Estimation of structural parameters using static loading tests

E. Covián & M. Casero

Escuela Politécnica de Mieres, Universidad de Oviedo, Asturias, Spain
A. González
School of Civil, Structural and Environmental Engineering, University College Dublin, Dublin Co., Ireland

ABSTRACT: The increasing stock of aging infrastructure demands new convenient and efficient methods for
early damage detection. In this paper, focus is placed upon methods that use static measurements via modern
surveying techniques. Theoretical simulations play an important role in the development of damage detection
methods. These simulations use noise-free data or corrupt the ideal response with noise to resemble a field
scenario. However, the nature and true uncertainty of the measured structural response is commonly ignored.
Here, the uncertainty associated to modern surveying techniques employed for measuring deflections and ro-
tations is quantified and used to propose site-specific noise models. The latter are used to assess the accuracy
of a Cross-Entropy static damage detection algorithm.

(Sanayei & Scampoli, 1991; Hjelmstad & Shin,

1997; Bakhtiari-Nejad et al., 2005; Caddemi & Mo-
1 INTRODUCTION rassi, 2007, Walsh & Gonzalez 2009 and Yang &
Sun, 2010). Most of these methods use a finite ele-
An increasing number of large structures are ap- ment (FE) mathematical model as reference, and de-
proaching the end of their design life, while others flections and rotations as inputs. However, these in-
may have been built based on wrong assumptions puts need to be measured and “no measurement is
(i.e., traffic flows underestimated in bridges), which ever exact” (Ghilani & Wolf, 2006). In order to ad-
could represent an aging acceleration factor. Health dress the later, preliminary theoretical testing of new
monitoring of these structures is required to ensure algorithms may pollute the ideal structural responses
their safety. There is now technology to live-monitor with random errors in an attempt to simulate a more
characteristics of a structure such as acceleration, realistic scenario. However, these simulated errors
deflection, rotation, strain and temperature and to are not in agreement with their own nature, and their
provide an accurate image of the structure’s health. uncertainty for a particular confidence level is com-
However, this monitoring is often overlooked for monly ignored.
many structures due to cost and design/installation Although previous researchers (Marchamalo et al.
complications. Therefore, there is a need for a con- 2011; de Luis, 2009; Gordon & Lichti, 2007; Olalla,
venient and efficient method of monitoring, so that 2007; Psimoulis et al. 2006; Li et al. 2006; Boehler
damage can be detected sooner and propagation & Marbs, 2004; Mills & Barber, 2004 and Cheng et
minimized. This paper focuses on damage detection al. 2002) have investigated the accuracy of meas-
techniques based on static measurements, and how urements for many techniques, this accuracy has of-
the number of measured points (measurement densi- ten been expressed as nominal values or referred to
ty) and the uncertainty of these measurements affect observable measurands (as opposed to objective
the accuracy of these techniques. measurands), and as a result, the suggested values
For a given load, the static response is related to the are not comparable. In this paper, the uncertainty as-
boundary conditions and distribution of structural sociated to modern surveying instruments and tech-
stiffness throughout the structure. As damage can be niques, such as programmable and reflector-less to-
characterized as a local reduction in structural stiff- tal station and terrestrial laser-scanner is quantified.
ness, analysis of the structure’s static response has Then, noise models are proposed for simulating dif-
been used extensively to locate and quantify damage ferent field scenarios. Finally, the impact of meas-
urement uncertainty and measurement density on a the difference between computed and measured dis-
static damage detection algorithm is assessed using placements. In addition, a measured data perturba-
theoretical simulations corrupted with the noise tion scheme is presented, where estimated parame-
models. ters are considered as random variables and two
damage indices (calculated using the mean and
standard deviation of their distributions) are intro-
2 DAMAGE DETECTION METHODS duced.
Bakhtiari-Nejad et al. (2005) introduce an algorithm
2.1 General Overview based on the comparison of the load vectors of the
Damage detection methods can be classified in two damaged and undamaged structure. An error vector
main groups: those that subject the structure to static is formulated as the difference between both load
loading and those that use measurements from dy- vectors and then a minimization problem is stated
namic test data. A third category could be consid- subjected to two constraints: a non-linear equality
ered for those methods that combine both static and that reflects the measured change in displacement
dynamic test data. and an inequality ensuring negative change in stiff-
The main differences between both approaches are: ness. Moreover, they include selection criteria for
i) equations of dynamic equilibrium depend on load cases, baseline values and measurements loca-
mass, damping and stiffness matrix whereas equa- tions. The measurements error is found to have no
tions of static equilibrium depend solely on the stiff- effect on the algorithm performance. Nevertheless,
ness matrix; ii) High order of modal data, as well as no particular noise treatment is proposed. Caddemi
certain mode shapes, that are needed in some dy- & Morassi (2007) develop closed-form expressions
namic methods, can be difficult to obtain; iii) In stat- for the determination of the location and magnitude
ic methods, damage at certain locations of the struc- of cracks in beams, subjected to different boundary
ture can be concealed due to load paths associated conditions (pinned-pinned, pinned-clamped and
with the load applied. The latter reflects the im- clamped-free). Cracks are modelled as rotationary
portance of adequately choosing the applied load as springs of constant related to the severity of the
well as the need for a parameter that would help to damage. The accuracy of the estimated crack loca-
make that choice (Bakhtiari-Nejad et al. 2005; Yang tion is found to decrease significantly for low levels
& Sun, 2010). of damage.
A review on dynamic based methods can be found in Yang & Sun (2010) approach the problem of locat-
Yan et al. (2007), while an example of a method ing and quantifying damage in a decoupled fashion.
combining both static and dynamic data is provided The flexibility disassembly theory is used as the lo-
by Wang et al. (2001). This paper focuses on static cation algorithm, obtaining a damage location vector
based methods and how measurement uncertainty that identifies the damaged elements. Then, damage
can affect their performance. is quantified for those elements through a perturbed
stiffness parameter. Abdo (2011) conducts a para-
2.2 Damage Detection Methods based on static test metric study of the relationship between damage and
data displacement curvature. He states that damage repre-
Sanayei & Scampoli (1991) propose a method to sents a discontinuity in the curvature that can be lo-
calculate structural parameters from a subset of de- cated and quantified using Grey system theory to
grees of freedom, for which displacement measure- compare the curvature of damaged and undamaged
ments have been taken, by minimizing the error re- structures. Measurement noise is found to have a
sulting from the comparison between two stiffness negligible effect on the results.
matrix (analytical and measured). They analyse the Walsh & González (2009) apply cross-entropy (CE)
acceptable level of error in the measurements to to damage detection. They develop an iterative algo-
maintain the input/output error relationship in the rithm that creates a certain number of trial beams
linear range. Once established, they conduct a simu- and selects those with a deflection similar to the one
lation performing probabilistic parameter identifica- measured. Then, based on the statistics of the select-
tion with different subsets of parameters and affect- ed best trial beams, a new sample is generated and
ing measurements by the error previously calculated. the process is repeated until convergence is
Hjelmstad & Shin (1997) develop a static parameter achieved. Noise treatment and its influence on the
estimation algorithm that includes an adaptive pa- algorithm performance are yet to be explored. In a
rameter grouping scheme in order to deal with further section, the performance of this CE algo-
sparse data. Hjelmstad and Shin use the output error rithm is tested in the presence of field-like noisy
proposed by Banan et al. (1994), which is defined as measurements. Regarding the measurement errors,
the usual approach, when a numerical simulation is x = x ± Ux = x ± k ⋅σ x (1)
performed, is to consider an “artificial” random
noise to simulate real measurements. This noise is where x is the reference value, Ux is the uncertainty,
expressed as a percentage of the computed or analyt- σx is standard deviation of the measurand and k is a
ical value (examples in the literature range from coverage factor (k > 1) that depends on the selected
0,002% to 10%). However, it would be interesting to probability distribution. In accordance with the Bu-
choose the magnitude of the noise in relation to the reau International des Poids et Mesures recommen-
precision of the measurement device used in the dations (BIMP, 2008), 95% of confidence level (k =
field. This simulation will fulfil the purpose of test- 1.96) of the normal distribution is adopted here.
ing the algorithm in realistic conditions. Previous research has reported on the accuracy of
different measurement techniques. Some (Marcha-
malo et al. 2011; Olalla, 2007; and others) mention
3 ACCURACCY OF SURVEYING precision data as the nominal value given by the
TECHNIQUES manufacturers, in accordance or not with ISO rules,
and referred to observable but not to objective
3.1 Principles in deflection and rotation
measurand. Recent investigations on differential
measurements
GPS (Marchamalo et al. 2011; Cheng et al. 2002;
One of the goals of this paper is to determine if the Psimoulis et al. 2006; Li et al. 2006 and others) are
accuracy associated to a measurement technique is carried out in large scale structures (breakwaters,
sufficient to guarantee the correct performance of suspension bridges and skyscrapers) characterized
the CE static damage detection algorithm mentioned by being slender and flexible structures with large
in Section 2.2. Deflections and rotations are the in- displacements. While some research on HDS try to
puts used on this algorithm and reflector-less elec- determine precision through comparison with refer-
tronic tachymeter (TAC) and terrestrial laser-scanner ence values of point positions in laboratory (short
(TLS), commonly referred to as high definition sur- distances) (Gordon & Lichti, 2007 and Boehler &
vey (HDS), the measurement instruments selected. Marbs, 2004), others try to do it in real scenarios
Deflection and rotation are indirectly measured, (long distances) comparing results to other tech-
since observable measurands are acquired firstly and niques (de Luis, 2009 and Mills & Barber, 2004).
then a model is applied to calculate them. Generally, HDS real scenarios show precisions una-
As in every measurement, many error sources bring ble to capture deflections smaller than 15 mm accu-
differences between the true values and measures of rately.
a particular measurand (BIMP, 2012). The nature of
3.2 Surveying techniques to measure deflection
known error sources should be analysed and proce-
dures applied to reduce the error, in order to reach Both techniques —TAC and HDS— are based on
goodness of measures. These procedures are fo- the same observational measurands and relational
cused, firstly, on eliminating gross and systematic models in order to determine the deflections (ui) in
error components and, then, on minimizing random many points (i = 1, 2… n) distributed along a beam
error component in measurements (Covián & (Fig. 1). In Figure 1, A and B are the supports of the
Puente, 2013). beam under investigation (solid thick line represents
Here, the stages of instrument calibration, atmos- the initial position and solid dashed line represents
pheric corrections and, in the case of tachymeter, the lj loading case), E is the location of the meas-
Bessel compensation are supposed to be applied to urement instrument and ai is the height of the in-
reduce systematic errors. The most unfavourable E E E A A A
strument. x y z and x y z are the axes of the ref-
combination of random error sources is considered erence systems for the measurement instrument and
when quantifying, in pre-analysis, the accuracy of a for the beam analysis, respectively. Therefore, it is
given instrumentation (see Sections 3.2 and 3.3). necessary to apply a Helmert transformation to go
Pre-analysis error estimations are propagated in or- from measured data to structural data, in which the
der to calculate the final error of the objective meas- ordinate y becomes the same for all the points. The
urands —defections and rotations—. observational measurands are: geometrical distance
Finally the value of the measurand (x) should be ex- between the origin of the measurement instrument
pressed, in accordance with Guide to the expression and the viewed point (dg) and horizontal (θ) and zen-
of uncertainty in measurement (BIMP, 2008), show-
ithal (λ) angles. These measurands are considered
ing not only the reference value of the measurand 0
but also its uncertainty and associated confidence for the first position of every point ( Pi ) correspond-
level: ing to the initial condition of the structure and for a
 l
second position ( Pi j ) resulting from applying a Prior to the resolution of Equation 3, it is necessary
to establish the estimated errors of observable meas-
loading case ( ∀j = 1,2...m ).
urands: geometric distance error ( σ d ) and zenithal
While TAC is more accurate than HDS, the latter g

measures much more points (higher density) than angle error ( σ λ ), which can be calculated using
TAC. The next two equations show the relation be- Equations 4, 5:
tween observable measurands and the deflection
σ dg = a + b(ppm)⋅ dg (4)
(Eq. 2) and between the estimated errors of observa-
ble measurands and the error of the deflection (Eq.
(σ )
2
σλ = 3 + σ ang
2
(5)
3): comp

ui = dg0 ⋅ cos λ 0 − dgj ⋅ cos λ

l lj
(2) where a (usually expressed in mm) and b (in ppm),
determined in accordance with ISO 17123-3 (ISO,
( ) 2001), are the error component derived from the
2
σ ui = 2 ⋅ σ d2g ⋅ cos 2 λ 0 + σ λ2 ⋅ d g0 sin λ 0 (3)
phase-meter precision and the error derived from the
where dg0 , dgj , λ 0 and λ j are described above, and
l l instability of the wavelength of the EDM system re-
spectively (Covián & Puente, 2013); σcomp is the
σ ui , σ dg , σ λ are the estimated errors of deflections, precision of the vertical compensator and σang is the
geometrical distances and zenithal angles, respec- angular error component in accordance with ISO
tively. Equation 2 can be deduced from Figure 1, 17123-4 (ISO, 2001). This info is generally specified
and Equation 3 comes from the application of vari- in the equipment datasheets for k = 1 (68,26 % of
ance-covariance propagation law (Ghilani & Wolf, confidence level), so this is too for the geometric
2006). Equation 3 has been obtained considering no distance error ( σ dg ) and for zenithal angle error
0
variation between the first ( Pi ) and the second posi- ( σ λ ).
lj
tion ( P ) of every point, since the effect of this as-
i
sumption in deflection error ( σ ui ) is negligible.

Figure 1. Deflection measurement: reference systems, observable and objective measurands.

3.3 Surveying techniques to measure rotation

σ ρ = 2 ⋅ arctan
(
max σ xP , σ zP ) (6)
The determination of the rotations of each section of d′ / 2
the beam between two different load states (ρi),
through surveying measurements, require to measure where σ xP , σ zP are estimated errors of the abscise
the position of two reflectors, situated in the superior and applicate (z-coordinate) or height, respectively;
and inferior flanges of the beam, in both load states. and d’ is the distance between reflector centres.
Applying the same principles as in Section 3.2, the
estimated error of the rotations (σρ) is:
4 NUMERICAL SIMULATIONS ed symmetrically with respect to midspan (simulat-
ing 2-axle truck). Vertical deflections are simulated
4.1 Cross-Entropy algorithm for estimation of at every node of the FE model. Displacements from
structural stiffness from static measurements points within 2 m from the supports are not used as
Given the distribution of flexural stiffness (modulus inputs to the algorithm. These displacements near
of elasticity x inertia) of a beam of known boundary supports are ignored because they are small and of
conditions and fixed static loading conditions, its de- order of magnitude similar to the expected noise, in-
flection profile is unique. In the inverse problem creasing computational without necessarily improv-
consisting of a limited number of available dis- ing the results.
placements (possibly corrupted), a number of stiff- While HDS allows measuring a point every 1 mm
ness distributions may be able to reproduce these (100 elements/m), such a high number of elements
displacements. The CE algorithm relies on a FE increases the computational time in excess, and a
model to establish the global stiffness matrix of the measurement density of 50 displacements/m is
beam, the latter being function of the flexural stiff- adopted here for the HDS equipment. A measure-
ness of each discretized FE and its length. The CE ment density of 10 displacements/m is assumed to
method proposes an iterative procedure where a simulate data gathered by TAC equipment. Two dis-
number of FE trial beams (TBs) are generated based cretization levels are considered for the FE beam, 10
on stiffness values that are randomly sampled from a and 50 elements/m, to facilitate the implementation
predefined statistical distribution (i.e., normal) of of the algorithm when providing TAC and HDS
mean and standard deviation updated in each itera- measurements respectively. In addition, the same
tion. Thus, a different set of TBs with different stiff- stiffness value is assigned to all the elements con-
ness profiles and, consequently, different deflection tained within one meter to speed up calculations, in-
profiles, is obtained at each iteration of the algo- stead of considering an individualized value for each
rithm. In particular, a total of 1000 TBs are generat- discretized element.
ed in each iteration in this paper. The deflection of Four case scenarios are considered for both TAC
each TB is compared against that of the true beam and HDS data: (i) noise-free measurements from
and those TBs performing best (lowest mean square healthy beam, (ii) noisy measurements from healthy
error of the deflections at measured points) are se- beam, (iii) noise-free measurements from damaged
lected to update statistical distributions associated to beam, and (iv) noisy measurements from damaged
each discretized beam element for the next iteration. beam. Damage is modelled as a 40% stiffness reduc-
In this paper, the best 10% of all TBs are retained. tion between 12 m and 13 m. As such a sudden drop
The process is repeated until the rate of change of in stiffness is not likely in a real situation, a 10%
error of the objective function (sum of the squared stiffness reduction is applied in adjacent meters to
differences between target/measured deflections and smooth the stiffness profile.
predicted deflections using FE model for each ob- 4.3 Noise model
servation point) falls below 0.01% over 10 itera-
tions. Then, injection (artificial widening of the sta- The nominal accuracies of the selected instruments
tistical distribution) is applied to prevent the (table 1) that have any influence in the deflection er-
algorithm falling into local minima. This injection is rors are:
repeated for three convergences of the algorithm. Table 1. Nominal accuracies for Leica TCRP1201 tachymeter
Values for the initial statistical distributions of flex- and Leica ScanStation2 laser-scanner.
ural stiffness for each discretized FE are random, TAC HDS
(Leica TCRP1201) (Leica ScanStation2)
although the mean is selected to be within expected distance accuracy 1 2

theoretical values and the standard deviation is set at (σdg)

2 mm + 2 ppm · d g 4 mm
20% of the mean value. Further details can be found angular accuracy 1
1" 60 µrad = 12,38"
in Walsh & González (2009). (σang)
compensator accuracy
0.5" 1"
4.2 Theoretical scenarios for testing (σcomp)
All the accuracies correspond to one sigma (k = 1).
A simply supported beam, of span length 21 m, is 1
According with ISO 17123-3 and ISO17123-4.
used to test the performance of the algorithm. A 2
Out of rules and at a 0 - 50 m range.
flexural stiffness of 3450 MN·m2, typical of a 21 m Maximum deflection errors, estimated in the pre-
bridge, is adopted. In all cases, the applied loads analysis (see Section 3.1), at a 95% of confidence
consist of a uniformly distributed load (equivalent to level, are shown in table 2 for TAC and HDS fol-
the self-weight of the bridge deck) and two concen- lowing application of Equations 3, 4, 5. They are
trated loads of 10 t each separated by 5 m and locat- given for combinations of horizontal distances and
heights of the target points with respect to the meas- curve to the target profile in the figure. Figures 2a, b
urement instrument. For the theoretical testing in the are obtained using measurements every 10 cm along
section that follows, a 6 m high bridge is assumed beam and Figures 2c, d are obtained using measure-
(relevant values are in bold in table 2). It is also as- ments every 2 cm (excluding 2 m at each end sup-
sumed that the station is located at an optimal dis- port).
tance that minimizes the quadratic sum of the deflec-
tion errors along the beam. Finally, the theoretical
deflections that result from the beam model in Sec-
tion 4.2 are contaminated using an additive noise
model (i.e., adding a value randomly sampled from a
statistical distribution modelling noise to the true de-
flection). The noise model is given by a normal dis-
tribution of mean zero and a standard deviation that
depends on the geometrical distance from each
measurement point to the station.

Table 2. Deflection errors for Leica TCRP1201 tachymeter and

Leica ScanStation2 laser-scanner in many probable observation
conditions.
d HZ (horizontal distance in m)
σui, TAC, 95 % (mm)
15,00 21,00 30,00
0,00 0,289 0,405 0,578
h (height in m)

3,00 1,588 1,202 0,990

6,00 2,973 2,238 1,687
9,00 4,114 3,185 2,394
12,00 5,000 4,004 3,061
15,00 5,669 4,692 3,669
horizontal distance (m)
σui, HDS, 95 % (mm)
15,00 21,00 30,00
0,00 3,529 4,941 7,058
3,00 4,681 5,416 7,229
height (m)

6,00 6,809 6,555 7,699

9,00 8,805 7,910 8,374
12,00 10,411 9,216 9,151
15,00 11,635 10,367 9,949

5 THEORETICAL TESTING

5.1 Noise-free simulations Figure 2. Target and estimated stiffness profiles using noise-
free deflection measurements: (a) 10 measurements/m in a
When deflection measurements are noise-free, the healthy beam, (b) 10 measurements/m in a damaged beam, (c)
50 measurements/m in a healthy beam, (d) 50 measurements/m
CE algorithm is reliable in both no damage and in a damaged beam.
damage situations for the load and structural model
under investigation (Section 4.2). The estimated An increase in measurement density from 10 (Fig.
stiffness profiles are consistent with the expected 2a) to 50 measurements/m (Fig. 2c) improves the es-
targets and able to capture the localised loss of stiff- timation of stiffness distribution in a healthy beam
ness (Fig. 2). The estimated peak loss of stiffness is of uniform stiffness (i.e., leading to a smoother and
relatively smaller and spread over a slightly wider closer profile to the target). However, no significant
length than the target value. However, the overall differences appear in the stiffness estimation of the
loss of stiffness and critical location where it occurs, damaged beam when using 10 (Fig. 2b) or 50 meas-
are correctly identified. urements/m (Fig. 2d). As it has been previously no-
The difference in stiffness profiles shown in Figure ticed (Walsh & Gonzalez, 2009; Gonzalez et al.
2 for different runs of the CE algorithm (i.e., test1, 2013), stiffness predictions for the elements closest
test2, test3, test4 and test5 in the figure) is a result of to the supports are unreliable due to the low sensitiv-
the random nature of the sampling through each iter- ity of deflection to the stiffness of these areas.
ation (Section 4.1). The uncertainty associated to the Measured rotations are included as additional inputs
nature of the algorithm can be reduced by using an to the CE algorithm to check their impact on the es-
average of a number of CE tests as shown in the fig- timated stiffness (Fig. 3). A small improvement is
ure. Therefore, the ‘average’ curve is the closest noticeable in the results of the healthy beam, i.e., the
profile in Figure 3a being smoother and closer to the it acceptable for accurate stiffness prediction using
target value than in Figure 2a. CE in a structure of characteristics and loading simi-
lar to the one described in Section 4.2.
A simplified noise model, where all points are at-
tributed the same maximum measurement error is
adopted. Figure 5 illustrates the ability of the CE al-
gorithm to predict damage for various deflection er-
rors. The noise limit to identify the stiffness profile
with reasonable degree of accuracy varies in a
threshold between 0.4 mm and 0.8 mm. Beyond 0.8
mm, damage is not unambiguously identified.

Figure 3. Estimated stiffness profiles using 10 measurements/m

including displacements and rotations: (a) healthy beam, (b)
damaged beam. Figure 5. Target and estimated stiffness profiles (average of 5
5.2 Noisy simulations simulations) using corrupted deflection measurements for vari-
ous deflection errors and 10 measurements/m in a damaged
Noise is taking into account following the modelling beam.
procedure of Section 4.3 and it depends on the loca-
tion and precision of the equipment being simulated It must be noticed that the noise limit previously
(TAC or HDS). recommended has been determined for a specific
When simulated TAC measurements are considered structure (geometry, boundary conditions and ele-
(10 measurements/m), the CE algorithm generally ment mechanical properties), loading conditions and
shows that the stiffness drops close to damaged ele- set of CE parameters. This noise limit may clearly
ments (between 12 and 13 m) as shown by Figure 4, be increased for more flexible structures (or larger
but drops can also be found in healthy elements. loads) leading to higher displacements or decreased
Therefore it could be said that the predictions by the for stiffer structures (or smaller loads) leading to
CE algorithm are inconclusive for this particular smaller displacements.
structure and loading.
6 CONCLUSIONS

The interest of using surveying measurements for

structural assessment as opposed to other sensors
that need direct contact with to the structure is be-
yond any doubt. This paper has used simulated de-
flection and rotation surveying measurements to as-
sess the accuracy of a CE algorithm in estimating the
Figure 4. Target and estimated stiffness profiles using simulat-
ed TAC measurements from damaged beam. stiffness profile of a structure. For this purpose,
sources of error related to the equipment (TAC and
Simulated HDS measurements (50 measurements/m) HDS) used to collect static measurements have been
leads to higher deviations with respect to the target quantified. For the selected structure (21 m simply
stiffness profiles than TAC measurements as a con- supported bridge) and loading case (self-weight + 2-
sequence of higher noise associated to this measur- axle truck), the CE algorithm has shown to be relia-
ing device. ble when applied to noise-free data, using measure-
If simulated noisy rotations are included together ment densities of 10 (TAC) and 50 (HDS) points per
with displacements in the CE calculations, there is meter. However, the corruption of the displace-
not an improvement in the estimated profiles with ments by noise associated to TAC or HDS meas-
respect to Figure 4. urements has prevented the accurate estimation of
Finally, several different values of deflection errors the stiffness profiles by the CE algorithm.
are tested, with the aim of establishing the noise lim-
The noise limit that would allow reliable stiffness Cheng Penggen, Shi Wenzhong & Zheng Wanxing. 2002.
Large Structure Health Dynamic Monitoring Using GPS
profiles and to detect damage through the CE algo- Technology. Engineering Surveys for Construction Works
rithm is not achievable under the test conditions in and Structural Engineering FIG XXII International Con-
this paper. Further developments in surveying tech- gress; Washington, DC (USA), 19-26 April 2002.
niques appear to be necessary to achieve an adequate Covián, E. & Puente, V. 2013. Fundamentos del ajuste de ob-
servaciones topográficas. Oviedo (Spain): Publication Of-
prediction of structural parameters using CE and fice of the Oviedo University.
static measurements as inputs. It should be noted Ghiliani, P. & Wolf, R. 2006. Adjustment computations. Spa-
that accuracy of deflection and rotation measure- tial data analysis. New Jersey (USA): John Wiley & Sons.
ments might be higher in a real case than in the pre- González, A., Covián, E., Casero, M. & Cooper, J. 2013. Ex-
perimental Testing of a Cross-Entropy Algorithm to Detect
liminary analysis carried out in this paper. Noise can Damage. 10th International Conference on Damage As-
be reduced through the use of higher precision ta- sessment of Structures. Dublin (Ireland), 8-10 July 2013.
chymeters or digital levels, or through the applica- Gordon, S.J. & Lichti, D. 2007. Modeling Terrestrial Laser
tion of sophisticated filtering algorithms to the HDS Scanner Data for Precise Structural Deformation Measure-
ment. Journal of Surveying Engineering 133-2: 72-80.
measurements. Therefore, more flexible structures or Hjelmstad, K. & Shin, S. 1997. Damage detection and assess-
higher loads will lead to larger displacements where ment of structures from static response. Journal of Engi-
noise will affect stiffness estimations to a lesser ex- neering Mechanics 123(6): 568-576.
tent. International Organization for Standardization (ISO). 2001.
ISO 17123, Optics and optical instruments — Field proce-
dures for testing geodetic and surveying instruments.
Geneve (Switzerland): ISO copyright office.
ACKNOWLEDGMENTS Li, X. Ge, L., Ambikairajah, E., Rizos, C., Tamura, Y. & Yo-
shida, A. 2006. Full-scale structural monitoring using an in-
The authors wish to express their gratitude for the tegrated GPS and accelerometer system. GPS Solutions 10:
233–247.
support received for Scientific Research under Grant Luis, de, J.M. 2009. Contraste en la ejecución de auscultacio-
No. BIA2011-26915, from the “Programa Nacional nes geodésicas por métodos clásicos y con láser escáner.
de Proyectos de Investigación Fundamental, VI Plan Tesis doctoral, Departamento de Ingeniería Geográfica y
Nacional de Investigación Científica, Desarrollo e Expresión Gráfica, Cantabria University, Santander (Spain)
Marchamalo, M., Galán, D., Sánchez, J.A. & Martínez, R.
Innovación Tecnológica 2008-2011” of the Spanish 2011. La tecnología DGPS en la construcción: control de
Government towards this investigation. movimientos en grandes estructuras. Informes de la Con-
strucción 63-522: 93-102.
Mills, J. & Barber, D. 2004. Geomatics Techniques for Struc-
tural Surveying. Journal of Surveying Engineering 130-2:
REFERENCES 56-64.
Olalla Marañón, C. 2007. Auscultación de laderas. Jornadas
Abdo, M. 2011. Parametric study from static response in struc- técnicas sobre estabilidad de laderas en embalses, Confe-
tural damage detection. Engineering Structures 34: 124- deración Hidrográfica del Ebro, Zaragoza (Spain) June
131. 2007.
Bakhtiari-Nejad, F., Rahai, A. & Esfandiari, A. 2005. A struc- Psimoulis, P.A., Pytharouli, P. & Stiros, S. 2006. Experimental
tural damage detection method using static noisy data. En- monitoring of oscillations of major flexible structures using
gineering Structures 27(12): 1784-1793. GPS and RTS. 3rd IAG / 12th FIG Symposium. Baden (Aus-
Banan, M. R., Banan, M. R., and Hjelmstad, K. D. 1994a. Pa- tria) 22-24 May 2006.
rameter estimation of structures from static response. Part I: Sanayei, M. & Scampoli, S. 1991. Structural element stiffness
computational aspects. Journal of Structural Engineering identification from static test data. Journal of Engineering
120(11): 3243-3258. Mechanics 117(5): 1021-1036.
Banan, M. R., Banan, M. R., and Hjelmstad, K. D. 1994b. Pa- Walsh, B.J. & Gonzalez, A. 2009. Assessment of the condition
rameter estimation of structures from static response. Part of a beam using a static loading test. Key Engineering Ma-
II: numerical simulation studies. Journal of Structural En- terials 413-414: 269-276.
gineering 120(11): 3259-3283. Wang, X., Hu, N., Hisao Fukumaga & Yao, Z.H. 2001. Struc-
BIMP (Bureau International des Poids et Mesures). 2008. tural damage identification using static test data and chan-
Evaluation of measurement data - Guide to the expression ges in frequencies. Engineering Structures 23: 610-621.
of uncertainty in measurement. Paris (France): Joint Com- Yan, Y.J., Cheng, L., Wu, Z. Y. & Yam, L.H. 2007. Develop-
mittee for Guides in Metrology. ment in vibration-based structural damage detection techni-
BIMP (Bureau International des Poids et Mesures). 2012. In- que. Mechanical Systems and Signal Processing 21: 2198-
ternational vocabulary of metrology – Basic and general 2211.
concepts and associated terms (VIM). 3rd edition. Paris Yang, Q. & Sun, B. 2010. Structural damage localization and
(France): Joint Committee for Guides in Metrology. quantification using static test data. Structural Health Mo-
Boehler, W. & Marbs, A. 2004. Investigating Laser Scanner nitoring 10(4): 381-389.
Accuracy. i3mainz, Institute for Spatial Information and
Surveying Technology, FH Mainz, University of Applied
Sciences, Mainz, Germany. On line: http://scanning.fh-
mainz.de.
Caddemi, S. & Morassi, A. 2007. Crack detection in elastic
beams by static measurements. International Journal of
Solids and Structures 44: 5301-5315.