Structural Dynamics

Instrumentation for civil engineering dynamic testing

Prof Pilate Moyo, PrEng, FSAAE, FSAICE
 Overview

• This lecture outlines some of the principles for selecting and using
instrumentation for dynamic testing of civil structures. In the second
lecture examples are given of a number of applications which can include:

2
 Introduction

The choice of best equipment depends on the application but theses are the
necessary system components:
• Vibration sensors
• Shaker or instrumented hammer
• Cabling to acquisition system
• Acquisition system comprising:
Power supply, amplifier and analog filtering module
Analog to digital converter
Acquisition control and processing software
3
 Accelerometer

The choice of components depends on the range of

applications, the experience of the tester and the available
budget. The most important choice is likely to be of the
vibration sensor, which for civil dynamic testing is (with rare
exceptions) an accelerometer

4
 Accelerometer

There is now a bewildering range of products including:

Strain gauge accelerometers (Kyowa)
Capacitive accelerometers (Setra)
Quartz-flexure servo accelerometers (e.g. Honeywell)
Piezo-electric accelerometers (B&K, Endevco, PCB, Kistler,
Dytran)
MEMS accelerometers (Applied Mems, Crossbow)
Seismometer (Guralp)

5
 Accelerometer

The factors affecting choice are:

Cost: this should consider all costs resulting from the choice e.g. special considerations
on mounting, cabling and power supply
Range: Civil structures rarely experience more than 1g so a larger range may have a
negative effect on ‘dynamic range’.
Sensitivity: This is important for matching the signal range to the input range of the
acquisition system, but a high sensitivity (e.g. 10V/g) is not always necessary for sensing
weak signals since acquisition system ranges can be varied and pre-amplification
applied.

6
 Accelerometer

Frequency range: Some accelerometers (e.g. Piezoelectric types) will not sense
very low frequency response e.g. below 1Hz which are important for low-
frequency structures. Piezo-electric accelerometers with better low frequency
performance are more expensive (and fragile).
Cabling limitations: Strain gauge sensors have limits of around 100m on cable
lengths, and for some accelerometers (especially piezoelectric), cables can pick
up noise or worse if mishandled.
Size: This is rarely a problem as civil structures are large; it will usually be
necessary to make a baseplate for locating the accelerometers.
Attachment: Since civil structures rarely experience more than 1g there is
usually no need to fix to horizontal surfaces. Magnetic mounts or plates for
gluing onto vertical/inclined surfaces may be useful.

7
 Accelerometer

Noise floor: For measurement of buildings, dams and other structures under
ambient conditions, and particularly for measuring ground-borne vibrations,
users should choose sensors with low noise floors. The best accelerometers
can detect signals with amplitude around 1g which is enough for all but the
most demanding application where seismometers my be needed.
We predominantly use Honeywell QA650/700/750 quartz-flex accelerometers
because they can be used with almost unlimited cable lengths, do not pickup
noise in the cables, measure down to DC and can resolve down to 1µg. They
are, however relatively expensive (about USD R15k), require a small custom-
built stage of signal conditioning and a US export license.

8
 Accelerometer

Operation of Quartz-flex accelerometer:

As the base plate moves the proof mass will remain

stationary (hence move relative to the base). Any
relative deflection sensed by the position detector will
lead to a current being passed through the ‘torquer
coil’ to zero the relative deflection. This current is sent
through the external load resistor: its value is
proportional to the force needed to accelerate the
proof mass with the base and the resulting voltage
dropped across the load resistor is recorded by the
acquisition system

9
 Excitation and force measurement

For civil forced vibration testing the most commonly used forms of excitation are
shaker and instrumented hammer.
Instrumented hammer
The Dytran 5803A ’12-pound’ instrumented hammer is a widely used excitation
device. The Piezoelectric force crystal mounted between the coloured impact tip
and hammer body has a sensitivity of about 4.5kN/V. The tip controls the
distribution of energy. For civil structures the brown tip, which is the softest, is
optimal. The softest tips deteriorate over time, the brown tip decomposing to a
toffee-like material, the green crumbling (depending in climate).

10
 Excitation and force measurement

The plots show the time history and frequency content for green (medium) tip on left and black (hard) tip on
right. Brown has an even wider force pulse and faster decay of energy with frequency.

green (medium) tip black (hard) tip

11
 Excitation and force measurement

The correct method to use the hammer is to let it impact

the flat surface (with a little assistance) with a clean,
square hit, and to catch it on the rebound to prevent a
double hit (Figure 6.1). Ideally the hammer operator
should not be on the structure being tested but in the
case of a floor, as shown, this is an impossible
requirement Figure 6.1)
Using a hammer: let drop to generate a clean, square
impact and catch on the rebound.

The method shown in Figure 6.2 is not advised.

Figure 6.2)
12
 Excitation and force measurement

Shaker
Rotating mass shakers have in the past been
used for civil structure testing. They comprise a
pair of masses on arms that rotate about the
same axis in opposite directions. As shown in
Figure 6.30) the shaker has a horizontal axis,
and load cells may be used between the shaker
body and the structure
The magnitude of the force (but not the timing) can be estimated using the formula shown on the
figure where Mo is the value of the unbalanced mass. For low frequencies extra masses are
attached but even for the maximum mass the low frequency characteristic goes with frequency
squared, while the maximum force depends on the shaker bearings and attachment; the unit
shown generates about 10kN maximum force.
13
 Shaker

• Electrodynamic shakers (Figure 6.4) are more common and can

be bought directly (from APS dynamics in the USA). The rubber
bands support the moving armature of the shaker; the force
applied to the structure is a reaction to the force generated in
the shaker. The magnitude of the force depends on the stroke of
the shaker; like the rotating mass shaker. The force is given by
-Mω2x where M is the total mass of armature and all attached
masses, x is the displacement and the (circular) frequency.

• To measure the force generated, either a force transducer may

be used between the shaker body and the structure being
excited, or an accelerometer can be attached to the armature to
record the acceleration time series. Knowing the mass, the
phase and magnitude of force signal can be recorded together Figure 6.4
with the structure response.
14
 Excitation and force measurement

The smaller APS113 shaker generates up to 150N force, the larger (and VERY
heavy) APS400 generates up to 450N. The advantage over the rotating mass
shaker is that force signals are not confined to sinusoids, they can be random,
chirp or user defined.

15
 Acquisition system

Again, there is a bewildering choice depending on combination of user

skill, budget and also the number of signal channels (i.e. number of
accelerometers).
A number of complete systems are available which take care of anti-alias
filtering, analog to digital version, data storage and some data processing.
We have chosen National Instruments equipment. With these solutions it is
necessary to build a system from components such as:
• Termination panel
• Filters
• ADC

16
 Acquisition system

Features that should be considered in an acquisition system are:

• Number of channels: in line with number of sensors
• Size/weight/power requirements: field work is easier if mains is not required and
several boxes do not have to be connected together
• Anti-alias filtering: should have sharp cut-off characteristics, preferably automatically
tracking the sampling rate
• Analog to digital (ADC) converter specification: in terms of number of bits and
simultaneous sample and hold (SSH) capability.
• For ADCs with 18 or less bits variable ADC input ranges will be necessary to optimise
dynamic range. 24 bit systems are now common.

17
 Acquisition system

ADCs sample a signal at exact

equal intervals (delta-t) and
compare the signal to a set of
2^n levels where n is the
number of bits.
Next page shows (clockwise
from top left) DP Quattro, OROS
OR35, NI-USB 9233, Geosig
GCR-16, DP mobilyser; all but
Geosig use 24-bit ADCs

18
 Acquisition system

Dynamic range
The MATLAB script alias.m is used to demonstrate issues relating to limited number of bits and aliasing.
The first plot shows a sine wave with unit amplitude sampled with an ADC having resolution of 0.5V per
bit. Signals in the ranges are truncated to
-1V to -0.5000001V -1V
-5V to 0.0000001V -0.5V
0V to 0.4999999V 0V
5V to 0.9999999V 0.5V
Hence any signal lying within the range 0 to just less than 0.5V will be read as exactly 0V. This system
represents a 4-bit ADC with range +/-4V: there are 24 =16different values that can be represented by 4
bits:
Bit level resolution is 8V/16=0.5V.
The sample rate is 16Hz i.e. 16 samples in 1 second.
19
 Acquisition system

0.5

-0.5

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time/seconds

Fourier amplitude spectrum a

n

1
0.8
0.6

0.4
0.2
0
0 10 20 30 40 50 60
frequency/Hz

4-bit ADC for +/-4V

20
 Acquisition system

The second plot shows a 12 bit ADC and a sampling rate of 64Hz. The sine wave is
resolved very well in the time axis as well as in the Voltage (vertical) axis.
With this system the bit-level resolution is 8/212 =0.002V.
The sample rate (64Hz) is faster so the better bit-level resolution is clearer.
With this system the size of signal that cannot be resolved is much smaller, now signals
between bit levels increasing by 0.002V cannot be distinguished.
The best resolution available is 24 bits. For a +/-5V range, the bit level resolution is
10/224= 0.12V. For typical accelerometer configurations with 10V/g sensitivity the
resolution of the ADC will be 0.12m/sec2 which is smaller than the noise level of even
the best-performing accelerometers which can then operate at full +/-1g range with
typical ADV voltage input ranges and retain maximum signal resolution.

21
 Acquisition system

0.5

-0.5

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time/seconds

Fourier amplitude spectrum an

0.8
0.6

0.4
0.2

0
0 50 100 150 200 250
frequency/Hz

12-bit ADC for +/-4V

22
 Acquisition system

Aliasing
The two figures show sinusoidal signals of 1Hz and 15Hz acquired by the same 16Hz sample rate. In the
case of the 15Hz signal, what the ADC sees is a signal at 1Hz which is the ‘alias’ of the 15Hz signal.
For each signal below half the sample rate, there will be one signal above half the sampling rate (Nyquist
frequency). In fact signals at 17Hz, 31Hz, 33Hz etc. are all aliased at 1Hz: there is an infinite number of
aliases.
Unless you know for sure that there cannot be any signals above the Nyquist frequency it is impossible
to know if the 1Hz signal is really 1Hz or is in fact 15Hz, 17Hz, 31Hz, 33Hz etc. etc..
This certainty is provided by inserting an analog anti-alias filter (AAF) before the ADC. The AAF ‘cut off
frequency’ above which no signal should get through, should be less than the Nyquist frequency.
If AAF is absent or imperfect a test for aliasing is to change the sample rate. Any frequencies below
Nyquist frequency that shift significantly are aliases.

23
 Acquisition system

1 1

0.5 0.5

0 0

-0.5 -0.5

-1
-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time/seconds
time/seconds
Fourier amplitude spectrum a
n Fourier amplitude spectrum an
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0 10 20 30 40 50 60 0
frequency/Hz 0 10 20 30 40 50 60
frequency/Hz

Original signal at 1Hz; no alias. Original signal at 15Hz; aliased at 1Hz.

24
 Acquisition system

Typical acquisition systems

Data acquisition using custom-built QA700

interface, DP mobilyser and PC for Tamar bridge
test

QA700 triad,
Endevco 7754-1000 triad,
Kistler coupler,
custom-built QA700 interface (with AAF)
and PC with 16 bit 16 channel PCMCIA ADC

25
 Data quality assurance

Noise levels, signal corruption, drift and tell-tale signs

Unless you use perfect equipment there is a very good chance

that transducer signals (usually accelerations) will have some
kind of noise present. This comes in several forms; the
measurements from a tall building (first page) illustrate this.

26
 Data quality assurance

SETUP9_4.MAT
0.5 0.2 0.2 -0.2

Ambient vibration data:

ch 1 (V)

ch 2 (V)

ch 3 (V)

ch 4 (V)
0 0 0 -0.3
Figure 6.5: Channel 3 (top
row, 3rd from left) has drifting -0.5 -0.2 -0.2 -0.4

and jumps in the signal, 0 1.5 0.4 -0.02

channel 16 has a long spike

ch 6 (V)
ch 5 (V)

ch 7 (V)

ch 8 (V)
1.4 0.2 -0.025
at the beginning of almost an -0.2

hour of data. 1.3 0 -0.03

0.095 -0.195 0.465

0.31

ch 10 (V)

ch 11 (V)

ch 12 (V)
ch 9 (V)

0.09 -0.2 0.46

0.085 -0.205 0.455

Figure 6.5
27
 Data quality assurance

SETUP9_4.MAT
Ambient vibration data 0 0.2 0.1 -0.3

ch 1 (V)

ch 2 (V)

ch 3 (V)

ch 4 (V)
-0.2 0 0 -0.35
Zooming in the time
axis (Figure 6.6) -0.4 -0.2 -0.1 -0.4

shows quantisation 0 1.5 0.4 -0.02

noise in channels 9- ch 5 (V)

1.45

ch 6 (V)

ch 7 (V)

ch 8 (V)
0.2 -0.025
12. The signals to -0.2 1.4

these channels are 1.35 0 -0.03

badly connected. 0.095 -0.196 0.465 0.31

ch 10 (V)

ch 11 (V)

ch 12 (V)
ch 9 (V)

0.09 -0.198 0.46 0.308

0.085 -0.2 0.455 0.306

Figure 6.6
28
 Data quality assurance

SETUP9_4.MAT

Figure 6.7 shows the

0.2 0.03 0.03 0.03

frequency spectra; Apart 0.02 0.02 0.02

ch 1 (V)

ch 2 (V)

ch 3 (V)

ch 4 (V)
0.1

from channels 8-13 there 0.01 0.01 0.01

are clear recurrent peaks in 0 0 0 0

-3
x 10
the plots; channels 8-13 are 0.03 0.03 0.03 1

noise which appears as a ch 5 (V)

0.02 0.02 0.02

ch 6 (V)

ch 7 (V)

ch 8 (V)
0.5
uniform but bumpy 0.01 0.01 0.01

spectrum with no clear 0 0 0 0

features. Channels 8-13

-3 -4 -3 -4
x 10 x 10 x 10 x 10
1 6 1 6

were not connected. All 4 4

ch 10 (V)

ch 11 (V)

ch 12 (V)
ch 9 (V)

other channels have strong 0.5

2
0.5
2
coherence with each other 0 0 0 0
around the peaks ion the 4

spectra. Figure 6.7

29
 Data quality assurance

FAB1A_1.MAT
0.04
ICP/LIVM type 0.03
accelerometers suffer 0.02
from significant drift. 0.01
Figure 6.8 shows an 0

example of ground

ch 2 (V)
-0.01

vibrations at a quiet -0.02

site at night: -0.03

-0.04

-0.05

-0.06
0 200 400 600 800 1000
seconds
Figure 6.8
30
 Data quality assurance

For stronger signals this is

not a problem, but it is one
reason why ICP type
accelerometers should not
be used for very weak
response measurements.

Figure 6.9
31
 Data quality assurance

p1_8
0.5 0.01
Forced vibration testing

ch 4 (m/sec2)
ch 1 (kN)
Figure 6.10 shows a shaker 0 0
signal (chirp) used on a short
span highway bridge. It looks -0.5 -0.01

OK but the spectrogram 0.01 0.01

ch 14 (m/sec2)
ch 8 (m/sec2)
shows first of all that the
0 0
shaker response is not linear
(generating weak harmonics) -0.01 -0.01
and that structure response 0.01 0.05

ch 15 (m/sec2)

ch 16 (m/sec2)
occurs in all modes twice
during the sweep, due to 0 0

passing vehicles. (Figure 6.11)

-0.01 -0.05
290 295 300 305 290 295 300 305
seconds seconds

Figure 6.10
32
 Data quality assurance
p1_8

40 40
Hence the shaker should

ch 1

ch 4
not be overdriven, also 20 20
the frame should be
0 0
rejected since the
response is not caused
by the measured input 40 40

ch 14
ch 8
and the frequency 20 20
response function will be
corrupted. 0 0

40 40
ch 15

ch 16
20 20

0 0
290 295 300 290 295 300
seconds seconds
33
 Data quality assurance

LD000A 22/11/1996 58:13

0.3 0.01 0.04
Figure 6.12 shows hammer 0.2 0.02
testing of a bridge (using a 0.1
0
0

t1

t2

t3
brown, soft, tip) in a case 0
-0.01
-0.02
where ambient response is
-0.1 -0.02 -0.04
high and the signal to noise 1
-3
2 3
F/Hz, t/sec
4 5 1
-4
2 3 4 5 1
-3
2 3 4 5
x 10 x 10 x 10
ratio is poor. In such cases 8
4
a strong exponential 3 6
2

window should be used to f1

f2

f3
2 4 1
reduce the noise 1 2

contribution. 0
5 10 15 20
0
5 10 15 20
0
5 10 15 20
F/Hz, t/sec F/Hz, t/sec F/Hz, t/sec

Figure 6.12
34
 Data quality assurance

LD000A
0.5

ch 1 (V)
0
The effect of an exponential
window of the form -0.5

W= (
( t ) exp − 4t T ) 0.02

ch 2 (V)
0

where T is the frame size (just -0.02

over 20 seconds) is seen 0.05

ch 3 (V)
0

-0.05
0 5 10 15 20
Figure 6.13 t/sec
35
 Data quality assurance

In frequency domain this window LD0004 ch 1

5
smoothes the response peaks and
removes noise contribution.
Other noise problems from hammer
0
testing include: LD0004 ch 2
1
• Double hits
0.5
• Saturation/overload of accelerometers
located near the driving point due to 0
LD0004 ch 3
large local deformations 4

• Failure to capture the hammer 2

transient as it rises and without
enough resolution, resulting in biased 0
5 10 15 20
Frequency response functions F/Hz
36