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

net/publication/237350967

Experimental methods used in system identification of civil

engineering structures

Article · January 2004

CITATIONS READS
128 2,077

1 author:

Reto Cantieni
rci dynamics Structural Dynamics Consultants
39 PUBLICATIONS 355 CITATIONS

SEE PROFILE

All content following this page was uploaded by Reto Cantieni on 20 November 2017.

The user has requested enhancement of the downloaded file.

 EXPERIMENTAL METHODS USED IN SYSTEM
IDENTIFICATION OF CIVIL ENGINEERING STRUCTURES

Reto Cantieni, rci dynamics, Structural Dynamics Consultants, Switzerland

[email protected]

Abstract
To experimentally identify the dynamic characteristics of a structure, also referred to as system
identification, two methods are available: Forced Vibration Testing (FVT) and Ambient Vibration
Testing (AVT). The basic ideas of these methods are shortly presented. The main part of the paper
deals with practical problems which are to be overcome when performing such a system identifica-
tion test. To achieve reliable, good quality results, a number of parameters have to be optimized
when planning an experimental system identification investigation. Among others, such parameters
may be the means and location of excitation, the density of the measurement point grid, the sam-
pling rate and the length of the time window. As an illustration, several examples of good and
worse tests on bridges and other civil engineering structures are discussed.

1 Forced Vibration Testing

1.1 Basics
With Forced Vibration Testing (FVT) the structure to be identified is artificially excited with a
forcing function in a point i and its response yk(t) to this excitation is measured together with the
forcing signal xi(t) (Fig. 1). Transformation of these time signals into the frequency domain and
calculation of all Frequency Response Functions (FRF’s) Hik between the response and the forcing
function time signals yields the Frequency Response Matrix, also referred to as Transfer Matrix,
H(iω) (Figs. 2 and 3).

Figure 1 Forced Vibration Testing Scheme. Figure 2 Calculation of the Frequency Re-
sponse Function Hik.
For linear systems, the Frequency Response Matrix is diagonal. This means that it suffices to either
determine one row or one column of this matrix (Fig. 3). The choice is to either keep the excitation
point constant and rove the response points over the structure or vice versa. Because it is not so
easy to move the exciters used in civil engineering investigations, the first method is preferred here.
In mechanical engineering, where the structures to be tested are comparatively smaller and easy to
excite, e.g. with a hammer, the latter of the procedures mentioned is more common.
 The Frequency Response Matrix contains all the in-
formation necessary to determine the dynamic natural
properties of the structure under investigation (natural
frequencies and the associated mode shapes and
damping coefficients). Dedicated software packages
are available on the market to extract these modal
parameters from the results of a Forced Vibration
Test.

Figure 3 The Frequency Response Matrix.

1.2 Excitation
Generally speaking, the means of excitation has to be chosen such as to
• excite all natural frequencies of interest,
• be significantly larger in effect than any other “unwanted” excitation
(because: the processing procedures are based on the assumption that the measured, artifi-
cial excitation is the only source of excitation during the tests).
Broad-band vibration generators excite all natural vibrations of the structure in this frequency band
at the same time. Examples are impulse hammers and servo-hydraulic or electro-dynamic shakers
generating random or swept-sine type forces. Narrow-band vibration generators excite one specific
frequency at a time. Mechanical devices using counter-rotating masses can be mentioned here. Of
course, hydraulic or electric shakers can also be used as narrow-band exciters.
Broad-band exciters are very time effective, but they have to have (relatively) more energy dispos-
able than narrow-band exciters. These devices distribute their energy on many frequencies at a
time. Using a narrow-band exciter is very time consuming, but such a device concentrates all the
energy available into a specific frequency.
To excite civil engineering structures, hydraulic and electric shakers are better suited than ham-
mers:
Compared with mechanical structures, the fundamental natural frequency of a civil engineering
structure is low. The average value, e.g. for some 200 highway bridges in Switzerland is f ≈ 3 Hz
[1]. The frequency resolution to be achieved with an FVT investigation has hence to be high, let's
say Δf ≈ 0.01 Hz. This resolution is directly related to the length of a time window to be trans-
formed into the frequency domain: Δf = 1/T. The length of this window has hence to be at least
T = 100 s in this case. And: The quality of the FRF's determined also depends on the number of
averages which can be performed when transforming the time data into the frequency domain.
Something like 10 is a good value here. Considering an overlap of 50%, at least 500 s of stationary
structural response has then to be generated to determine a reasonably well averaged FRF when
investigating a "standard" bridge with a fundamental natural frequency f ≈ 3 Hz. This is not possi-
ble with using a hammer.
The equipment shown in the Figures 4 and 5 produces vertical dynamic forces with an amplitude of
F = 5 kN for frequencies f > 2.3 Hz. This is well suited to excite e.g. bridges with a total length of
up to 100 m. For larger bridges, larger shakers are available (F = 20 kN for f > 1.8 Hz). To excite
structures like e.g. concrete floors with f > 5 Hz, electro-dynamic shakers are used. To excite struc-
tures like e.g. dams in the horizontal direction, either servo-hydraulic shakers or devices operating
with counter-rotating masses can be used.
 Figure 4 Servo-hydraulic Figure 5 Hydraulic power pack used to drive the shaker shown
shaker. in Figure 4.

1.3 Response
The type of sensor chosen for the response measurement has to fit the requirements concerning
sensitivity and frequency range. Also because they are much easier to apply and rove over a struc-
ture, accelerometers are the best choice in most cases. Measuring displacement in many points is a
very cumbersome task for civil engineering structures. Velocity transducers are well suited for
structures exhibiting a fundamental natural frequency f > 4.5 Hz. (Recent attempts try to take care
of the "bad" low frequency amplitude and phase behavior of such sensors with using digital correc-
tion procedures. The practical performance of this still has to be looked at in detail.) Most civil en-
gineering structures exhibit lower frequencies. Therefore, highly sensitive accelerometers are
mainly chosen to investigate such structures (10 V/g). Piezoelectric sensors are suited for structures
with a fundamental natural frequency f > 1 Hz. For structures exhibiting lower frequencies, sensors
of the force balance type should be used.
As a next point, the measurement direction(s) and the measurement point grid density have to be
chosen. The basic rule here is: Information on the mode shapes is available in measured points and
directions only. This choice can be made in a much more reliable way when based on the results of
a preliminary Finite Element analysis of the structure. In most of the cases discussed later, this FE
analysis was anyway the first step of the procedure, because the major goal of the experimental
system identification was to update the preliminary FE model based on the experimental results.
This updated FE model could subsequently be used as a basis to identify problem solutions per-
forming parameter studies.
It can be seen from the examples discussed later that the number of measurement points can be as
high as 200 to 300. The number of degrees-of-freedom to be measured is even higher in cases
where it is necessary to measure in two or three directions per point. It is therefore standard prac-
tice to simultaneously use a limited number of sensors and to rove this set of sensors over the struc-
ture until the measurement point grid is completed. A test is therefore separated into several setups.
As the forcing signal is always measured too, there is no problem for the processing software to
subsequently glue together the information gathered from the different setups.
2 Ambient Vibration Testing
2.1 Basics
No artificial exciter is used with Ambient Vibration Testing (AVT), also referred to as output-only
modal analysis or natural-input modal analysis. The response of the structure to ambient excitation
is measured instead. With civil engineering structures, ambient excitation can be wind, traffic or
seismic micro-tremors. The more broad-band the ambient excitation, the better the results. Other-
wise, there is some risk that not all natural frequencies of the structure are excited.
Generally speaking: The information resulting from the force input signal xi(t) with FVT investiga-
tions is replaced with the information resulting from the response signal yR(t) measured in a refer-
ence point R (Fig. 6).
The first software package to extract modal parameters from AVT investigations has been devel-
oped by a civil engineer in the early nineties of the last century. Today, there are several packages
on the market making use of the frequency domain procedures shown schematically in Figure 7.
One of them offers more sophisticated methods like FDD (Frequency Domain Decomposition) and
EFDD (Enhanced FDD), the latter also including estimation of damping values. These methods
have been protected by a US patent recently (www.svibs.com).
However, the most recent signal processing tools are not based on an analysis in the frequency do-
main as shown in the figures below. Stochastic Subspace Identification (SSI) is a method working
completely in the time domain. Basically, a multi-order model is looked for which synthesizes the
measured time signals in a optimum way. This method has especially been developed for AVT
investigations.
Concerning response measurement requirements, the same basic rules apply as for FVT investiga-
tions. In addition, it is wise to use more than one reference point unless the structure to be tested is
very simple. The risk of the reference point sitting in a node of one or more modes can thus be re-
duced significantly. If response measurements are three-dimensional, at least one 3D-point has to
be chosen as a reference point.
As a rule of thumb, the length of the time windows acquired should be 1'000 to 2'000 times the
period of the structure's fundamental natural vibration. This a simple but very important rule of
thumb. Experience shows that many investigators do not care about this. But: You cannot harvest
feathers from a frog! Therefore: please make sure that your time windows are long enough!

Figure 6 Ambient Vibration Testing Scheme; R is Figure 7 Calculation of the cross relationship
a reference point, k is a roving point. between the reference point R and roving re-
sponse point k signals.
3 Forced Versus Ambient Testing
The main advantage of FVT is the fact that this method provides "scaled" results. Because the input
force is measured, information on the mass and stiffness matrices of the structure is gathered. This
allows automated updating of FE models. Model updating using the results of an AVT investiga-
tion is possible with manual techniques only.
The main advantage of AVT is the fact that no artificial excitation is necessary. This makes such
tests comparatively cheap. In addition, AVT investigations can be performed without embarrassing
the normal user. This fact is very important e.g. for highway bridges.
Ambient excitation is of the so-called multiple-input type. Wind, traffic and micro-tremors are act-
ing on many points of a structure at the same time. In the contrary, a forced vibration is usually of
the single-input type. For small structures, this difference is not important. The "large-
structure"example "Westend Bridge" is presented here to illustrate the limits of a single-input FVT
investigation. For large and complex structures, AVT has hence an advantage on the excitation
side. AVT offers multiple-input excitation "free of charge".
Ambient excitation being non-controllable usually results in a lack of stationarity. This may lead to
problems due to the non-linearity of the structure (no civil engineering structure behaves in a really
linear way). In case of the excitation amplitude being significantly different for each of the setups,
a certain scatter in the results may occur. This is not the case for FVT where the structural vibra-
tions induced can be kept stationary.

4 Examples of FVT Investigations

4.1 A Short Bridge: Bridge on the Aare River at Aarburg [2]
This single span arch bridge
crosses the Aare River at Aar-
burg. The completely
clamped-in arch is a rein-
forced concrete structure
from 1912 (Robert Maillart)
whereas the bridge deck has
been rebuilt in 1968 as a pre-
stressed structure. The bridge
deck is horizontally free at
both abutments. This results
in the structure being simple Figure 8 Aare Bridge Aarburg.
(one span) and exhibiting simple and clearly defined boundary conditions. Therefore, the results of
the single-input FVT investigation were of very good quality. Figures 9 and 10 show the compari-
son between an FE-analysis and the FVT-investigation results.
The shaker shown in Figure 4 was located on the downstream (right-hand side in Figure 8) curb
10 m from the abutment. The forcing signal was of the random type with an upper band limit of
fc = 36 Hz. The measurement point grid consisted of 105 and 41 3D-acceleration measurement
points on the bridge deck and the arch respectively. One 3D-point was always located in the driving
point (underneath the shaker). Another five 3D-points were roved over the structure. Figure 9 gives
an idea of the resulting grid density. The sampling rate was s = 80 Hz, the length of a time window
T = 51.2 s, the number of averages was 8, the total net testing time per cycle was 7 minutes, the
total test took four nights between 10 pm and 5 am.
 Figure 9 Frequency and shape of the first three Figure 10 MAC matrix (FVT versus FE
bridge modes; FVT results to the right, updated model) for the first seven bridge modes. (scal-
FE model to the left. ing of the vertical axis: 0 to 1.).

4.2 A Long Bridge: Westend Bridge in Berlin [3]

This eight-span continuous beam bridge is part of the heavily trafficked Berlin city belt. The rela-
tively short single columns are all clamped into a flat foundation. With the exception of one column
in the middle of the bridge, which is also clamped into the bridge deck, the columns are pinned at
their top end. There are several problems with this bridge from the point of view of a single-input
FVT investigation. The bridge is too long and has too many spans. Especially bad is the short 31-
m-span in the middle of the bridge which cuts the structure into two dynamically almost "decoup-
led" parts (at least what vertical bending modes is concerned).

Figure 11 Westend Bridge Berlin. The view is op- Figure 12 Measurement point grid for the
posite to the one used in Figure 12: Measurement Westend Bridge. #75 is the driving point
points #1 to #5 are on the right-hand side of the (shaker position); the length of the individual
photo shown here. spans is indicated in m.
The reason for this somehow strange span arrangement is the fact that the spans in the left-hand
part of the photo (and in the right-hand part of the drawing shown in Fig. 12) are crossing several
railway lines. The columns had to respect the position of the rail tracks. It was not possible to in-
stall a second shaker in this part of the bridge because access to the area underneath these spans
was, a) not allowed and, b) not possible with vehicles.
Due to the heavy traffic on the bridge, the time window for the tests was 10 pm to 4 am only. Out-
side of this window, nothing could be left on the bridge. There was hardly enough time to in-
stall/remove one shaker at position #75 (Fig. 12) with the infrastructure (hydraulic power pack,
power generator) being located just underneath this point on the ground.
The main results of the test are discussed in Figure 13. Here, the view is the same as used in Figure
12.

Figure 13
Frequencies and shapes
of the first five (out of
nine [3]) modes of the
Westend Bridge as de-
rived from the FVT in-
vestigation and an up-
dated FE-model.
(MAC for the modes
shown is in the MAC =
0.74…0.83 range.)
It can be seen that:
a) the frequency of f1 is
too low to properly be
excited,
b) the second torsional
mode f2 is not properly
excited at the right-hand
side of the bridge,
c) the bending modes are
well excited in the area
of the shaker but not at
the right-hand side of the
bridge,
d) the natural vibration
dictated by the 31 m mid-
dle span is properly ex-
cited (f5).
4.3 A Dam: Norsjö Dam in Sweden [4]

Norsjö Dam is a cylindrically shaped

reinforced concrete structure with a
length at the crest of 170 m and a
maximum height of roughly 46 m
(Fig. 14). The radius of curvature of
the dam upstream face is r = 110 m.
The dam width is 2.5 m at the crest and
5.5 m at the bottom (Fig. 15).
The results of an FVT investigation of
this dam using a servo-hydraulic
shaker producing a horizontal force
amplitude f = 32 kN for f < 1.5 Hz
were of extraordinary good quality.
Considering the results of a prelimi- Figure 14 Norsjö Dam.
nary FE analysis, the shaker was lo-
cated some 60 m from the abutment at the first spillway wall (left-hand side on the photo). The
forcing signal was of the random type with a band-limit at fc = 20 Hz. The tests could only be per-
formed when the powerhouse was out of operation.
The measurement point grid consisted of 270 3D-acceleration measurement points distributed over
the dam crest and eight further levels between crest and foundation (it can be seen from Figure 15
that the whole downstream face of the dam is accessible), spillway walls and powerhouse. Four
3D-points were roved over the structure. The sampling rate was s = 100 Hz, the length of a time
window T = 41 s, the number of averages was 8, the total testing time per cycle was about 6 min-
utes, the total test took two days (of a weekend) and four nights.
Twelve modes with f = 3.0…13.5 Hz could be identified. Updating of an FE model was possible
with very good results [4].

Figure 15 Norsjö Dam cross Figure 16 Results for mode No. 5 (FEM and FVT shapes overlaid!).
section.
5 Examples of AVT Investigations
5.1 A long Bridge: Ganter Bridge [5], [6]

This two-lane highway bridge with a total length of

678 m has eight spans with a length between 35 m and
174 m. The height of the tallest pylon is 176 m. Figure
18 shows the measurement point grid chosen for an
AVT investigation. Including 192 points, this covered
the bridge deck on its entire length plus the two tallest
piers.
At the bridge deck, the upstream sensors were of the
3D-, the downstream sensors of the vertical 1D-force
balance type. As traffic remained open during the tests
the accelerometers were placed inside the box girder.
3D sensors were fixed to the piers through profes-
sional climbers rappelling from the pier top (see Fig.
17). The two reference points chosen were equipped
with a 3D and a 1D sensor respectively. Three 3D and
three 1D sensors were roved in pairs along the struc-
ture.
The big challenge of this test was to organize the cable
management. The cable length available was 300 m
per 3D and 1D sensor and the bridge deck was not Figure 17 The Ganter Bridge
accessible for any piece of equipment (no curbs). It
was therefore necessary to break down the test into four phases. Phase 1: The measurement center
was located close to pier 2 on the ground (Station 1 in Fig. 18). Measurement of the bridge deck
between the north abutment and the reference points. Phase 2: The measurement center was located
between piers 4 and 5 (Station 2). Measurement of the bridge deck between the south abutment and
the reference points. Phases 3 and 4: Measurement of piers 4 and 3 with the measurement center
being located at stations 2 and 1 respectively.

Figure 18 Ganter Bridge: The measurement point grid. Figure 19 The first eight modes.
The sampling rate chosen was s = 20 Hz, the length of a time window 53 minutes for the bridge
deck and 27 minutes for the piers. The test took ten working days. A total of 25 modes could be
identified in the frequency band f = 0.40…3.88 Hz. AVT proved to be a very good method to iden-
tify the dynamic parameters of such a large structure exhibiting very low natural frequencies.

Figure 20 Shapes of the Modes No. 1, 3 and 5 (elevation looking upstream and plan view).

5.2 A Short Bridge: Regensdorf Bridge Zurich [5]

The Regensdorf Bridge is a 30 m long

and 14 m wide, skew steel-concrete
composite structure. It consists of six
riveted main girders and a concrete
deck.
The bridge undersight being inaccessi-
ble (railway lines), the challenge of the
AVT investigation was to organize the
measurements without disturbing the
(heavy down-town Zurich) traffic too
much. The existence of sidewalks is Figure 21 The measurement point grid.
very helpful in such a case.
Figure 21 shows the the measurement
point grid. The two reference points are
indicated with solid squares. Seven
sensors were roved over the bridge
deck. In a first phase, the curbs and the
bridge centerline were measured with-
out any interference to the traffic.
Based on the intermediate (on-line)
results it was decided to measure two
additional lines. For this, the traffic
flow was restricted to the remaining
part of the bridge deck.
The sampling rate was s = 80 Hz, the
length of a time window 7 minutes.
The test was accomplished in one day. Figure 22 The first five bridge modes
The results are summarized in Figure 22. A total of 7 modes could be identified in the frequency
band f = 4.80…18.55 Hz . Extraction of the mode shapes was based on vertically oriented meas-
urements only, although some of the sensors were of the 3D-type.

5.3 A Bell Tower: St. Peter and Paul Church, Zurich

The tower of St. Peter and Paul Church is a 52 m high stone structure built in 1895. The measure-
ment point grid used for an AVT investigation consisted of three horizontal degrees of freedom
(accelerations) measured on the levels 2 to 6 as shown in Figure 23. Three sensors remained as
reference points at level 6 while three further sensors were roved over the levels 2 to 5 in four set-
ups.

Figure 23 Measurement point grid. Three sensors per meas- Figure 24 The measurement center
urement level. located at level 5.
To define the instrumentation necessary to determine the mode shapes of a structure like a tower or
a tall building, several assumptions can be made: a) the vertical components can be neglected, b)
the rectangular shape of the structure's cross section for a certain level above ground remains un-
changed, and, c) the movements of the structure in the horizontal plane are small. It then suffices to
measure three of the eight possible degrees of freedom of a rectangular cross section: x and y in
one corner and y in a neighboring corner (Fig. 23). Advanced software packages allow to determine
the remaining DOF's using so-called "slave node equations".
The sampling rate was s = 25 Hz, the length of a time window 30 minutes. The test was accom-
plished in one afternoon. Nine modes with frequencies f = 2.30…8.9 Hz could be identified (Fig.
25).
 Figure 25 St. Peter and Paul bell tower: Modal parameters for modes No. 1, 4, 5 and 8

6 References

[1] Cantieni, R. "Dynamic Load Tests on Highway Bridges in Switzerland - 60 Years Ex-
perience of EMPA". EMPA Report No. 211, (1983).
[2] Cantieni, R., Deger, Y., Pietrzko, S., "Modal Analysis of an Arch Bridge: Experiment,
Finite Element Analysis and Link". Proc. 12th International Modal Analysis Confer-
ence (IMAC), (1994) 425-432.
[3] Deger, Y., Cantieni, R., Pietrzko, S.J., Rücker, W., Rohrmann, R., "Modal Analysis of
a Highway Bridge: Experiment, Finite Element Analysis and Link". Proc. 13th Inter-
national Modal Analysis Conference (IMAC), (1995) 1141-1149.
[4] Cantieni, R. Assessing a Dam's Structural Properties Using Forced Vibration Testing,
Proc. IABSE International Conference on Safety, Risk and Reliability - Trends in En-
gineering, Malta, March 21-23 (2001).
[5] Felber, A.J. Cantieni, R., Introduction of a new Ambient Vibration System - Descrip-
tion of the System and Seven Bridge Tests, EMPA Report No. 156'521, (1996).
[6] Felber, A.J., Cantieni, R., Advances in Ambient Vibration Testing: Ganter Bridge,
Switzerland, Structural Engineering International (6), Number 3, (1996) 187-190.

View publication stats