Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Special Issue Article

Structural Health Monitoring


12(5-6) 411–429

Monitoring of a civil structure’s state Ó The Author(s) 2013


Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
based on noncontact measurements DOI: 10.1177/1475921713487397
shm.sagepub.com

Piotr Kohut1, Krzysztof Holak1, Tadeusz Uhl1, Łukasz Ortyl2, Tomasz


Owerko2, Przemysław Kuras2 and Rafał Kocierz2

Abstract
In this article, the comparison of two noncontact measurement methods dedicated to civil engineering structures’ state
examination is presented. The vision-based method computes the displacement field of the analyzed structure by means
of the digital image correlation coefficient. The system consists of one or more high-resolution digital cameras mounted
on a head or on portable tripods. The developed methodology and created software application embedded in an MS-
Windows operating system are presented. The second system measures the deflection of the structures by means of a
radar interferometer. In both cases, it is possible to measure many points on the structure simultaneously. This article
presents a comparison of the displacement field measurement performed on a field setup, as well as the span of a steel
bridge designed for tram traffic. Both systems are described, with special attention given to their application in measure-
ments of civil engineering structures. This article demonstrates a preliminary test performed to verify both of the non-
contact systems in relation to high-accuracy measurement devices, the precise surveying level, and the electronic dial
indicator of displacement. The experiment was designed intentionally to simulate the geometric conditions of the real
structure, but the displacement values were generated and controlled by the operator. As a key study, a steel viaduct
subjected to an operational load was measured, as a type of structure for which the observation is required in terms of
structural health monitoring. It was subjected to the operational load caused by tram traffic. Both examined systems
were applied. The accuracy analysis of both systems was investigated, and the obtained results were discussed.

Keywords
Vision systems, digital image correlation, non-contact measurement systems, deflection measurement, radar interfero-
metry, bridge state monitoring

Introduction the static or dynamic test loads. This can be performed


for new structures before putting them into operation,
Structural health monitoring (SHM) involves the inte- as well as for previously used structures in order to
gration of sensors, data transmission, processing, and determine the current structure condition, for example,
analysis in order to detect, localize, and assess damage after repair. However, tests during operation are often
within a structure, which can lead to its failure at the conducted with randomly varying loads and may be a
present time or in the future.1 SHM methods can be useful basis for experimental modal analysis.4,5
divided into two groups: local and global methods. The In order to assess the correctness of their work and
latter are applied if a global change in the geometry of condition, structures are submitted to static loads.6 The
a structure can be observed. The analysis of the displa- results of regularly repeated static tests allow the
cement field or mode shapes plays a key role in the eva-
luation of constructions’ static states and dynamic 1
Department of Robotics and Mechatronics, AGH University of Science
characteristics. and Technology, Krakow, Poland
In bridge engineering, diagnostic tests are complex 2
Department of Engineering Surveying and Civil Engineering, AGH
processes that use broad knowledge of the considered University of Science and Technology, Krakow, Poland
object. Among the various sources of information on
the structure state, in addition to the technical and Corresponding author:
Piotr Kohut, Department of Robotics and Mechatronics, AGH University
operational documentation and structure inspections, a of Science and Technology, Mickiewicza 30 Avenue, 30059 Krakow,
particularly important role is played by research under Poland.
load.2,3 Analysis of structure response may be related to Email: pko@agh.edu.pl

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


412 Structural Health Monitoring 12(5-6)

comprehensive evaluation of changes, which can indi- displacement calculation.13 Therefore, it is recom-
cate the appearance of damage, often difficult to reveal mended to avoid the integration process and to use the
with other methods. A separate problem is the analysis appropriate device for direct measurement of the
of bridge response under dynamic loads. The detection required quantity.14
of changes in the dynamic characteristics of structures A significant drawback of all the above-mentioned
may be used to discover bridge defects, including defor- methods is the necessity of fixing sensors or additional
mation, destruction, or loss of material continuity. devices directly to the examined parts of objects. In
Each bridge structure investigated under operational many practical cases, it may be very difficult or even
loads demands individually designed procedures of reg- impossible on account of the working conditions.
ular tests to investigate its condition. The concepts of Moreover, they provide only discrete information of
monitoring systems must take into consideration a few structure displacement. To get full information about
items: the structure type, operational conditions, and the position of the structure, it is necessary to install a
the rate of degradation processes. The techniques that number of sensors or make measurements at many
are applied to determine the response of the loaded points. Therefore, there is a necessity for the design and
structure are very diverse. implementation of noncontact measurement systems.
Nowadays, a wide variety of measurement tech- Two noninvasive techniques, which allow the displace-
niques, which are based on displacement measures, are ment measurements at many points simultaneously, are
applied in the monitoring of bridge deformations and described in detail below.
vibrations.7 Among them are surveying techniques: glo- The first technique is radar interferometry. It is
bal positioning system (GPS) or global navigation sat- widely used for imaging changes in the earth’s surface
ellite system (GNSS) and electro-optical instruments. by satellites.15 Recently, it has also been used in
The satellite technique is particularly useful in the ground-based imaging. It can provide information
dynamic measurements of long-span bridges, which are about the movements of surfaces, such as slopes or
submitted to displacements of several centimeters.8,9 landslides,16 and also enables the static and dynamic
However, the accuracy of about 61 cm in the horizon- tests of engineering structures.
tal direction (and 62 cm in the vertical direction) may The ground-based synthetic radar aperture
be disturbed by some effects. An error of a few centi- (GBSAR) technique has been developed since 2000.17
meters may be caused by multiple signal reflections Its main applications are land surface topography ima-
from construction elements that cause the multipath ging, and land deformation monitoring. The radar used
effect, the improper modeling of ionospheric and tro- in this technique performs the measurement by sending
pospheric delay, and the unfavorable configuration and a modulated wave and analyzing the returning signal
visibility of satellites. On the contrary, measurements reflected from the object. The radial component of the
made by total stations or digital levels are not disturbed object displacement is observed as a phase change of
with these effects and enable deflection measurements the returning signal.
with an accuracy better than 0.2 mm.10,11 Nevertheless, The first implementation of the GBSAR technique
this technique cannot be applied in bad weather condi- was the linear synthetic aperture (LISA) radar, used for
tions that limit visibility or to perform the dynamic measurement of static bridge deflection or deformation
measurements because of the low sampling rate. of building facxades, among others.17,18 The concept of
Other devices, used particularly in dynamic measure- using the ground-based radar to study the dynamics of
ments, are electronic sensors such as accelerometers, structures was proposed by Pieraccini et al.19 in 2004.
displacement transducers (e.g. linear variable differen- In further research, the appropriate frequency of the
tial transformer (LVDT) and laser beam), and strain carrier wave was investigated. The higher the frequency,
gauges.12 These devices have to be mounted directly on the higher the sensitivity of the displacement measure-
the structure and require connection with reference ment, but the lower the range arising from the phase
points to determine the absolute displacements.11 In ambiguity. The commercial version of the system,
the case of displacement transducers, a fixed base is named ‘‘Image by Interferometric Survey’’ (IBIS),
necessary to measure relative displacements. On the implemented in 2007, adopts the center frequency of
contrary, the accuracy of data obtained with acceler- 17.2 GHz (Ku-band of microwaves) as a compromise.
ometers depends strongly on sampling rate, internal This means that the maximum displacement, unam-
noise of a device, sample length, and frequency content biguously determinable, is equal to 64.36 mm (61/4 of
of the signal. In order to determine the displacement the wavelength). Greater values require methods for
values, a double integration of acceleration data must removing the ambiguity phase.20 Due to its features, the
be calculated. This process, however, requires the use IBIS system may be applied in static and dynamic mon-
of digital filtering and the definition of the initial condi- itoring of structures, especially of elongated shape,
tions. Their improper assumption may lead to errors in including bridges and towers.21,22

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 413

The second technique is digital image correlation, standard deviation (SD) of the absolute difference
applied in vision-based systems (VBs). The vision sys- between the displacements measured by a vision system
tems can be a good alternative to contact type transdu- and a network of tensometers was 0.31 mm. In a previ-
cers23–26 as easy to use, accurate, low cost, and ous article by Tung et al.,38 the method of crack detec-
universal tool systems, which can be applied in vibra- tion in masonry walls was presented. The deflection
tion and deformation measurements. Nowadays, an measurement was carried out by means of a digital
increase in the availability of vision systems using digi- image correlation coefficient. In the experimental tests,
tal image correlation for the measurement of displace- the wall was subjected to compressive stresses. The
ments, strains, and stresses of objects has been initiation and growth of the cracks were observed. The
observed on the world markets. The most important cracks were detected as places in which the correlation
companies offering vision measurement systems exist- coefficient decreased its value abruptly. Kujawińska
ing at present on the market are Correlated Solution, et al.39 proposed a hybrid, distributed system dedicated
GOM, LIMESS Messtechnik & Software, Dantec to construction monitoring. The images were acquired
Dynamics, and Metris (Krypton 9000).27–34 The men- by two types of optical sensors: one- or two-camera sys-
tioned commercially available vision systems are not tems using digital image correlation as a measurement
dedicated to large-scale civil engineering structure mon- technique and an infrared camera. The fusion of data
itoring. They can perform only manual measurements from two types of sensors allowed the efficiency of the
and have to be operated by a skilled user. They cannot structure’s monitoring to be increased. Part of the mea-
be easily used for fully automatic long-term monitoring surement system, using cameras and the digital image
of the structure. Some of the systems belong to active correlation method, was applied in the monitoring of
vision devices that involve additional sources of radia- the elements of civil engineering constructions, and in
tion, making them expensive. However, the passive the art gallery for paintings’ state observation.40 Canon
devices require special geometric markers or random EOS 5D Mark II cameras were applied as image acqui-
texture and cannot carry out measurements without sition devices. Kujawińska et al.41 proposed an intelli-
the object’s preparation. Researchers have proposed gent, hybrid vision system consisting of optical sensors
many solutions for the systems dedicated to civil engi- and a server controlling the operation of all measure-
neering state evaluation. ment devices. The first sensor is an active device and
In a previous article by Olaszek,35 the method of consists of a structured light projector and charge-
geometric marker’s displacement tracking is presented. coupled device (CCD) camera. The device reconstructs
The proposed method was applied for the measurement the geometry of objects in the form of a cloud of points.
of a bridge’s vibrations and displacements under static The second device makes use of the Moire Pattern anal-
and dynamic loads. The subpixel technique of tracking ysis. The pattern, consisting of parallel lines, is placed
was introduced. An additional reference point was on the flat surface of the structure. When the construc-
placed on the bridge’s structural element in order to tion is under the load, it deforms along with the pattern.
correct errors induced by the vibration of the camera. The reference pattern is superimposed on the image of
The marker had the shape of a circle, and its centroid the deformed pattern. The interference phenomenon
was tracked during investigation. A different monitor- creates the Moire Pattern, carrying information about
ing system for bridges was presented in a previous arti- the deformation. Devices of the system were used for
cle by Lee and Shinozuka.36 The system was applied for monitoring of the structural elements of the Temple of
train bridge state examination. The 1 MPixPointGrey Divine Providence in Poland. An SHM system for
Fleas2 camera was used for observation of the geo- bridge monitoring was presented in the study by
metric marker’s displacement. The positions of the Zaurin.42 The system consists of a digital camera and a
tracked marker’s centroid generated the trajectory of set of strain sensors. The vision part of the system is
the measurement point of the bridge, which was sub- used for monitoring the traffic on the bridge by detect-
jected to dynamic loads. The second device, a Canon S7 ing, tracking, and classifying vehicles on the bridge. The
camera with a lens with a focal length equal to f = 140 contact sensors measure the strains and displacements
mm, was applied to observe the girder’s deflection. of the construction’s elements. The registered data are
Yoneyama et al.37 proposed a method of bridge deflec- correlated with images from the cameras. The SHM
tion measurement using digital image correlation. The system detects abnormal states, such as exceeding of the
camera’s position was fixed. The random speckle pat- strain thresholds.
tern was placed on the analyzed plane of the structure. In this article, the developed vision system and
The examination of the structure’s deflection was car- image processing methods dedicated to in-plane mea-
ried out during a loading test. The load of 20 tons was surement of civil engineering structure displacement
provided by a truck. The authors reported that the fields are presented. The developed vision system

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


414 Structural Health Monitoring 12(5-6)

Figure 1. Ability of observing many points (range bins) on the tested structure according to the radar range resolution.
SNR: signal-to-noise ratio.

presented is characterized by many novel features with Ground-based RI


respect to the existing systems. First, it is dedicated to
The IBIS system was developed by the Italian company
large-scale civil engineering structures, which have two
IDS in order to monitor the movements of land masses
basic modes of operations. It can be used as a manual
and engineering structures. The IBIS-S version is
measurement device, like the other commercially avail-
applied to measure the displacements of buildings, of
able systems, by a skilled user. On the contrary, it can
which one dimension is significantly larger than others,
be part of a fully automatic measurement system used
that is, tall buildings, towers, or bridge structures.
for long-term monitoring of the structure. The method
The system is based on the Ku-band radar, which
makes it possible to acquire the images of the construc-
generates, transmits, and receives electromagnetic
tion from two distinct points in space, which makes the
waves of 17.2 GHz central frequency. Basing on the sig-
use of costly positioning devices unnecessary.
nals reflected from elements of a tested structure, their
Moreover, if the analyzed structure has a highly ran-
displacements can be determined. The most important
dom texture, the method can be applied even without a
abilities of the system are the remote measurement of
special set of markers. The system can operate with
range up to 1000 m and sampling rate up to 200 Hz.21
inexpensive single-lens reflex (SLR) digital cameras,
The direct access to the structure is not necessary, and
which reduces the cost of the system’s installation on
many points may be observed simultaneously. The
the object even further.
detailed description of the applied radar techniques can
This article presents the results of the preliminary
experiment performed on an anti-flood embankment, be found in the previous studies.15–22 The most impor-
the field test on a tram viaduct deflected under the tant features, necessary to explain tests presented in this
weight of passing vehicles and also the comparison of article, are summarized below.
the data acquired by the radar interferometer (RI), and The bandwidth used by the radar is B = 200 MHz
the developed VB. The preliminary tests were per- (from 17.1 to 17.3 GHz). In special cases, it can be
formed to verify the mutual compatibility of methods expanded to 300 MHz. This allows the maximum range
in relation to the reference methods. As the key study, resolution DR = 0.5 m to be obtained. The concept of
the tram viaduct under load conditions was investigated range resolution shall be understood as the minimum
in order to verify the usefulness and the compatibility separation between two points on the structure at which
of the deflections determined by both systems. The they may be observed as different points. This means
obtained results were submitted to statistical assessment the ability to observe points on the structure separated
in order to evaluate the correlation and accuracy of by not less than 50 cm along the radial direction, that
methods. For the preliminary test, the SD of differences is, direction of wave propagation (Figure 1). If the dis-
and the Pearson correlation coefficient were calculated. tance between reflecting points is smaller than DR, they
In the main experiment, the Bland–Altman test was are treated as one point. The single interval of DR is
employed. called a range bin. Selection of range bins for further

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 415

Figure 2. Relation between radial and vertical displacements.

analysis is carried out based on the range profile. This


is a graph that shows the echo intensity, expressed in
decibel as the signal-to-noise ratio (SNR), in relation to Figure 3. Relation between frequency and time in SFCW
modulation.
the distance to the observed point. SFCW: stepped-frequency continuous wave.
Two radar techniques are implemented in the IBIS-S
system: microwave interferometry and stepped-
structure at the same time. The radar wave, reaching
frequency continuous wave (SFCW) modulation. The
the observed structure, is scattered on the construction
first one allows high accuracy of displacement measure-
elements. Subsequently reflected signals are subjected
ment to be obtained. Displacement of the observed
to the appropriate processing to obtain the displace-
point in the direction of wave propagation causes the
ment values. In order to increase the reflectivity of a
phase difference between signals received by the radar
structure or to point at selected places, it is possible to
at different times. The value of radial displacement dR
use microwave reflectors installed on the structure.
can be written as
This requires, however, direct access to the structure
and may be difficult in practice, especially for high
l  Du
dR = ð1Þ tower structures.
4p The IBIS radar transmits signals in the form of short
where l is the wavelength and Du is the phase pulses. The relation between pulse duration t and the
difference. used microwave bandwidth B can be written as
An important fact is that the IBIS-S is a one- tB = 1. According to equation (2), shorter pulse dura-
dimensional (1D) system. The high accuracy of the dis- tion t means the higher range resolution DR can be
placement measurement is related to the radial direc- obtained:
tion (radar–object). If the movement occurs in the
ct c
other direction, an assumption is required to calculate DR = = ð2Þ
the real displacement value. For example, to calculate 2 2B
the dV value from the dR measured, the H and D dis- where c is the speed of light.
tances have to be known, for example, using surveying Improving the range resolution DR value is reached
methods, and only the vertical movement of the by reducing the value of t or increasing the value of B.
observed point must be assumed. To calculate the dV SFCW radars, instead of using short pulses, achieve
value, dR must be multiplied by the projection factor p, the wide bandwidth through the stepped, linear
which is p = D/H (Figure 2). increase of discrete frequency Df values. Bandwidth can
It should be noted that the accuracy of displacement be expressed as
measurement refers to the radial (line-of-sight) direc-
tion. The conversion causes a decrease in the accuracy B = ðN  1Þ  Df ð3Þ
of the determined real displacements. This is particu-
larly unfavorable for low, but long-span bridges, for where N is the number of discrete frequency values
which often p . 4.43 The most preferred ratio p = 1, within the bandwidth B (Figure 3).
obtained by the accordance of dV and dR vectors, is dif- Excluding the impact of changing weather condi-
ficult to achieve in the field, especially for bridges above tions and the effect of signal multipath, the radial dis-
rivers, and practically eliminates the possibility of placement measurement accuracy is 0.1 mm in typical
observing a lot of points at the same time. conditions.19 However, the low SNR value, occurring
The second technique is SFCW modulation, which for structures that weakly reflect the radar wave,
allows the measurement of many points on the decreases the accuracy of displacement determination.20

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


416 Structural Health Monitoring 12(5-6)

Figure 4. Digital image correlation applied to the deflection measurement. The displacement of the intensity pattern is found by
ZNCC coefficient. The method applied to all span of the beam results in the deflection curve.
ZNCC: zero-mean normalized cross-correlation coefficient.

2 3
Vision-based in-plane deflection fsx su u0
measurement method K=4 0 fsy v0 5 ð4Þ
0 0 1
The developed vision-based method of the in-plane
deflection measurement consists of the following steps: where f is the focal length of the lens; u0, v0 are coordi-
system calibration, image acquisition, and deflection nates of the principal point; sx, sy are the size of the
measurement using the digital image correlation coeffi- pixel along horizontal and vertical directions, respec-
cient (Figure 4). Additionally, perspective distortion tively; and su is a skew coefficient of the pixels. The
removal can be carried out if the images of the struc- chessboard planar pattern with black and white
ture are captured from different points in space. The squares, even number of rows and odd number of col-
system is calibrated by calculation of the scale coeffi- umns, is used. The set of radial and tangential lens dis-
cient amm/pix from an object with known geometric tortion coefficients is also calculated by means of
dimensions. The scale coefficient gives information internal calibration. It is assumed that distortions
about the length corresponding to one pixel on the exhibited by lens can be modeled by fourth degree
image. In the next step, the reference image and one or polynomial given by the following equation (5)
more images of the construction under the load are
acquired. If the photographs are taken from distinct xd = R  xu + Dx
points in space, the homography transformation H, R = 1 + k1 r 2 + k2 r 4
  ð5Þ
calculated from a set of rectangular markers, trans- 2k3 xu yu + k4 (r2 + 2xu 2 )
forms the images in order to remove projective distor- Dx =
k3 (r2 + 2y2 ) + 2k4 xu yu
tions from the image of one particular plane of the
structure. Photographs transformed in such a way can where r2 = x2 u + y2 u ; xu = ½ xu y u T are the coordinates
be spatially overlaid with the reference image. This step of undistorted points on the image; xd are the coordi-
is optional and can be omitted if it is not necessary. nates of distorted points; k1, k2 are the coefficients of
The method has been presented by authors in their pre- radial distortions; and k3, k4 are the coefficients of tan-
vious articles.1,26,44–49 The detailed description of the gential distortions.
mentioned rectification algorithm can be found in the After computation of distortion coefficients, lens
previous studies.44–49 distortions can be removed from the image of the con-
The procedure of camera calibration is carried to struction. It must be done in order to reduce the mea-
obtain intrinsic calibration matrix K (4): surement error of the vision-based method.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 417

The scale coefficient amm/pix is computed from an with a corresponding pixel subset on the image of the
object on the scene with known geometric dimensions. loaded structure by means of the ZNCC coefficient.
The value of amm/pix can be obtained from planar circu- The displacement vector for each of the measurement
lar or rectangular marker or from the certified length point is computed as difference between calculated
standard. The scale calibration introduces Hough trans- positions of the pattern on two images. The method
form50,51 for circle detection and high-accuracy ellipse performed on all points of interest gives a complete dis-
fitting technique. placement field of the analyzed object.
In general, the digital image correlation method con- Correlation coefficient computes the position of the
sists of the following three steps: object preparation, pixel on discrete pixel grid on the image. When the sub-
acquisition of the object’s image before and after load- pixel methods are introduced, one can interpolate posi-
ing, and processing the acquired images to obtain the tion of the pixel to floating point value, which increases
displacement field. The speckle pattern on the struc- measurement accuracy up to 0.01–0.1 parts of pixel.
ture’s surface can be a natural texture or an artificially Developed algorithm takes integer value of pixel’s posi-
made special random intensity pattern attached to the tion given by correlation coefficient and fits a second-
structure or painted on it. The object’s surface must be degree polynomial into the four-element neighborhood
planar flat, and the out-of-plane motion of the object of the pixel. The new subpixel location of the correla-
during loading should be negligible. The basic principle tion coefficient maximum is the point for which the
of digital image correlation is matching of the same fitted parabola has its maximum value. An overview of
image path between the two images acquired before the mentioned deflection measurement is presented in
and after deformation.52–57 In order to compute the Figure 4.
displacement of a construction’s point, a square or rec-
tangular reference subset of pixels centered at a point is
chosen from the reference image and used to match its Software for the construction’s deflection
corresponding location in the deformed image. measurement
Assuming that deformations of the object are small,
The developed software enables construction deflection
only the local search is performed. The matching is car-
measurement using digital SLR cameras for remote
ried out by searching for the peak position of the corre-
image acquisition and provides image processing algo-
lation coefficient function over the search window. In
rithms to calculate the displacement field (Figure 5).
the developed method, the zero-mean normalized cross
The software implementation has been developed in the
correlation (ZNCC) has been chosen as a correlation
C + + /C# programing environment. ED-SDK libraries
criterion. This coefficient is more robust and is not sen-
provided by Canon have been used to control one or
sitive to changes in the scene’s illumination. The ZNCC
more cameras in the system. Two modes of operation
coefficient is given by equation (6).58,59
are available: online and off-line. In the first case, the
P   user specifies the date and number of measurements
fn ðx, yÞ  f n fd ðx  u, y  vÞ  f d
x, y and then the system works fully automatically carrying
ZNCC(u, v) = rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P 2 P  2ffi out the image acquisition and deflection measurement.
fn ðx, yÞ  f n fd ðx  u, y  vÞ  f d
x, y x, y The off-line mode provides analysis of the images
ð6Þ stored on hard disk. There are two methods of deflec-
tion calculation implemented in the software: in the
where fn(x, y) is an intensity value for a pixel with coor- first one, the user can specify points in which the mea-
dinates (x, y) on the reference image; fn is the mean surement will be carried out. The displacement field is
value of intensities of the pattern on image before computed in these sets of points only. However, one
deformation; fd(x2u, y2v) is the intensity value for a can use the dense measurement method in which one
pixel with coordinates (x, y) on the image after defor- specifies N—the number of points between the selected
mation; fd is the mean value of intensities of the pattern ones. The program automatically divides the structure
after deformation; x, y are the positions of the pattern between the user specified points into N equally spaced
on reference image; and u, v are the displacements of measurement locations. The result browser module car-
the pattern between two images. ries out the visualization of calculated curves of deflec-
In the process of deflection measurement, the refer- tion and report generation in popular formats, such as
ence image of the part of interest of the unloaded con- PDF, XML, CSV, and TXT. An additional feature of
struction is divided into set of rectangular intensity the software is detection of exceeding the allowed level
patterns, whose centers are taken as measurement of a deflection and the sending of alerts to a client by
points (Figures 4 and 5). Each of the pattern is matched an e-mail.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


418 Structural Health Monitoring 12(5-6)

Figure 5. The developed software for vision-based measurement of construction’s deflections.

Preliminary tests
Determination of the reference method in the
laboratory
Before conducting measurement of the real structure
deformation, it was decided to perform a preliminary
test. This was carried out in order to measure the user-
defined displacements of selected points of the investi-
gated structure and to control them using different
devices. For this purpose, a heavy photographic tripod
with adjustable head height was used. A special plat-
form was attached to the upper surface of the tripod
head. This was the base for fixing the microwave reflec- Figure 6. A moving platform constructed for the preliminary
test.
tor and a plate covered with the pattern used for digital
image analysis during measurements. The aim of this
experiment was not only the comparison of the pre- Ni002 precise analog surveying level was used together
sented methods but also the determination of their with the glass surveying staff (usually used in industrial
actual accuracy in relation to the precise devices. surveying), which was attached to the moving platform.
Therefore, the vertical movement of the tripod head Its SD equals 60.2 mm/km of double-run leveling.
was controlled using the Helios DIGI-MET electronic This gives 60.09 mm error for the performed displace-
dial indicator (DI), which ensures the measurement ments (Figure 6).
range of 50 mm and the SD of 0.01 mm (and the maxi- In the case of such precise measurements, the geo-
mum permissible error of 0.02 mm). In order to per- metric conditions of the created measurement system
form an additional control of displacement and are important. The Leica TCR407 total station was
stability of the whole construction, a Carl Zeiss Jena used to control whether the attached platform was

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 419

perpendicular to the tripod vertical axis. Moreover, the where n is the number of measured displacements.
condition to set the DI axis in parallel to the performed The Pearson product-moment correlation coefficient
displacements was achieved. r was also determined:
In order to verify the correctness of the displacement
sensor attachment and the mutual compatibility of the covðdZDI , dZLev Þ
r= = 0:99999887 ð8Þ
two reference instruments, an independent test was car- sDI  sLev
ried out in the laboratory. The tripod head, together with
the glass staff (and microwave reflector—to simulate real where cov(dZDI, dZLev) is the covariance movements
conditions), was moved vertically over a distance of 40 from both measurement methods and s(dZ) is the SD
mm (not more than the DI range) with increments of 10 of measured displacement.
mm. The displacement values were recorded with the The obtained results fully confirm the high compat-
DIGI-MET DI and the Ni002 level. The level values are ibility of the two instruments and their use in checking
marked by the Lev symbol and the DI values by the DI the accuracy of displacements determined by the radar
symbol, and dZ means the vertical displacement. or VB.
This measuring procedure was performed twice. The
results are presented in Table 1. Simulation of measurement
In order to evaluate the agreement of both methods,
the SD of differences between the recorded displacements The geometry of test required to simulate the creation
was calculated, using the method of observation pairs: of the typical conditions that occur when measuring
the deflection of a bridge structure span (Figure 7,
sffiffiffiffiffiffiffiffiffiffiffi shown from a side view, and Figure 8). Therefore, the
P 2
D main part of the preliminary test was performed on an
sD = 6 = 60:025 mm ð7Þ
n anti-flood embankment. The construction simulating
displacements of the real structure was set at the top of
Table 1. Results of verifying the geometrical stability of the the embankment, while the RI was located beneath.
moving platform. The horizontal distance between the radar and the
microwave reflector was 9.92 m (d2 in Figure 8) and the
Set Displacement values (mm) D = dZDI2dZLev (mm) height difference was 4.26 m. For this setup, the projec-
dZDI dZLev tion factor p equaled 2.5365. This was the initial value.
For every position of the moving platform, the actual
1 0.00 0.00 0.00 value of p was calculated.
10.11 10.10 0.01
20.07 20.11 20.04 During the field tests, a flat aluminum plate of the
30.14 30.11 0.03 dimensions 0.42 m 3 0.30 m was fixed to the moving
40.08 40.10 0.02 platform. The random intensity pattern was placed on
2 0.00 0.00 0.00 the measured plane of the plate. Photographs of the
9.77 9.77 0.00 object were acquired by a system of a Canon 5D Mark
19.81 19.84 20.04
30.19 30.17 0.02 II digital camera with 21.1 Mpixel image resolution
41.00 40.99 0.01 and a Canon 24–70 mm f/2.8 L lens with focal length
f ’ 70 mm adjusted. The position of the camera was

Figure 7. Field test equipment arrangement (side view).

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


420 Structural Health Monitoring 12(5-6)

Figure 8. Field test equipment arrangement.


IBIS: Image by Interferometric Survey.

fixed in space. The camera had its optical axis at the The values of displacements recorded by the Lev, DI,
height of the displaced plate and perpendicular to it in RI, and VB systems are presented in Table 2. Due to
order to ensure the uniform imaging scale. The distance the high agreement of Lev and DI results, their averages
between the camera and the plate was 15.4 m (d1 in were assumed as the reference displacement values: Ref
Figure 8). Regarding scene illumination, there were = (Lev + DI)/2. Moreover, the differences D between
natural lighting conditions with sunlight and cloudy displacements obtained by different methods were cal-
variations, so there was no need for artificial lighting. culated in Table 2. Four groups of differences were
For the purposes of an independent verification of determined:
both measurement systems (radar and vision based), an
electronic DI and precise Ni002 level were used. The  Level versus DI (both assumed later as the
mutual agreement of the precise NI002 level and the reference);
DIGI-MET DI had been previously confirmed. A sche-  RI in relation to the reference;
matic layout of all elements of the measurement system  VB in relation to the reference;
is shown in Figure 8.  RI in relation to the VB.
In the case of the vision system, the scale coefficient
was computed from an object on the scene with known The SDs sD of four groups of differences were also
geometric dimensions.44 The value of 1.53 mm/pix was calculated according to equation (7). They are pre-
obtained from an image of the certified surveying staff sented in the last row of Table 2.
using a WizScale module of the developed software. All determined correlation coefficients indicate good
As in the case of laboratory tests, in this experiment, agreement of measurement results (more than 99.96%).
the displacements were generated at the range of 45 mm Their values indicate the highest compatibility of results
(four movements of 10 mm and the last movement of 5 obtained by the Lev and the DI. This confirms the cor-
mm). The measurement was performed in two steps. rect selection of these devices as reference sensors. The r
The first three runs of the moving platform were per- coefficients in the last row of Table 2 (in columns 7, 9,
formed in the following directions: downward, upward, 10) show lower compatibility of the VB and RI systems.
and downward. Displacements were observed by three
devices: precise level (Lev), DI, and RI (Table 2, tests
no. 1–3). In the second step, two runs were performed Key study—tram viaduct
(downward and upward). Displacements were mea-
sured additionally by the VB (Table 2, tests no. 4 and Field test
5). In total, five runs of five displacements were carried In order to perform the final verification of both non-
out. contact measurement techniques, an experiment was

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 421

Table 2. Vertical displacements of the moving platform recorded by the systems.

Test no. Lev DI D = Lev2DI Ref RI D = RI2Ref VB D = VB2Ref D = RI2VB


(mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm)

1 210.29 210.22 20.07 210.26 29.50 0.76 — — —


220.29 220.22 20.07 220.26 219.89 0.37 — — —
230.39 230.19 20.20 230.29 230.09 0.20 — — —
240.24 240.02 20.22 240.13 240.14 20.01 — — —
245.59 245.38 20.21 245.48 245.58 20.10 — — —
2 239.84 239.85 0.01 239.84 240.21 20.37 — — —
229.99 230.01 0.02 230.00 230.11 20.11 — — —
220.20 220.28 0.08 220.24 219.97 0.27 — — —
29.44 29.51 0.07 29.48 29.36 0.12 — — —
0.24 0.14 0.10 0.19 0.41 0.22 — — —
3 29.89 29.89 0.00 29.89 29.44 0.45 — — —
220.54 220.48 20.06 220.51 220.32 0.19 — — —
231.00 230.90 20.10 230.95 230.83 0.12 — — —
239.89 239.70 20.19 239.80 239.80 0.00 — — —
— 245.47 — 245.47 245.85 20.38 — — —
4 210.58 210.52 20.06 210.55 29.90 0.65 29.72 0.83 0.18
221.71 221.60 20.11 221.66 221.32 0.34 220.91 0.75 0.41
230.34 230.25 20.09 230.30 230.29 0.01 230.48 20.18 20.19
240.06 240.02 20.04 240.04 240.59 20.55 240.03 0.01 0.56
244.88 244.84 20.04 244.86 245.51 20.65 244.63 0.23 0.88
5 239.92 240.08 0.16 240.00 240.47 20.47 239.33 0.67 1.14
230.29 230.48 0.19 230.39 230.42 20.03 229.93 0.46 0.49
220.48 220.58 0.10 220.53 220.49 0.04 220.29 0.24 0.20
29.90 210.03 0.13 29.96 29.95 0.01 29.52 0.44 0.43
0.18 0.04 0.14 0.11 0.91 0.80 0.67 0.56 20.24
sD (mm) 0.121 0.378 0.506 0.561
r (—) 0.999970 0.999879 0.999811 0.999792

Lev: precise level; DI: dial indicator; RI: radar interferometer; VB: vision-based system.

performed on a real engineering structure. The test


involved monitoring of a 28-m-long steel span
(Figure 9), which is part of a tram viaduct. The
observed deflections were forced by trams of various
mass, periodically passing in the south–north direction.
The span deflections were measured with two devices:
an IBIS-S RI and a VB (in the following tables marked
as RI and VB, respectively). Span displacement mea-
surements were performed from two radar positions
relative to the structure and a constant camera
position.
The main part of the VB is a Canon EOS 5D Mark
II SLR full-frame digital camera. The camera was
equipped with a Canon EF 24–70 mm f/2.8 L lens set
to work with approximately 50 mm of focal length.
The lens was focused at infinity, and the shutter speed
was set to 1/100 s at f = 18 aperture. The high value of Figure 9. The measurement setup consisting of a digital SLR
aperture was set intentionally to obtain a large depth of camera and IBIS-S radar interferometer.
field, bearing in mind that the shutter speed was high SLR: single-lens reflex; IBIS: Image by Interferometric Survey.
enough to get blur-free images. The camera was situ-
ated at the distance of 27.77 m from the side plane of direction to the span axis. The shutter was triggered 3.9
the span to set the projection of the lens center in the times per second to record the full tram passing with
midspan of the bridge. In order to obtain a uniform the image series.
scale in the entire field of view (covering 14.5 m of the In addition, the continuous deflection measurement
span), the lens axis was aimed at a perpendicular was performed using the IBIS-S radar.11 In order to

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


422 Structural Health Monitoring 12(5-6)

Table 3. Deflections of 10 points (for the first radar position) measured by the RI and VB systems and differences between results.

Run no. Method Measured points (from the beginning of span) (m) Dmin; Dmax
(mm)
7.5 10.6 11.2 11.7 13.9 14.5 15.0 17.1 17.6 20.8

Maximum deflections (mm)

11 LD 27.62 20.26; 0.97


RI 25.28 26.42 26.30 26.78 27.28 27.17 27.38 26.24 25.65 24.09
VB 25.58 26.85 26.94 27.02 27.21 27.17 27.12 26.66 26.48 25.06
D 0.30 0.42 0.64 0.24 20.07 0.000 20.26 0.42 0.83 0.97
12 LD 26.62 20.47; 0.80
RI 25.02 26.07 25.87 26.33 26.85 26.76 26.97 25.88 25.34 23.98
VB 25.12 26.21 26.36 26.44 26.60 26.53 26.50 26.08 25.97 24.78
D 0.10 0.14 0.49 0.11 20.25 20.23 20.47 0.20 0.63 0.80
13 LD 26.72 0.29; 1.20
RI 24.95 25.95 25.81 26.30 26.39 26.26 26.46 25.63 25.01 24.50
VB 25.46 26.46 26.64 26.77 26.88 26.82 26.76 26.34 26.22 25.00
D 0.51 0.549 0.83 0.47 0.49 0.56 0.30 0.71 1.21 0.50
14 LD 26.32 20.42; 0.60
RI 24.52 25.40 25.37 25.80 26.25 26.16 26.35 25.41 24.90 23.87
VB 24.82 25.73 25.83 25.86 26.09 25.93 25.93 25.54 25.49 24.47
D 0.30 0.33 0.46 0.06 20.16 20.23 20.42 0.13 0.59 0.60
15 LD 26.82 20.36; 0.79
RI 25.16 26.13 25.94 26.42 26.92 26.81 27.01 25.97 25.39 24.05
VB 25.38 26.34 26.43 26.58 26.75 26.67 26.65 26.12 26.04 24.84
D 0.22 0.21 0.49 0.16 20.17 20.14 20.36 0.15 0.65 0.79

RI: radar interferometer; VB: vision-based system; LD: laser distance meter.

record the bridge span deflections with this device, the Knowledge of external orientation elements allows
radar unit was located in the vicinity of the line of the reconstruction of the spatial position of the camera
pillars. During the first five recorded runs (runs no. relative to the structure and thus the corrections neces-
11–15, Table 3) the radar was placed in the bridge girder sary to obtain proper displacements of the tested struc-
plane, while during the next three runs (runs no. 21–23, ture. The determination of orientation elements was
Table 4), it was set directly under the track-way axis. performed by the close-range terrestrial photogramme-
The independent control of measured deflections of try procedure involving capturing the pair of conver-
the center part of the investigated girder was carried ging images with the basis equaled approximately one-
out with a Leica DISTO D8 laser distance meter. By fourth of the distance to the observed bridge. Both the
connecting it in the online mode to a computer, it was targets fixed to tripods and elements of optical noise
possible to record measurements continuously at a fre- covering the bridge span were used as control points.
quency of 2 Hz. All points were measured in a uniform coordinate sys-
In the field, the lens axis was not exactly perpendicu- tem using the Leica TCRA1102 + total station.
lar to the span axis. This effect may be eliminated with Based on the knowledge of the local terrain coordi-
analytical methods. The orientation and position of the nates, for the pair of images, it was possible to calculate
camera are described by the external orientation ele- the components of vectors in the spatial coordinate sys-
ments. In the described case, there were three local ter- tem of images. The relation between the spatial coordi-
rain coordinates X, Y, Z, and three angles:60 nates and the vector components is given by the
formula:
v—vertical angle of the camera axis tilt; 2 3 2 3
u—horizontal angle that describes the camera direction x x9
in the coordinate system (azimuth); 4 y 5 = ½ M 4 y9 5 ð9Þ
k—angle of fiducial plane rotation. z ck

where M is the transformation matrix given by


2 3
coskcosu cosusink sinu
½ M ¼4 sinvsinucoskþcosvsink sinvsinusinkþcosvcosk sinvcosu 5 ð10Þ
cosvsinucosk þ sinvsink cosvsinusinkþsinvcosk cosvcosk

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 423
2 3

Dmin; Dmax (mm)


x
where 4 y 5 are the spatial image coordinates, (x#,y#)

20.47; 0.79

0.89; 2.21

20.20; 1.04
z
are the image coordinates, and ck is the focal length of
camera.
After adjustment of the internal orientation, the

24.16
23.69
20.47

24.49
25.51
1.02

23.37
23.80
0.43
deflection of the tested span can be expressed by the
20.9

formula:

Y
24.40
24.32
20.08

24.82
25.84
1.02

23.64
24.10
0.46
DZ = Dz ð11Þ
20.3

ck

where DZ is the beam deflection, Y is the distance from


25.06
24.90
20.16

25.32
27.16
1.84

24.54
24.38
20.16
19.3

the object; Dz is the image representation of the beam


deflection.
Table 4. Deflections of 14 points (for the second radar position) measured by the RI and VB systems and differences between results.

Basically, this type of measurement method is used


25.09
24.98
20.11

25.51
26.61
1.10

24.70
24.68
20.02
18.8

if it is possible to direct the camera axis perpendicularly


to the plane of the occurring deformation. In practice,
it must be assumed that the component of deformation
25.69
25.47
20.22

26.10
26.99
0.89

25.35
25.15
20.20
17.2

in the direction of the camera axis is not present or


negligible. This situation occurred in the described
cases, because the plane deformation of the viaduct
25.77
26.01
0.24

26.01
27.65
1.64

25.34
25.42
0.08

was almost perpendicular to the axis of the camera and


15.1

the movement occurred in the longitudinal vertical


plane of the structure. Therefore, both of the 1D sys-
25.56
26.02
0.46

26.08
27.64
1.56

25.46
25.54
0.08

tems—IBIS-S and vision-based—observed the same


14.6

deflections of span.
In the first radar position, the displacements mea-
25.55
26.03
0.48

25.98
27.63
1.65
25.68
25.37
25.49
0.12

sured by the radar were analyzed for 10 points of the


14.0

span. The position of the points is shown in Table 3.


For these points, the SNR value (describing their inten-
25.77
26.49
0.72

25.98
27.98
2.00

25.18
25.52
0.34

sity of reflection) is from 56.1 to 85.5 dB. In the second


12.4

radar position, the displacements measured by the


Measured points (from the beginning of span) (m)

radar were analyzed for 14 points of the span. The


25.43
26.15
0.72

25.55
27.70
2.15

24.85
25.38
0.53

position of the points is shown in Table 4. For these


11.3

RI: radar interferometer; VB: vision-based system; LD: laser distance meter.

points, the SNR value is from 57.7 to 82.7 dB. Figures


10 and 11 present the results of measurements per-
25.27
26.02
0.75

25.34
27.55
2.21

24.64
25.36
0.72

formed with the RI and VB systems during a single run


10.7

recorded from the first radar position. Displacements


recorded by the VB system are determined for places,
Maximum deflections (mm)

24.88
25.50
0.62

24.97
26.84
1.87

24.43
24.95
0.52

which correspond to the points (range bins) observed


9.0

by the RI system. This was performed by identifying


and calculating these places in acquired digital images.
24.64
25.11
0.47

24.75
26.51
1.76

24.22
25.01
0.79

Slight differences may arise only from selection of the


8.4

proper points submitted to correlation calculation in


the software during processing the data from the VB
23.77
24.56
0.79

23.93
25.77
1.83

23.49
24.53
1.04

system.
7.1

Tables 3 and 4 contain the comparison of maximum


deflection obtained from the RI and VB systems for all
Method

observed points. The differences between the methods


are calculated as D = VB2RI. The minimum and max-
VB
LD

LD

LD
VB

VB
RI

RI

RI
D

imum value of the D set is presented for all runs for the
first (Table 3) and second (Table 4) radar position. In
Run no.

addition, the values of maximum deflection of the cen-


21

22

23

tral span point, measured by the laser distance meter

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


424 Structural Health Monitoring 12(5-6)

Figure 10. Graph of the deflected bridge span obtained from the RI system illustrated every 0.5 s (observed points are marked).
RI: radar interferometer.

Figure 11. Graph of the deflected bridge span obtained from the VB system illustrated every frame (observed points are marked).
VB: vision-based system.

(LD), are also shown. All runs were observed during methods: radar interferometry and vision-based.
loading of the closer track-way. Calculations were carried out as follows.

1. The linear relation between the results of both


methods was verified.
Statistical assessment of results 2. The regression coefficient (the derivative of the
The aim of the assessment based on statistical models previously determined linear model) was calculated
is the evaluation of compatibility of two measuring as a necessity condition of method compatibility.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 425

Table 5. Results of statistical assessment for all eight observed runs.

Data set Linear regression Multiple t-test of F-test of Bland–Altman t-test of SD of the
R2 (—) averages variances plot fit? significance differences
Slope Intercept inequality equality for the expected (mm)
coefficient coefficient (p value) (—) (p value) (—) value
(—) (mm) of differences

11 0.620 22.218 0.884 0.591 0.230 OK No 0.352


12 0.649 22.295 0.921 0.577 0.260 OK OK 0.366
13 1.100 0.208 0.730 0.285 0.374 OK No 0.426
14 1.010 21.555 0.709 5.702e206 0.521 OK OK 0.435
15 0.679 21.815 0.736 0.180 0.411 OK OK 0.371
21 0.676 22.375 0.912 0.394 0.318 OK No 0.396
22 0.630 22.332 0.885 0.675 0.248 OK No 0.402
23 0.850 21.470 0.867 0.053 0.790 No No 0.251

SD: standard deviation.

3. The equality of the mean values from the RI data vector of differences RI2VB. The upper agreement
and VB data was calculated. limit (UAL) and lower agreement limit (LAL) indicate
4. The equality of the SD values from the RI data and the 62s value, where s means the SD of differences.
VB data was calculated. Afterward, the test of the distribution compatibility
5. The compatibility of methods was calculated using with the normal distribution for pairs of observation
the Bland–Altman test. differences was performed. The obtained result (p-value
= 0.0835) is close to the level of significance of the test.
The following calculations were made for a sample Therefore, it can be examined with the standard t-test,
run (no. 13). The values of linear regression for all eight which assumes that the sample has a normal distribu-
runs are presented in Table 5. Initially, the compatibil- tion. As a result, the p value = 0.0206 was obtained.
ity between span deflections obtained from the RI and This is lower than the assumed level of significance
VB systems was verified by examining the existence of level. Thus, it suggests that there is a systematic shift
a linear correlation coefficient between the results for between methods, so they are not compatible. In the
the corresponding points of span. For a sample run same way, as presented above, all eight runs were ana-
observed from the second radar position, a linear rela- lyzed. The results are summarized in Table 5.
tion is described by a slope coefficient, which is 1.100 For all eight analyzed cases, the test of significance
with 60.193 SD, and the intercept coefficient 0.208 of correlation between variables gave a positive result.
mm. In the discussed linear model, representing the Therefore, there is a correlation between results from
relationship between the results obtained with two sys- both methods, but the systems do not shift the results
tems, the residual standard error is 0.439 mm on 12 in relation to each other in only three out of eight cases.
degrees of freedom. The value of the correlation coeffi- The calculation of the relation between the RI and VB
cient is 0.730 (multiple R2).61,62 methods was also performed for the values of maxi-
The next step was to test the hypothesis of equality mum deflection of the midpoint of the span for all eight
of means. Due to the lack of assumptions about var- runs (Figure 13).The horizontal axis contains the vector
iances, the Welch two sample t-test was used.63,64 The of means of maximum deflections, and the vertical axis
calculated p-value is 0.285 (higher than the assumed contains the vector of differences of maximum deflec-
level of significance 0.05), so there is no reason to reject tions. The UAL and LAL indicate the 62s value,
the null hypothesis. Similarly, the equality of variances where s means the SD of differences of maximum
was tested on the basis of the Fisher–Snedecor F-test. deflections.
A good result of the p-value, equaling 0.374, was The expected value is zero. According to the data
obtained. The results of both tests allow the acceptance related to the maximum span deflections (Figure 13),
of the null hypotheses of equality of means and var- the compatibility of methods can be concluded.
iances. In order to assess the compatibility of the data
from the RI and VB systems, they were compared using
Conclusion
the Bland–Altman test.65,66 The results are illustrated
in Figure 12. The horizontal axis contains the vector of The developed vision measurement system based on the
means (RI + VB)/2 and the vertical axis contains the digital image correlation coefficient enables assessment

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


426 Structural Health Monitoring 12(5-6)

markers are available. The developed vision system is


applied with user-friendly software, which can help one
to quickly assess the state of the construction during
inspection. Furthermore, the developed software tool
provides a high level of measurement process automa-
tion, accomplished through operations such as image
acquisition, image preprocessing, and analysis. The sys-
tem monitors the state of the structure and informs the
user when the critical level of measurement estimates
has been exceeded and a message can be sent by an
e-mail. The system provides high measurement density
without the application of active optical methods.
On the contrary, the radar interferometry can also
be used to measure the deflection of the structures. The
RI enables the observation of many points on the struc-
ture simultaneously in the remote mode, without direct
access to the structure. It provides the high sampling
Figure 12. Bland–Altman plot for a sample run (no. 13).
UAL: upper agreement limit; LAL: lower agreement limit.
frequency, especially useful for dynamic tests. In addi-
tion, the system is independent on weather conditions.
The accuracy assessment is a key factor in the per-
formance of such kinds of systems. The performed field
tests allow the determination of the applicability of the
proposed vision-based technique for measuring the
deflection of civil engineering constructions. The vision
system measurement resulted in data that matched the
points’ displacement registered by the RI.
The preliminary test arranged to simulate the displa-
cements of the observed point of the defined values
enabled the accuracy of both systems to be calculated.
The SD of the differences amounted to 0.38 mm for the
RI and 0.51 mm for the VB. The mutual accuracy of
both methods was 0.56 mm. All examined methods indi-
cated a correlation of 99.96%. As a result of the
obtained correlation, the key test on the tram viaduct
was performed. The obtained graphs of deflection of the
central point seem to indicate the greater advantage of
Figure 13. Bland–Altman plot for maximum deflections of all the VB. The curve of deflection acquired by this system
eight runs. becomes more accurate and smoother than by radar
UAL: upper agreement limit; LAL: lower agreement limit. interferometry. The irregular shape of the radar results
arises from the uncertainty of the points reflecting the
radar wave. However, a theoretical, regular graph may
of a structure’s static states under a load during inspec- be fitted into this data. It must also be noted that the RI
tions, as well as continuous deflection monitoring, to be is a 1D system. This means that, for low structures,
carried out using the change in the shape of deflection results may be worse because of the unfavorable projec-
curve analysis. The main advantages of the proposed tion factor, so users must be careful to avoid improper
methods (embedded in the developed vision system) are scaling of displacements. In the main test, the SD of the
the simplicity of the measurement, the possibility of differences for the maximum deflection of the bridge
obtaining a dense field of a deflection using only one span was designated by a VB and RI and equaled 0.61
acquisition device, the application of commonly avail- mm. On the basis of performed tests, the usefulness of
able digital cameras, the easy image acquisition process both systems for displacement measurements was found
that can be performed from different points in space, as for the purposes required in terms of SHM.
well as the fast and easy analysis and interpretation of
measurement results. The method can be applied to the
structures with a texture in the form of noise or natural Declaration of conflicting interests
texture of materials as well as in the case where special The authors declare that there is no conflict of interest.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 427

Funding measurement of vibrations and evaluation of their effects


The research was supported by the OPIE project No: on structures. Geneva: International Organization for
01.01.02-00-013/08 co-financing by ERDF within a frame of Standardization.
Operational Program Innovative Economy and MNiSW 15. Gens R and Van Genderen JL. SAR interferometry:
(Polish Ministry of Science and Higher Education) under issues, techniques, applications. Int J Remote Sens 1996;
grant N N526 158838. 17: 1803–1836.
16. Pieraccini M, Luzi G, Mecatti D, et al. Ground-based
radar interferometry for monitoring unstable slopes. In:
References Proceedings of joint 12th FIG international symposium on
1. Uhl T, Kohut P, Holak K, et al. Vision based condition deformation measurements and analysis/3rd IAG sympo-
assessment of structures. J Phys Conf Ser 2011; 305: 9–10. sium on geodesy for geotechnical and structural engineer-
2. Sustainable Bridges. Inspection of railway bridges in Eur- ing, Baden, Austria, 22–24 May 2006.
ope. In: Helmerich R (ed.) Guideline for inspection and 17. Pieraccini M, Tarchi D, Rudolf H, et al. Structural static
condition assessment of existing European railway bridges. testing by interferometric synthetic radar. NDT&E Int
Sixth Framework Programme. Bam, Iran: European 2000; 33: 565–570.
Commission, 2007, pp. 19–35. 18. Tarchi D, Rudolf H, Pieraccini M, et al. Remote moni-
3. Ryall MJ. Bridge management. 2nd ed. Amsterdam: toring of buildings using a ground-based SAR: applica-
Elsevier. tion to cultural heritage survey. Int J Remote Sens 2000;
4. Alampalli S. Effects of testing, analysis, damage, and 21: 3545–3551.
environment on modal parameters. Mech Syst Signal Pr 19. Pieraccini M, Fratini M, Parrini F, et al. High-speed CW
2000; 14: 63–74. step-frequency coherent radar for dynamic monitoring of
5. Cunha Á, Caetano E and Magalhães F. Output-only civil engineering structures. Electron Lett 2004; 40: 907–908.
dynamic testing of bridges and special structures. Struct 20. Rödelsperger S. Real-time processing of ground based syn-
Concr 2007; 8: 67–85. thetic aperture radar (GB-SAR) measurements. Darm-
6. Łaziński P and Salamak M. Identification of computa- stadt: Technische Universität, Schriftenreihe der
tional models in load carrying structures of concrete Fachrichtung Geodäsie, Heft 33, 2011.
bridges on the basis of making load tests. In: Proceedings 21. Gentile C. Application of radar technology to deflection
of the 7th Central European Congress on concrete engi- measurement and dynamic testing of bridges. In: Koue-
neering CCC 2011, Balatonfüred, Hungary, 22–23 Sep- mou G (ed.) Radar technology. Vukovar: InTech, 2010,
tember 2011. pp. 141–162.
7. Ko JM and Ni YQ. Technology developments in struc- 22. Pieraccini M, Parrini F, Fratini M, et al. In-service testing
tural health monitoring of large-scale bridges. Eng Struct of wind turbine towers using a microwave sensor. Renew
2005; 27: 1715–1725. Energ 2008; 33: 13–21.
8. Larocca APC. Using high-rate GPS data to monitor the 23. Uhl T, Kohut P and Holak K. Structure deflection exami-
dynamic behavior of a cable-stayed bridge. In: Proceed- nation by means of correlation coefficient. In: Uhl T (ed.)
ings of the 17th international technical meeting of the ION Selected problems of modal analysis of mechanical systems:
Satellite Division, ION GNSS 2004, Long Beach, CA, 21– proceedings of the 13th school of modal analysis. Radom:
24 September 2004. National Research Institute, 2009, pp. 173–180.
9. Watson C, Watson T and Coleman R. Structural moni- 24. Kohut P and Kurowski P. The integration of vision based
toring of cable-stayed bridge: analysis of GPS versus measurement system and modal analysis for detection
modeled deflections. J Surv Eng: ASCE 2007; 133: 23–28. and localization of damage. In: Szpytko J (ed.) Engineer-
10. Casadei P, McCombie P, Galati N, et al. NDT monitor- ing achievements across the global village, Cracow, Glas-
ing of bridges using innovative high precision surveying gow, Radom: The International Journal of INGENIUM.
system. In: Proceedings of IABSE 2006 symposium, Buda- Monographic Series of the Library of Maintenance Prob-
pest, Hungary, 13–15 September 2006. lems, 2005, pp. 391–398.
11. Owerko T, Ortyl Ł, Kocierz R, et al. Investigation of dis- 25. Kohut P and Kurowski P. The integration of vision sys-
placements of road bridges under test loads using radar tem and modal analysis for SHM application. In: Pro-
interferometry—case study. In: Biondini F and Frango- ceedings of the 24th conference and exposition on structural
pol DM (eds) Bridge maintenance, safety, resilience and dynamics 2006 (IMAC XXIV). St Louis, MO, 30 Janu-
sustainability. London: Taylor & Francis Group, 2012, ary–2 February 2006.
pp. 181–188. 26. Uhl T, Kohut P, Holak K, et al. Vision based vibration
12. Paultre P, Proulx J and Talbot M. Dynamic testing pro- and deformation measurement. In: Proceedings of the 5th
cedures for highway bridges using traffic loads. J Struct European workshop on structural health monitoring (eds F
Eng: ASCE 1995; 121: 362–376. Casciati and M Giordano), Sorrento, Italy, 28 June – 4
13. Park K-T, Kim S-H, Park H-S, et al. The determination July 2010, pp. 1005–1010. Pennsylvania, PA: DEStech
of bridge displacement using measured acceleration. Eng Publications.
Struct 2005; 27: 371–378. 27. Jerabek M, Major Z and Lang RW. Strain determination
14. ISO 4866:2010. Mechanical vibration and shock— of polymeric materials using digital image correlation,
vibration of fixed structures—guidelines for the Polymer Testing 2010; 29(3): 407–416.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


428 Structural Health Monitoring 12(5-6)

28. Lee M, Precht M, Tyson J, et al. Application of 3D mea- 43. Kuras P, Owerko T, Ortyl Ł, et al. Advantages of radar
surement system with CCD camera in microelectronics. interferometry for assessment of dynamic deformation of
Adv Packag 2003; November: 33–34. bridge. In: Biondini F and Frangopol DM (eds) Bridge
29. Schmidt T, Tyson J and Galanulis K. Full-field dynamic maintenance, safety, resilience and sustainability. London:
displacement and strain measurement using advanced 3D Taylor & Francis Group, 2012, pp. 885–891.
image correlation photogrammetry. Exp Techniques 44. Kohut P, Holak K, Uhl T, et al. Structure’s condition
2003; 27(3): 41–44. monitoring based on optical measurements. Key Eng Mat
30. Johnson E and Woodard N. Low-cost electrical resis- 2012; 518: 338–349 (Structural health monitoring II: pro-
tance spot welding technique for micro-optical compo- ceedings of the 2nd international conference on smart diag-
nent mounting with sub-micron tolerance. In: nostics, Cracow, Poland, 14–16 November 2011).
Proceedings of photonics packaging symposium, San Fran- 45. Kohut P, Uhl T and Holak K. Image correlation tech-
cisco, CA, 21–25 April 2003. niques in structures deflection measurements. In:
31. Schmidt T, Tyson J and Galanulis K. Dynamic strain EVACES’09: proceedings of the international conference
measurement using advanced 3D photogrammetry. In: on experimental vibration analysis for civil engineering
Proceedings of IMAC XXI, Orlando, FL, 3–6 February structures (ed Jan Bień), Wrocław, Poland, 14–16 Octo-
2003. ber 2009, pp. 147–148. Dolnoślaxskie Wydawnictwo
32. Tyson J, Schmidt T, Shahinpoor M, et al. Biomechanics Edukacyjne.
deformation and strain measurements with 3D image cor- 46. Kohut P, Holak K and Martowicz A. An uncertainty
relation. Exp Techniques 2002; 26(5): 39–42. propagation in developed vision based measurement sys-
33. Tyson J, Schmidt T and Galanulis K. Advanced photo- tem aided by numerical and experimental tests. J Theor
grammetry for robust deformation and strain measure- Appl Mech 2012; 50(4): 1049–1061.
ment. In: Proceedings of SEM 2002 annual conference, 47. Sładek J, Ostrowska K, Kohut P, et al. Development of a
Milwaukee, WI, 10–12 June 2002. vision based deflection measurement system and its accu-
34. Bergmann D and Ritter R. 3D deformation measurement racy assessment, Measurements, 2013; 46( 3): 1237–1249.
in small areas based on grating method and photogram- 48. Holak K, Kohut P, Martowicz A, et al. An uncertainty
metry. In: Proceeding of SPIE 2782: optical inspection and analysis for developed measurement vision system aided
micromeasurements, Besancon, France, 10 June 1996, pp. by numerical simulations. Mech Control 2011; 30(2):
212–223. 65–72.
35. Olaszek P. Investigation of the dynamic characteristic of 49. Kohut P, Holak K and Uhl T. Monitoring of civil
bridge structures using a computer vision method. Mea- engineering structures supported by vision system. In:
surement 1999; 25: 227–236. Structural health monitoring 2012: proceedings of the sixth
36. Lee J-J and Shinozuka M. A vision-based system for European workshop (ed Christian Boller), Dresden, Ger-
remote sensing of bridge displacement. NDT&E Int 2006; many, 3–6 July 2012, pp. 1575–1582. DGZfP.
39(6): 425–431. 50. Gonzales R and Woods R. Digital image processing. New
37. Yoneyama S, Kitagawa A, Iwata S, et al. Bridge deflec- Jersey: Prentice Hall, 2002.
tion measurement using digital image correlation. Exp 51. Russ J. Image processing handbook. London, UK: CRC
Techniques 2007; 31(1): 34–40. Press, 1998.
38. Tung S-H, Shih M-H and Sung W-P. Development 52. Berfield TA, Patel JK, Shimmin RG, et al. Micro- and
of digital image correlation method to analyze crack var- nanoscale deformation measurement of surface and inter-
iations of masonry wall. Sadhana 2008; 33(6): 767–779. nal planes via digital image correlation. Exp Mech 2007;
39. Kujawińska M, Bukalska M and Dymny G. Hybrid digi- 47: 51–62.
tal image correlation/thermovision system for monitoring 53. Li N, Sutton MA, Li X, et al. Full-field thermal deforma-
of civil engineering structures. In: Proceedings of the 10th tion measurements in a scanning electron microscope by
international conference on quantitative infrared thermo- 2D digital image correlation. Exp Mech 2008; 48:
graphy, Québec, Canada, 27–30 July 2010. 635–646.
40. Kujawińska M, Malesa M, Malowany K, et al. Digital 54. Helm JD. Digital image correlation for specimens
image correlation method: a versatile tool for engineering with multiple growing cracks. Exp Mech 2008; 48:
and art structures investigations. In: Proceedings of SPIE 753–762.
8011: 22nd congress of the international commission for 55. Roux S and Hild F. Stress intensity factor measurements
optics: light for the development of the world, Puebla, Mex- from digital image correlation: post-processing and inte-
ico, 2 November 2011. grated approaches. Int J Fracture 2006; 140: 141–157.
41. Kujawińska M, Sitnik R, Dymny G, et al. Remote on-line 56. Réthoré J, Gravoil A, Morestin F, et al. Estimation of
monitoring and measuring system for civil engineering mixed-mode stress intensity factors using digital image
structures. In: Proceedings of SPIE 7389: optical measure- correlation and an interaction integral. Int J Fracture
ment systems for industrial inspection VI, Munich, Ger- 2005; 132: 65–79.
many, 15 June 2009. 57. Fujun Y and Xiaoyuan H. Digital speckle projec-
42. Zaurin R. Computer vision oriented framework for struc- tion for vibration measurement by applying digital
tural health monitoring of bridges. Struct Health Monit image correlation method. Key Eng Mat 2006; 326–328:
2011; 10(3): 309–332. 99–102.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015


Kohut et al. 429

58. Hartley R and Zisserman A. Multiple view geometry in 63. Welch BL. The generalization of ‘‘Student’s’’ problem
computer vision. Cambridge: Cambridge University Press, when several different population variances are involved.
2004. Biometrika 1947; 34: 28–35.
59. Ma Y, Soatto S, Kosetska J, et al. An invitation to 3D 64. Rees DG. Essential statistics. 4th ed. Boca Raton, FL:
computer vision. New York: Springer-Verlag, 2004. Chapman & Hall/CRC, 2001.
60. Kraus K. Photogrammetry. Vol. 2, advanced methods and 65. Altman DG and Bland JM. Measurement in medicine:
applications. Bonn: Ferd. Duemmlers Verlag, 1997. the analysis of method comparison studies. Statistician
61. Chambers JM. Linear models. In: Chambers JM and 1983; 32: 307–317.
Hastie TJ (eds) Statistical models in S. Boca Raton, FL: 66. Bland JM and Altman DG. Statistical methods for asses-
Chapman & Hall/CRC, 1991, pp. 95–144. sing agreement between two methods of clinical measure-
62. Wilkinson GN and Rogers CE. Symbolic descriptions of ment. Lancet 1986; 327: 307–310.
factorial models for analysis of variance. Appl Statist
1973; 22: 392–399.

Downloaded from shm.sagepub.com at RMIT UNIVERSITY on March 14, 2015

You might also like