to 10-9 m/s permeability, upstream
from the chimmey drain. The drain is connected to the downstream by drainage
cords surrounded by filters. Vertical relief wells emerge in the downstream foot
gutter, which recovers drained water. The first filling of the dam was made in 1984.
The system of hydraulic measurements comprises – among others - 13 stand pipe
piezometers with direct reading downstream from the dam and 7 pore water
pressures cells (vibrating cord) in the body of the dam, in the axis of the spillway
(fig. 1 and 2). The statistical analysis of these 20 instruments is carried out over the
1989/98 period, that is to say 9 years and approximately 100 measurements per
instrument. The precision is in the order of the centimetre.
On average, on stand pipe piezometers the levels variation is explained to 30% by
the variation of the water level in the reservoir, to 19% by rainfall, 15% by seasonal
effects and 6% by evolutions in time (fig. 3). For some piezometers, the
explanatory share of the rain can reach 38%. Among the six explanatory variables
of the rain model, it is primarily d1 which is explanatory: 12% on average, and up to
30% (fig. 4).

Figure 3. Contributions to R2 of HST and Figure 4. Contribution to R2 of the

P effects (average on piezometers) parameters of the rain model
(average on piezometers)

An example of detailed results is given in figure 5 (curve of rough measurements

and variations of level in the reservoir), and figure 6 (results of the analysis). The
variation of the levels of the piezometer is primarily ascribable to the variations in
the level of the reservoir. The seasonal effect is significant, with a maximum in
summer and a minimum right before the winter. It is to be noticed that it is the
positive exponential and not the negative exponential which describes the
downward trend with time (analyzes under constant conditions), with a minus
coefficient.
The values of R2 and F are written out on cross sections, as well as on a map view
(fig. 7-10). The quality of analyses is to be connected to the instruments location.
For a number of measurements n>60, the statistical threshold of significance fixed

140
at F1,n=4 (resp. 8), which corresponds at 5% risk (resp. 5‰) of wrongly select or
reject a variable with each step (fig. 11). The statistical indicators (R2,F) show
analyses of excellent quality, except for the cells close to the drain, and the
piezometers located downstream in bottom of valley, behind the grouting curtain,
which is completely acceptable because the variations of the levels are weak.

Figure 5. Water level in the reservoir and piezometric level (PD1)

Figure 6. Influence of reservoir water level, season, rain and time (PD1). The curves
represent the statistical model H, S, T, P. The symbols represent the effect corresponding to
the measurement, that is to say H+ε,S+ε,T+ε,P+ε.

141
The variations of the reservoir level are also analyzed according to S, T and P:
simple regression R2=0.495 and F(13,100)=7.55 and Stepwise regression
R2=0.410 and F(3,110)=25.55. Analysis of the reservoir level shows a seasonal
component (B=±74 cm/an), and a light increase with time of the mean level over
the analysis period (v=+16 cm/an), probably caused by two emptyings at the
beginning of period.
In the surroundings of the dam, the correct operation of the grouting curtain is
illustrated by the decrease, from banks towards the bottom of valley, of the
amplitudes of variation of the levels (fig. 12) and of the α coefficient (fig. 13). The
grouting curtain is circumvented by banks. It illustrated by values of α near to one
on the top of banks. However, it is done far from the dam, since the values of α
decrease towards the bottom of valley (fig. 13). The seasonal variations account
for 5 to 10% of the total variations and are of delicate interpretation (fig. 14). The
evolutions in time of the levels are weak compared to the amplitudes of variation,
which is a good indicator for safety (fig. 15). It is the rainfall of the previous week
which influences primarily the piezometric levels, in a little more significant way in
the bottom of valley than on the banks (fig. 16).
A piezometer located on the left bank, close to the peak, is remarkable: its
variations are much more significant than those of the level in the reservoir (549
against 381 cm, fig. 12), partially ascribable to the reservoir level (α=0.48, fig. 13),
highest seasonal variations (B=±102 cm/year, fig. 14), highest rainfall effect (fig.
16). This behaviour is due to the presence of ground water from the bank,
identified during the project phase. The levels do not present concern. They remain
sufficiently lower than the ground level, and do not show any evolution in time.
In the dam body, the lowering of the groundwater, which illustrates the
effectiveness of the vertical drain, is highlighted by the decrease, from the
upstream towards the downstream, of the amplitudes of variation of the levels (fig.
17) and of the coefficient α (fig. 18). The measured seasonal variations are
ascribable to the reservoir level (fig. 19). The evolutions in time of the levels are
weak compared to the amplitudes of variation which is a good indicator for safety
(fig. 20). Rainfall has a negligible influence on the piezometric cells located in the
dam body.

Figure 7. Indication of R2 on the map view Figure 8. Indication of F on the map view

142
Figure 9. Indication of R2 on the equipped Figure 10. Indication of F on the equipped
cross section cross section

Figure 11. F (Fisher-Snedecor) related to Figure 12. Indication of the total variation
2
R and the number p of explanative amplitude on the map view (cm)
variables for n>60 measurements.

Figure 13. Indication of α on the map view Figure 14. Indication of B on the map view
(±cm/year)

Figure 15. Indication of v on the map view Figure 16. Indication of d1 on the map
(cm/year) view

143
 Figure 17. Indication of the total Figure 18. Indication of α in the equipped
amplitude variation in the equipped cross cross section
section (cm)

Figure 19. Indication of B in the equipped Figure 20. Indication of v in the equipped
cross section(±cm/year) cross section (cm/year)

5.2 Application to displacement measurements on the cracks in

side walls of a spillway (Lac des Sapins)
Measurements of three 3D crackmeters located on cracks of the side walls of the
spillway of an earthdam are analyzed using the HST model. This 16 meters high
dam contains a reservoir of 2 million m3, and has a monitoring system comprising
four cells of pore water pressure, six drains, six stand pipe piezometers, three 3D
crackmeters, and a system of topographic bench marks. All measurements are
analyzed with HST model, except for topographics for the reasons explained
before in this paper. The analysis of measurements of the 3D crackmeters is
commented on hereafter; values are usable only since 1996, because the system
had before been improved by successive stages, before obtaining regular
(problems of access) and reliable (problem of fixings) measurements.
For each 3D crackmeter, X measurement indicates an opening in the horizontal
direction, Y measurement (most interesting) indicates a possible movement of the
side wall towards the interior of the spillway, Z measurement a shift in the vertical
direction. For each of the 3D crackmeters, a drift exists to the Y measurements
since 1996.
The result of the analysis of measurements of the 3D crackmeters with HST model
appears in Table 2. Detailed results are represented in figures 21 to 23 (curves of
rough measurements), and figure 24 (results of the analysis). The variations are
primarily due to time, for measurements Y, and to seasonal effects for the other
measurements. Pluviometry (influencing the saturation of embankments) is
probably an explanatory factor of the variations of the Y measurements of the 3D
crackmeters. Unfortunately, measurements of pluviometry are not available.
Y measurement: this is the most interesting measurement for analysis, as it
indicates a possible movement of the side walls towards the inside of the spillway.
The coefficient of correlation R2 is good to very good for the three measurements

144
(0.967, 0.988, 0.800). As previously supposed by visual observation of rough
measurements, time effect is in all cases the most explanatory (respectively 90, 98
and 42% of explanation for the total phenomenon). Seasonal effect is significant for
V3 (23%). The hydrostatic effect (16%) noted for V3, is to be connected to the
measurements remained "high" for this apparatus during the emptying of the
reservoir. It is to be noted that, contrary to the other two apparatuses, this 3D
crackmeter is located on the reservoir side of the spillway. The noted drifting of
measurements show that the chamber walls undergo the pressure from the earth
they support and move towards the inside of the channel.

Figure 21. X measurements

Figure 22. Y measurements

145
 Figure 23. Z measurements

Vinchon V2Y aval droit

Figure 24. Measurements, influence from water level, season, and time (V2Y).
The curves represent the statistical model H,S,T.

146
 Instr. V1Y V2Y V3Y V1X V2X V3X V1Z V2Z V3Z

Ampl. 2.54 2.80 2.50 1.12 1.00 1.55 1.06 0.86 0.37
(mm)
R2 0.967 0.988 0.800 0.693 0.510 0.808 0.716 0.565 0.352

α 0.000 0.000 -0.084 0 0.011 -0.013 0 0 0

(mm/m)
B
(mm) ±0.29 ±0.14 ±0.59 ±0.40 ±0.20 ±0.54 ±0.12 ±0.16 ±0.08

v 0.341 -0.363 -0.223 0 0 0 0.050 0 0

(mm/year)
Table 2. Result of the statistical analysis of the 3D crackmeters over the 1996-
2001period

X measurement: analysis quality is poor for V2 (R2=0.51), it is average for V1

(R2=0.69), it is good for V3 (R2=0.81). For the three instruments the seasonal effect
is the only one which influences the movements explained by analysis. So, there is
no irreversible movement. The hydrostatic effects are negligible.
Z measurement: statistical analysis quality is average for V1 and V2 (R2=0.72 and
0.57), it is poor for V3 (0.35). Seasonal effect is the only significant one for V2 and
V3. V1 very slowly drifts (+0.05mm/year). In the three cases, there is no
hydrostatic effect.
Broadly, statistical analysis carried out with HST model (115 measurements) gives
satisfactory results on the three 3D crackmeters. It makes it possible to follow the
evolution of the irreversible movements of the side walls of the spillway.

6. Discussion
HST analysis is a traditional approach in data analysis. It is used in many other
fields (Young, 1998). One of the oldest known examples is Forbes' sinusoidal
adjustment (1846) to reproduce the cyclic variations of the temperature of the
ground. For dams, this is a robust approach which leads in the majority of the
cases to suitable results. It has however two gaps: 1) its parameters have little
mechanical signification, 2) it does not take into account the particular structure of
the analyzed time series.
The polynomial expression of the influence of the water level in the reservoir is
historical. It results from applied material strength to analyze displacements of an
arch dam. This explanatory variable is often used by default for the analysis of
hydraulic measurements, but a polynomial relation of degree four (or more)
between a piezometric level and the level in the reservoir at the same moment is
not mechanically justified.
An example of measurement putting at fault HST model is given figures 25 and 26.
Figure 25 gives piezometric measurements monitoring the foundation, and located
on the downstream platform of a dam. A priori, variations during the first years are

147
proportional to the water level in the reservoir, which should result in a linear
relation between piezometric variation and variation of water level in reservoir. But
it is definitely not like this (figure 26), and a polynomial law of type (1), even of
higher order, would not give an account of it. Moreover, if measurements
frequency is not dependent on the water level in the reservoir (which is common in
practical cases), these variations can be statistically and accidentally well
explained by the polynomial model (1), without any real mechanical explanation.
Indeed, when the water level lowerings are slow because of exploitation,
measurements are numerous. When the rises are fast because of significant
precipitations, measurements are rare. Adjustment will then be carried out to the
most common measures, and the result will not be interpretable.
Figure 26 shows that a cycle of raising/lowering of water level in the reservoir is
dissipative (hysteresis): the drawn curve is not the same one for a rise (phase 3)
and for a lowering (phase 5). Some measurements can indicate an increase in
pore water pressure while the level of the reservoir decrease, and conversely. This
well-known phenomenon is due to the stocking capacity: variation of porosity or
presence of air. It is observed on works (Kjaernsli and Al, 1982; Myrvoll and Al,
1985) and was reproduced in laboratory (Windish and Høeg, 2000). This example
illustrates what is the delay effect and the fact that the model (1) cannot give an
account of it.

Figure 25. Water level in the reservoir and Figure 26. Rough piezometric according
rough level in a downstreal piezometer, to level in the reservoir.
according to time.

A mechanical approach based on the Impulse-Response-Function of the work was

developed to analyze hydraulic measurements (Bonelli and Royet, 2001; Bonelli,
2004). It is based on an integral representation - starting from the Green function -
solution of the associated linear parabolic problem. The statistical model which
results from it is of type ARMA (Auto Regressive Moving Average), well adapted to
time series studies. This type of model is used in many fields (Young, 1998), but
still little applied to problems in civil engineering (Owen and Al, 2001). The same
approach can moreover be used to analyze the temperature or displacement
measurements in an arch dam (Bonelli and Felix, 2001).

148
7. References
Bonelli S., Félix H., Tourment R., 1998, Interprétation des mesures d'auscultation
des barrages par régression linéaire multiple HST, Proc. 2ème Conférence
Nationale Fiabilité des matériaux et des structures, Marne la Vallée, Hermès, pp
189-198.
Bonelli S., Royet P., 2001, Delayed response analysis of dam monitoring data,
Proc. Int. Symposium on Dam Safety, Geiranger, Balkema Rotterdam, pp 91-
100.
Bonelli S., Félix H., 2001, Delayed analysis of temperature effect, 6th ICOLD
Benchmark Workshop on Numerical Analysis of Dams, Salzburg, 6 p.
Bonelli S., 2004, Analyse retard des mesures d'auscultation de barrages, Revue
Française de Géotechnique, à paraître.
Carrère A., Colson M., Goguel B., Noret C., 2000, Modelling: a means of
assisting interpretation of readings. XXth International Congress on Large Dams,
Beijing, ICOLD, vol. III, pp. 1005-1037.
Crépon O., Lino M., 1999, An analytical approach to monitoring. International
Water Power & Dam Construction, June, pp. 52-54.
Dibiagio E., 2000, Monitoring of dams and their foundations, XXth International
Congress on Large Dams, Beijing, ICOLD, vol. III.
Diday E., Lemaire J., Pouget J., Testu F., 1982, Eléments d'analyse des
données, Dunod.
Ferry S., Willm G., 1958, Méthodes d'analyse et de surveillance des
déplacements observés par le moyen de pendules dans les barrages, VIth
International Congress on Large Dams, New-York, ICOLD, vol. II, pp.1179-1201.
Forbes J.D., 1846, Account of some experiments on the temperature of the earth
at different depths and in different soils near Edinburgh. Transactions of The
Royal Society of Edinburgh, vol. 16, pp. 189-236.
Guedes Q.M., Coelho P.S.M., 1985, Statistical behaviour model of dams. XVth
International Congress on Large Dams, Lausanne, ICOLD, vol. I, pp. 319-334.
Kjaernsli B., Kvale G., Lunde J., Baade-Mathiesen J., 1982, Design, construction,
control and performance of the Svartevann earth-rockfill dam, XIVth International
Congress on Large Dams, Rio de Janeiro, ICOLD, vol. IV, pp. 319-349.
Londe P., 1990, La sécurité des barrages, Revue Française de Géotechnique,
n°51, p. 41-49.
Lugiez F., Beaujoint N., Hardy X., 1970, L'auscultation des barrages en
exploitation au service de la production hydraulique d'Électricité de France, des
principes aux résultats, Xth International Congress on Large Dams, Montréal,
ICOLD, vol. III, pp. 577-600.
Marazio A., 1989, Monitoring of dams and their foundations, State of the art,
Bulletin n°68, ICOLD, 327 p.
Mary M., 1948, Research methods and instruments or measuring stresses an
deformations in earth and concrete dams, IIIrd International Congress on Large
Dams, Stockholm, ICOLD, vol. II.

149
Myrvoll F., Larsen S., Sande A., Romsol N.B., 1985, Field instrumentation and
performance observations for the Vatnedalsvatn dams, XVth International
Congress on Large Dams, Lausanne, ICOLD, vol. I, pp. 1039-1069.
Oberti G.,1964, Results and interpretation of measurements made on large dams
of all types, including earthquake observations, Q.29, VIIIth International
Congress on Large Dams, Edimbourg, ICOLD, vol. IV.
Owen J.S., Eccles B.J., Choo B.S., Woodings M.A., 2001, The application of
auto-regressive time series modelling for the time-frequency analysis of civil
engineering structures, Engineering Structures, n°23, pp. 521-536.
Plancke V., 1986, Auscultation des barrages, logiciel de traitement statistique des
mesures, mémoire ENITRS, Cemagref.
Post G., 1985, Dams and foundation monitoring, Q.56, XVth International
Congress on Large Dams, Lausanne, ICOLD, vol. I.
Poupart M., De Lustrac J., Bourgey P., and Bonelli S., 2000, Les enjeux
économiques de l'auscultation pour la maintenance des barrages, XXth
International Congress on Large Dams, Beijing, ICOLD, vol. III, pp. 1063-1073.
Silva Gomes A.F., Silva Matos D., 1985, Quantitative analysis of dam monitoring
results, state of the art, applications and prospects, XVth International Congress
on Large Dams, Lausanne, ICOLD, vol. I, pp. 749-761.
Willm G., Beaujoint N., 1967, Les méthodes de surveillance des barrages au
service de la production hydraulique d'Electricité de France, problèmes anciens
et solutions nouvelles, IXth International Congress on Large Dams, Istanbul,
ICOLD, vol. III, pp. 529-550.
Windisch E., Høeg K., 2000, Pore pressure in the till core of Oddatjorn dam, 53rd
Canadian Geotechnical Conference, Montreal, pp. 231-238.
Yoshida T., 1958, Observations des contraintes et des déformations dans les
barrages, dans leurs fondations et dans leurs appuis latéraux, comparaison de
ces observations avec les calculs et les essais sur modèles réduits, VIth
International Congress on Large Dams, New-York, ICOLD, vol. II.
Young P., 1998, Data-based mechanistic modelling of environmental, ecological,
economic and engineering systems, Environmental Modelling & Software, n°13,
pp. 105-122.

150