Tracking deflection in the field using optical system: a case study

Yan Xu1, Farhad Huseynov2,3, James M. W. Brownjohn1, Eugene J. O’Brien3, David Hester4
1
College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, UK
2
Full Scale Dynamics Ltd, Sheffield, UK
3
School of Civil & Structural Engineering, University College Dublin, Dublin, Ireland
4
School of Planning, Architecture and Civil Engineering, Queen’s University Belfast, Belfast, UK
email: [email protected], [email protected], [email protected], [email protected],
[email protected]

ABSTRACT: Monitoring displacement of in operation bridges is practically challenging but potentially very useful for
condition assessment and decision support. The primary difficulties are in finding fixed physical reference points and, for the
majority short span bridges under normal operation, the mm-level magnitudes of displacement under normal operating
conditions (e.g. standard truck loading). With rare possibility for physical connection between a reference and a bridge, non-
contacting technologies such as GPS need to be used. Other options include total station and more exotic technologies of laser
interferometer and radar have also been tried.
There are drawbacks for each technology related to limited sample rate (for total station) and signal to noise ratio (for GPS)
while radar and laser are expensive and require specialist users. With advances in computing power, optics-based systems are
becoming popular, relying on a standard lens but with capability to track multiple positions with potential to recover
deformation with high spatial resolution.
This paper reports the experiences of the authors exploring the suitability of a commercially available optics-based system in
terms of spatial and temporal resolution and sampling and in challenging field conditions required for long term monitoring.
For example issues such as stability of camera mounting (e.g. in wind) and varying lighting conditions while not problematic
in a laboratory govern performance in the field. The paper tracks a sequence of experiments moving from lab to field, ultimately
moving up to a field test on a road bridge in Devon.
In each case the capabilities and limitations of the system have been critically examined. The study has defined both
limitations and capabilities, while defining best approaches for use and at the same time providing some useful performance data
on the subject bridges.

KEY WORDS: Bridge Field Testing, Strain Measurements, Finite Element Modeling, Load Distribution Factors, Case Study

1 INTRODUCTION even total stations. However, mainly due to cost and reliability
Deformation of a structure under operational and extreme issues none of these has become a standard feature of a
loads represents its performance, and is a rich source of modern structural health monitoring system, where GPS
information that enables both diagnosis of present condition receivers and LVDTs are standard.
and prognosis for future management. A number of research teams have explored the capability of
Often the time and space derivatives of deformation are vision-based systems using standard optical lenses. Probably
measured using accelerometers and strain gauges but direct the earliest implementation of this technology was at Humber
measurement of displacement is feasible using a range of Bridge to track deformations of the main span using two
technologies all of which measure relative motion between independent systems, one based on a threshold detection with
two points, almost always one which is a fixed reference point a high-contrast image, the other (the ‘vision system’) based on
off the structure. Differences among the technologies relate to: predictive target tracking [3]. There have since been several
systems based on image processing, but the most robust
 Distance between structure and reference point
system and capable system appears to the commercial video
 Range, resolution and achievable accuracy
gauge’ developed from the Humber ‘vision system’ whose
 Achievable sample rate evaluation is described in this paper.
 Cost, reliability and robustness
 Effects of atmospheric conditions 2 CAMERA SYSTEM USED AND LABORATORY TRIALS
 Effects of local disturbances
 Requirements for physical location on a structure In preparation for deploying the system in the field some
Historically, optical systems have been used for static simple lab trials were undertaken. Figure 1(a) shows the
positional surveys were used for continuous monitoring of the Imetrum ‘video gauge’ with a single camera set up in the
new Tagus River bridge [1], before which seismometers had laboratory. The camera is mounted on a tripod in the right of
been used in California [2]. the figure, it is pointed at a paper sliding target attached to a
Other methods have included fluid levelling systems ruler. The image seen by the camera is visible on the
(installed, but no longer operation, Tsing Ma and Tamar computer screen in the left of the figure. In this test the
suspension bridges), Radar and laser Doppler vibrometer and camera is fitted with 16 mm lens and is approximately 2.2 m
from the target. In the software the user defines a
measurement plane (in essence this allows the system to 150mm square stuck on the side of the bridge at mid-span. In
calculate the size of a pixel in mm). Then the user identifies a this test there was an 85mm lens on the northern camera, and
zone in the frame that they would like to track, this is referred a 180mm lens on the southern camera.
to as the ‘target’. This can be a ‘natural’ feature of the
structure or an artificial target similar to the one shown in the
figure and the system just tracks the movement of the target
between successive frames. A ‘target’ needs to represent a
minimum size of 40 pixels by 40 pixels in the camera CCD
array. Figure 1(b) shows a close up view of the paper target
used on the ruler. In the test presented here the target was
moved 100mm down-wards in increments of 10mm, then the
target was moved up again in increments of 10mm. Once
returned to its starting position it was left stationary for
approximately 100 seconds and then the test was terminated.
The results of the test are shown in Figure 2.

Figure 3. Test setup.

(a) (b)

Figure 1. Experimental setup (a) general view of experimental

set up (b) close up of paper target on ruler.
20
Vertical movement (mm)

-20

Figure 4. Test configuration.

-60
3.2 Results
-100 Rather than using the distance form camera to target,
dimension lines marked on the target were used to calibrate
0 100 200 300 400
Time (s)
the coordinate frames in the video gauge. Figures 5(a) & (b)
show the (mid-span) horizontal displacement returned by the
system (over 400 seconds) for the north and south sides of the
Figure 2. Displacement plot resulting from the test.
bridge respectively. In both plots there appears to be
3 EARLY FIELD TRIALS extraneous high frequency noise and a low frequency drift, i.e.
engineering judgement tells us the bridge should not be
Following the lab trials the camera system was evaluated to moving ±1.5mm horizontally (i.e. longitudinally).
test its ability in the field and to develop procedures to obtain Figures 6(a) & (b) show the mid-span vertical displacement
the best results. To begin, the system was used to track the returned by the system for the north and south sides of the
displacement of a small road bridge close to the University in bridge respectively. The plots in Figure 6 show the raw data
Exeter, and the results of this early monitoring is reported in and an ‘Average’ plot, (the ‘Average’ plot is obtained using a
this section. moving average filter of an appropriate span). If the average
plot in Figure 6(b) is examined it can be seen that the
3.1 Test setup displacement starts at zero, increases to 1mm after 250
The video gauge can record data from up to four cameras seconds and then falls back toward zero. During this time
simultaneously and this capability was examined in this test there were only a few cars crossing the bridge i.e. there was
by deploying a pair of cameras, one each side of the bridge as no loading that could have caused such vertical deflection.
shown in Figure 3, (dimensions are approximate). Figure 4 The amplitude of the high frequency movement in Figure 6
shows the test set up, with artificial targets are approx. look too large to be dynamic bridge movements and to
confirm this, the frequency content is shown in Figure 7. On discussion with manufacturer Imetrum it was brought to
From a previous vibration study of the bridge it is known that our attention that the noise/ drift experienced was probably a
the first bending and first torsion frequency of the bridge are result of varying lighting conditions. Therefore in the next
at 3.1 Hz and 4.9 Hz respectively but there are no any peaks at field test, provision was made to use a target under the bridge
these frequencies in Figure 7, which shows only noise. So, as and not subject to varying lighting conditions. The results of
with the horizontal displacement, the vertical data obtained in this testing are given in section 4.
this video gauge deployment are not usable.
(a) Horizontal displacement of north side (85mm lens) 4 DISPLACEMENT MONITORING ON CONCRETE BRIDGE
3

2
Following the early field trials described in section 3 we
1
moved on to a test where we were able to mount the target on
0
the soffit of the bridge which minimised issues with varying
light levels.
-1

-2
0 50 100 150 200 250 300 350 400 4.1 Bridge used
Horizontal displacement of south side (180mm lens)
The bridge used in this test was the three span concrete road
Horizontal displacement (mm)

3
(b)
2
bridge shown Figure 8(a). To be able to install the optical
1
target, a bridge with a relatively low soffit and safe access was
0
required. Ideally it would have been good to use natural
texture and the paper target, however panels had been
-1
installed on the soffit of the bridge which meant there was not
-2
0 50 100 150 200 250 300 350 400 a sufficiently large vertical surface to work with. The flow of
Time(s)
the river in the photo is controlled by a weir downstream of
Figure 5. Horizontal displacement (mm) reported by system in the bridge. Normally the water level is held constant just
first 400 seconds of the test, (a) north side of bridge, (b) south below the level of the footway on the right of the photo.
side of bridge. However, to facilitate river works the level of the level of the
river was dropped for a short period, thereby exposing the
Vertical displacement of north side (85mm lens)
(a) 3 sand bank on the right of the photo which we were able to use
Vertical displacement (mm)

2
as a working platform. The bridge has beam and slab deck
and Figure 8(b) shows a cross section through the deck.
1
Essentially the beams are steel beams that have been encased
0
in concrete, the beams span 19m and the transverse spacing
-1 between the beams is 1.5m.
Raw Average
-2
0 50 100 150 200 250 300 350 400
Time(s) 4.2 Test setup
Vertical displacement of south side (180mm lens)
(b) 3 The deflection monitoring was carried out at mid-span in
Vertical displacement (mm)

2 the northern most span, on the 3rd beam from the east side.
1 Figure 9 shows a schematic of the test site, the location of the
0
cameras and the position of the target. In Figure 11 the camera
tripods are visible in the right of the figure and the step ladder
-1
Raw Average under the bridge approximately indicates the position where
-2
0 50 100 150 200 250 300 350 400 deflection is being monitored.
Time(s)

Figure 6. Vertical displacement (mm) reported by system in

first 400 seconds of the test, (a) north side of bridge, (b) south
side of bridge.
30
North side (85mm lens)
20 South side (180mm lens)
Power spectral density (dB)

10
(a) Bridge used
0

-10

-20
(b) Cross section of deck
-30
Figure 8. Bridge information, (a) bridge used in the test, (b)
-40
0 5 10 15 20 25 Cross section through the deck
Frequency(Hz)

Figure 7. Frequency content of vertical displacement signals

seen in Figure 6.
 Figure 9. Schematic of test setup.

A few days before the main test some preliminary testing

was carried out where the camera was placed on the sand bank
in the fore ground of Figure 8(a), however the week of the test Figure 12. Equipment mounted on bridge soffit.
was particularly windy and the data from this preliminary 4.3 Results
testing were poor, i.e. there was a lot of drift and noise, see
The test lasted approximately 15 minutes and the results
Figure 10. It was felt that this was at least partly due to the
from the camera system are shown in Figure 13. In that time a
cameras being shaken by the wind. Based on the these data, a
number of trucks passed and these generated maximum
better method of mounting the lens was implemented which is
deflections of approximately 3mm. These peaks are evident in
reported later in section 5.2, but not in time for repeat
the top row of Figure 13. The results in Figure 13 are far
measurements on the same bridge. For this test the camera
clearer than those shown in Figure 6 or Figure 10. There is
recording the target (circled in Figure 11) was placed in the
minimal drift, with the displacement steady at 0mm except
lee of the abutment wing wall thereby sheltering it from the
when a heavy lorry crossed the bridge (on three occasions).
wind, (the camera shown uncircled was video-recording
Based on the approximate weight of the truck and the
traffic so buffeting by the wind was not an issue). Figure 12
properties of the bridge, a displacement in the region of 3mm
shows the testing equipment mounted on the bridge soffit.
would be expected, so the magnitude of the displacement
Essentially a large clamp was used to fix a steel angle to the
predicted by the camera system is credible.
underside of the beam. The optical target was attached to the
steel angle using a magnet.
Vertical displacement

Displacement time history 1

Displacement(mm)

3
0

2 -1 Swipe1 Swipe2 Swipe3

-2
Displacement(mm)

1 -3
600 650 700 750 800 850
0 Time(s)
Swipe 1 Swipe 2 Swipe 3

-1 1 1 1
Displacement(mm)

Displacement(mm)

Displacement(mm)

0 0 0
Horizontal Vertical
-2 -1 -1 -1
0 100 200 300 400 500 600
Time(s) -2 -2 -2

-3 -3 -3
Figure 10. Results of preliminary testing in windy conditions.
575 580 585 590 700 705 710 715 810 815 820 825
Time(s) Time(s) Time(s)

Figure 13. Results from the test.

5 SUBSEQUENT IMPROVEMENTS IN PROCEDURE

5.1 Improved results with auto exposure
The cause of the drift problem described in section 3 was
believed to be due to changing lighting conditions. A
Figure 11. Test arrangement,
software update with auto exposure feature was released after
the measurements described in sections 3 and 4, and the test
described in Section 3 was repeated, with much clearer
results. Figure 14 shows the test set up, this time both cameras
were pointing at the same target, and Figure 15 shows the
results. Comparing Figure 15 to Figure 6 shows much reduced
noise and drift. The displacement stays around 0mm, except at
50, 60, 100 and 110 seconds where single or multiple car/van
events resulted in displacements in the region of 0.2mm. 6 DISCUSSION
Based on the properties of the bridge and the approximate Accurate and reliable measurement is a prerequisite for
weight of a car/van this magnitude of displacement makes structural condition assessment and health monitoring and
sense. new technologies require careful evaluation in real world
conditions to iron out the bugs, identify limitations and factors
on performance and build confidence in their use. So far
optics-based monitoring has not been widely adopted but we
see it as a major capability for the future of asset management
through performance observation. Because the Imetrum
‘vision system’ has a long pedigree (since 1990) and appears
to be highly capable we have gone to considerable lengths to
evaluate it through a series of experiments. We have
confirmed the capability but have found this to be strongly
affected by the way it is deployed. Issues with lighting
conditions have been mitigated with software upgrades but it
remains necessary to take extreme care in providing a stable
mounting for the lens and camera. Future evaluations will
focus on extreme range and resolution, plus comparison with
standard technologies of GPS and accelerometers.
Figure 14. Test configuration.

0.4
85mm 180mm
ACKNOWLEDGEMENT
Displacement(mm)

0.2

0
The research leading to these results has received funding
from the People Programme (Marie Curie Actions) of the
-0.2
European Union's Seventh Framework Programme
-0.4
0 20 40 60 80 100 120 140
(FP7/2007-2013) under grant agreement n° 330195. The
Time(s) authors would also like to acknowledge the Bridge Design
Figure 15. Results from test using auto exposure feature of the Group of Devon County Council, led by Kevin Denthtih, for
software. their support of the work.

5.2 New camera mounting

As mentioned in section 4.2 the results obtained in windy
conditions were unsatisfactory (see Figure 10) so an REFERENCES
improved, rigid mounting was designed. Figure 16(a) shows
the original mounting of 300mm lens, which is in effect a [1] J. Marecos, M. Castanheta, and J. T. Trigo, “Field observation of Tagus
River suspension bridge,” ASCE J. Struct. Eng., vol. 95, no. 4, pp. 555–
cantilever and figure 16(b) shows the new rigid mounting. 583, 1969.
Although wind is still an issue, better results have been
[2] D. S. Carder, “Observed vibrations of bridges,” Bulletin Seismological
obtained with the new mounting and with the realisation that Society of America, vol. 27. p. 267–, 1937.
stable lens mounting is a critical feature of optics-based
[3] J. M. W. Brownjohn, M. Bocciolone, A. Curami, M. Falco, and A. Zasso,
measurements. “Humber Bridge Full-Scale Measurements Campaigns 1990-1991,” J.
Wind Eng. Ind. Aerodyn., vol. 52, pp. 185–218, 1994.

(a) (b)
Figure 16. Lens mounting, (a) old approach, (b) new approach