Experimental Modal Analysis

15

Abstract

This chapter deals with the sequence of the individual steps of the experimental modal
analysis. The focus is on the practical operational execution and implementation with
the aid of MATLAB®. The theoretical principles required for this are presented in
summary form with references and without further derivation. This chapter does not
claim to contribute to the discussion of the theoretical foundations of modal analysis.
For this purpose, reference is made in particular to the literature (Ewins DJ (2003)
Modal Testing: Theory, Practice and Application, Aufl. 2. Research Studies Press,
Baldock; Døssing O (1988) Structural Testing Part I: Mechanical Mobility
Measurements. Brüel & Kjær, Nærum (Revision April 1988); Døssing O (1988)
Structural Testing Part II: Modal Analysis and Simulation. Brüel & Kjær, Nærum
(March 1988); Kokavecz J (2010) Messtechnik der Akustik, Kapitel 8, Modalanalyse.
Springer, Berlin)

" This chapter deals with the sequence of the individual steps of the experi-
mental modal analysis. The focus is on the practical operational execution
and implementation with the aid of MATLAB®. The theoretical principles
required for this are presented in summary form with references and
without further derivation. This chapter does not claim to contribute to
the discussion of the theoretical foundations of modal analysis. For this
purpose, reference is made in particular to the literature [1–4].
There are differences in the notation of equations and formula designators
between the notation used so far in this book and that used in the field of
modal analysis. Where notations differ, they are listed in Table 15.1.

# The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, 513
part of Springer Nature 2023
T. Kuttner, A. Rohnen, Practice of Vibration Measurement,
https://doi.org/10.1007/978-3-658-38463-0_15
514 15 Experimental Modal Analysis

Table 15.1 Comparison of different notations

Notation
according to DIN Alternative Notation in exp.
Designation 1311 spelling Source Modal analysis
Mass m m
Spring constant k k
Damping constant d c
Deflection x(t) ξ Literature x(t)
[4]
Receptance, dynamic HxF ndyn Literature HxF
compliance [4]
Admittance, agility, HvF Y Literature HvF
mobility, mobility [4]
Acceleration, Inertance, HaF acc Literature HaF
acceleration, inertia [4]
Force F(t) f(t) Literature F(t)
[3]

In order to analyze vibration problems, it is necessary to know the natural

frequencies, the associated vibration mode and the respective modal damping
of the structure under consideration. Here, one depends on computer-aided
measurements and analyses. The most important method in this field is experi-
mental modal analysis. This is not a self-contained computational process
realized by a single function in the analysis software, but a process consisting
of methods of measurement and functions for analysis. The experimental
modal analysis is the process with which this is determined for the considered
frequency range.
Scanning synchronous measurements of the structure excitation and the
structure response(s) are required. Transfer functions are calculated from the
measured values, which are used to identify system parameters of the object
structure. Classical exciters for object excitation are the impulse hammer and
the shaker.
With the 2017a release, Mathworks introduced features of the MATLAB®
Signal Processing Toolbox for modal analysis, which are applied to experimen-
tal modal analysis in this chapter.
Additional material for this chapter can be found at http://
schwingungsanalyse.com/Schwingungsanalyse/Experimentelle_Modalanalyse.
html.
15.1 Assumptions and Explanations of Terms 515

15.1 Assumptions and Explanations of Terms

A mechanical structure usually has several natural frequencies with associated vibration
patterns. A mode (mode of appearance, mathematically eigenfunction ψ) is understood as a
natural frequency with its oscillation pattern, the oscillation mode. From this it is derived
that a structure has a first mode, second mode and so on. This form of representation lists
the modes of the structure in the sequence of the natural frequency. It is common to have a
description that verbally describes the vibration pattern of the mode. The subscript r is used
for the mode number.

Example
• (first) torsional mode, for a mode whose vibration pattern indicates torsion along
the principal axis of the structure with the lowest (1st) torsional natural frequency.
• (second) bending mode, for a mode whose vibration pattern indicates bending of
the structure in the second bending natural frequency.

One of the most important assumptions for the experimental modal analysis is the
orthogonality of the eigenfunctions and thus the independence of the modes from each
other. For each mode, a separate closed system is assumed in the form of the single-mass
oscillation system described in Chap. 3 consisting of the mass m, the spring stiffness k and
the damping constant c, with one degree of freedom (Fig. 15.1). This means for the
boundaries of the respective system that they have either ideally free or absolutely rigid
edges. In practice, this assumption is fulfilled if the natural frequencies can be determined
via the determined transfer functions.
The single mass oscillating system is also called single mass oscillator or SDOF (Single
Degree Of Freedom) system.
Another important assumption is that they are linear time invariant systems.1
If this is not the case, the analysis shall be limited to such an extent that this assumption
is fulfilled. For example, by reducing the frequency range considered or shortening the
measurement time to a time interval in which the condition of time invariance can be
fulfilled.
The principle of superposition is assumed. The superposition of individual system
responses is to be understood as the sum of the system responses without the individual
system responses influencing each other. This also means that the response of the system to
simultaneous excitation with several signals is identical to the sum of the system responses
to the individual excitation signals.

1
Linear means that a doubling of the excitation leads to a doubling of the response. Time invariant
means that the system properties do not change over time.
516 15 Experimental Modal Analysis

m1 m2 m3 mn

k1 c1 k2 c1 k3 c3 ... kn cn

Fig. 15.1 Model for the assumption of a structure with n uncoupled modes

Proportionality is assumed. Proportionality exists, for example, when the excitation

force is changed by a certain factor y and the system response also changes by the same
factor y.
There is reciprocity. This means that the location of the excitation can be swapped with
the location of the response. This is a very important assumption to verify when performing
experimental modal analysis, since the exchange of locations between excitation and
response is performed very often. Locations of excitation are given index n, locations of
response are given index m.
There is causality. Accordingly, there is no system response without a system excitation.
Therefore, the required measurements must take into account that the system response and
the system excitation are recorded as a whole.
The system is stable. If the system excitation is terminated, the oscillations of the system
decay. The decay behaviour is determined by the damping of the system.

15.2 Summary of the Analytical Principles of Modal Analysis

The basis of the theory for the experimental modal analysis is the equation of motion of the
forced oscillation of the single-mass oscillator with velocity-proportional damping, a
differential equation of second order (compare Eq. 15.1). Considering the different
notations and multiplying by (-1), Eq. 4.1 becomes.

m€x þ c_x þ kx = F ðt Þ ð15:1Þ

For the description of the oscillation form, deflection values are required at several points of
the examined structure. In addition, it can be assumed that the real system is not a system
with one degree of freedom, but a system with several degrees of freedom, which is called
MDOF (Multi Degree Of Freedom) system, for which the assumption made in Fig. 15.1
applies. From the measurements carried out, the transfer function matrix discussed in [1, 4–
7] and the notation of the equation of motion results in.

M€x þ C x_ þ Kx = F ð15:2Þ

Herein is
15.2 Summary of the Analytical Principles of Modal Analysis 517

• M is the matrix of the modal mass

• C the damping matrix
• K is the stiffness matrix

A mathematically more manageable model provides the consideration in the frequency

domain.
By using the transfer function(s), (compare Sect. 4.3 or Table 4.2)

• dynamic stiffness, H xF ðωÞ, when measuring the applied force and the deflection as
system response, which is operationally less common
• Mobility,H vF ðωÞ, when measuring the force applied and the velocity as the system
response.
• Inertia, H aF ðωÞ, when measuring the applied force and acceleration as the system
response.

a modal parameter model for the single-mass transducer can be derived. The transfer
function H xF ðωÞ is required for this. This can be determined either directly from the Fourier
transforms of the measurement or by derivation from the transfer functions H vF ðωÞ and.
H aF ðωÞ.

X ðωÞ H vF ðωÞ H aF ðωÞ

H xF ðωÞ = = = ð15:3Þ
F ð ωÞ jω ðjωÞ2

The transfer function can be determined by a practical measurement (compare Sect. 14.5).
In [3], the transfer function H xF ðωÞ is obtained by substituting the equation of motion in.

1 1 1
H xF ðωÞ = þ - ð15:4Þ
k jωc ω2 m

is transferred. This describes mathematically the characteristic course of the transfer

function of the single-mass oscillator with velocity-proportional damping (Fig. 15.2).
In the frequency range up to the resonant circuit frequency of the damped oscillation ωd
this is characterized by the spring and is described by the stiffness. The higher the stiffness
of the system, the lower the magnitude of the transfer function.
In the frequency range above the resonant frequency of the damped oscillation ωd, the
angular frequency and the mass determine the course of the transfer function. The higher
the angular frequency, the lower the magnitude of the transfer function.
In the frequency range of the resonant circuit frequency of the damped oscillation ωd,
also referred to as modal frequency (see [4]), the damping of the system is determinant for
the magnitude of the transfer function. Systems with pronounced and clearly identifiable
resonant peaks have low damping. Here, the resonant circuit frequency of the damped
oscillation ωd can be easily determined, but the determination of the damping is more
518 15 Experimental Modal Analysis

1/j c
|HxF| in m/N

1/k

- 1/ 2 m

in 1/s
d

Fig. 15.2 Characteristic curve of the transfer function of the single-mass oscillator with velocity-
proportional damping

difficult due to the narrow frequency range. The opposite is true for systems with high
damping. These show a flatter and broader course of the resonance peak.
For the determination of the modal parameters

• Modal frequency ωd(r)

→
• Vibration mode described by the deflection vector ϕ ðr Þ
• modal damping δ(r)

the eigenvalues from Eq. 15.1 are required.

Three cases are distinguished:

• The limiting case of critical damping ck

p
This is present at c = ck = 2 mk where the two eigenvalues are equal and become
s = - c=2m.

p
• Aperiodic damping c ≥ 2 mk
Here the deflection decays non-oscillatory exponentially and the eigenvalues
become real.

p
• Non-critical damping c < 2 mk
Here the deflection decays exponentially and the solution has conjugate complex
eigenvalues.
15.2 Summary of the Analytical Principles of Modal Analysis 519

c k c 2
s= - þj - = - δ þ jωd ð15:5Þ
2m m 2m

c k c 2
s = - -j - = - δ - jωd ð15:6Þ
2m m 2m

From.

s = Refsg þ Imfsg ð15:7Þ

it is derived that.

c
δ = - Refsg = ð15:8Þ
2m

and

k c 2
ωd = Imfsg = - = ω20 - δ2 ð15:9Þ
m 2m

Where ω0 is the natural angular frequency of the undamped system.

The eigenvalues of the system under consideration are referred to as pole positions2 and
consist of the modal damping δ and the modal frequency ωd. The eigenvalues and thus the
pole positions can be found in the transfer function H xF ðωÞ at ωd(r).
If one performs the solution of the eigenvalues as a matrix calculation (see [4, 6]), then
each individual element of the transfer function matrix can be given by.

N
ϕm ð r Þ  ϕn ð r Þ
H xF,mn = ð15:10Þ
r=1
- ω2 þ jω2δðr Þ þ ω20 ðr Þ

can be determined.
For the consideration of the individual mode r this means in mathematical notation.

ϕm ð r Þ  ϕn ð r Þ N ϕm ðqÞ  ϕn ðqÞ
H xF,mn = þ q=1 ð15:11Þ
- ω2 þ jω2δðr Þ þ ω20 ðr Þ q≠r
- ω2 þ jω2δðqÞ þ ω20 ðr Þ
Bmn

2
The zeros of a characteristic polynomial are called poles.
520 15 Experimental Modal Analysis

Due to the assumption of orthogonality, the transfer function in the interval around the r-th
resonance is determined by the share of the r-th mode alone. The influence of the
neighboring modes, which is described in Eq. 15.11 with Bmn, is neglected in the parameter
determination by Bmn = 0.
The determination of the modal parameters is trustworthy if the individual modes are
clearly identifiable as resonance points in the transfer function, which is referred to as weak
modal coupling. If this is not the case, strong modal coupling is present and ωd(r) and δ(r)
cannot be determined unambiguously.
For Eq. 15.11 as transfer function H aF,mn holds with Bmn = 0 and ω = ωd(r).

ω2d ðr Þ  ϕm ðr Þ  ϕn ðr Þ
H aF,mn = ð15:12Þ
- ω2d ðr Þ þ jω2δðr Þ þ ω20 ðr Þ

With the damping ratio D.

δðr Þ
D ðr Þ = ð15:13Þ
ω d ðr Þ

and ω20 = ω2d ðr Þ þ δ2 ðr Þ from Eq. 15.9, the amplitude jHaF(d(r))j becomes.

ϕm ð r Þ  ϕn ð r Þ
j H aF ðωd ðr ÞÞ j = ð15:14Þ
D4 ðr Þ þ 4  D2 ðr Þ

To determine the deflections ϕm(r), the deflection at the response point, and ϕn(r), the
deflection at the excitation point, a measurement is required in which the response point m
is equal to the excitation point n. This is called the driving point measurement. This
measurement is called the driving-point measurement. In this case, Eq. 15.15 becomes.

ϕn ð r Þ 2
j H aF ðωd ðr ÞÞ j = ð15:15Þ
D4 ðr Þ þ 4  D2 ðr Þ

and it can be found at ϕn(r) via.

ϕn ð r Þ = j H aF,mn ðωd ðr ÞÞ j  D4 ðr Þ þ 4  D2 ðr Þ mit m = n ð15:16Þ

can be determined. The following then applies for ϕm(r).

15.3 Operational Performance of the Experimental Modal Analysis 521

j H aF,mn ðωd ðr ÞÞ j
ϕm ðr Þ = D4 ðr Þ þ 4D2 ðr Þ  mit m ≠ n ð15:17Þ
ϕn ð r Þ

" Mnemonic
To determine the mode shapes of modes r of a structure with the capabilities
presented here using MATLAB® are required:

• the amplitudes from the transfer functions j H aF,mn ðωd ðr ÞÞ j at ω = ωd(r) for the
excitation response point m = n, in order to determine the deflection of the
excitation point ϕn from them
• the amplitudes from the transfer functions j H aF,mn ðωd ðr ÞÞ j at ω = ωd(r) for all
response measurement points m ≠ n, in order to determine the deflections of the
response points ϕm from them
• the range around od(r) of the transfer functions j H xF,mn ðωd ðr ÞÞ j, in order to
determine from this the damping D(r) and the exact resonant circuit frequency
ωd(r) or alternatively the resonant frequency. fd(r)
• the sign of the imaginary part of the pole s, in order to determine the direction of
the deflection.
This solution path requires that the underlying measurements are measurements with
one excitation point n and multiple response points m. If this is not the case, the index n can
be swapped with m in the equations due to reciprocity.

15.3 Operational Performance of the Experimental Modal Analysis

The measurement itself is carried out as a pulse hammer measurement (Fig. 15.3) with one
or more excitation points with one or more response measuring points, which are ideally
equipped with accelerometers. Alternatively, a shaker (Fig. 15.4) can be used, but then
usually at one excitation point.
The object of analysis must be discretized in a meaningful way for the modal analysis.
Individual measurement or excitation points must be defined on the structure to be
investigated – a process which initially appears to be simple. However, it must be
remembered that if the measurement points are evenly distributed, individual modes cannot
be detected. If, for example, excitation or measurement is performed in the vibration node
of a mode, this mode cannot be determined. Shannon’s sampling theorem also applies to
the determination of the vibration mode, according to which more than two measured
values are required for the determination of a vibration. Applied to experimental modal
analysis, this means that only those modes can be determined whose wavelength is longer
than twice the minimum distance between two measurement points.
522 15 Experimental Modal Analysis

Fig. 15.3 Measurement setup of the experimental modal analysis on a flywheel. The response signal
is measured at a response point m with an accelerometer on the inside of the flywheel. The excitation
is performed at a total of 10 points along the flywheel with an impulse hammer. The picture shows the
excitation-response point measurement

Fig. 15.4 Measurement setup

of the experimental modal
analysis on a beam. The
response signal is measured via
the two accelerometers on the
beam, while the excitation at one
point is performed by an
electrodynamic shaker, which is
coupled to the beam via a force
sensor

In the case of complex structures, care must be taken to ensure that all vibrations of each
characteristic component are recorded.
If the result of the experimental modal analysis is also used for the adjustment with a
simulation model, it must be ensured that the measuring points of the measurements match
the simulation points. Only then is an adjustment possible.
The number of measurement points depends on the geometry of the structure, the
frequency range considered and the number of modes. Accordingly, this cannot be
determined unambiguously at the beginning of an analysis. However, this task can be
adequately mastered through experience with similar objects and comparison with a
simulation model.
15.3 Operational Performance of the Experimental Modal Analysis 523

Each individual measuring point has six degrees of freedom, of which usually only one
to three translational degrees of freedom are measured.
If the natural frequencies and natural mode shapes are already known (e.g. from a
calculation or on the basis of similar components), a few measurement points are sufficient
for metrological verification. If, however, a complex model is to be analyzed from the
experimental modal analysis, a very large number of measurement points, up to several
hundred, are required. The operational effort in the measurement execution is
correspondingly high.

15.3.1 Storage

For the quality of the results of an experimental modal analysis, the boundary conditions
under which the required measurements are carried out are decisive. It makes sense to
investigate the structures to be analyzed in boundary conditions that are as realistic as
possible. Whenever the excitation and measurement points are difficult or impossible to
reach in the installed state, idealized storage and boundary conditions are used. However,
this is very often the case in practice.
If the measurement serves to adjust an FEM model, the framework conditions of the
FEM model must also be observed for the measurement. Often, the measurement and
simulation results do not match because the boundary and transition conditions of the
calculation and measurement models are different.
The influence of a high structural damping and nonlinearities will strongly influence the
result in the installed object state, since these are in contrast to the prerequisites of a modal
analysis. In the practical implementation, an attempt is made to get as close as possible to
the real boundary condition.
An alternative approach is to work in the extreme opposite direction by trying to achieve
the ideal free boundary conditions. For this, it must be achieved that the natural frequency
of the suspension is as far away as possible from the frequency ranges to be investigated,
e.g. natural frequency of the suspension at most 1/5 of the first mode of the structure to be
investigated.

15.3.2 Object Excitation by Means of an Impulse Hammer

The ideal impulse with a unique event at t = 0 is only a thought model. The Fourier
transformation of this thought model is the infinite broadband frequency excitation. With a
pulse hammer only a real pulse excitation is possible. The real time pulse causes a
broadband excitation in a limited frequency range (compare Sect. 14.5).
The simulated test with impulse hammer blows of different widths shown in Figs. 15.5
and 15.6 demonstrates that very narrow impulses must be taken into account when carrying
524 15 Experimental Modal Analysis

Evaluation of pulse widths

Pulse width
1
0.2 ms
2 ms
5 ms
0.8
Pulse height

0.6

0.4

0.2

0
0.01 0.015 0.02 0.025 0.03 0.035
Time

Fig. 15.5 Time series of simulated impulse hammer blows to evaluate the suitability of impulse
hammer measurements for experimental modal analysis

out impulse hammer measurements. Pulse widths of significantly less than 1 ms are
required for usable measurements.
The measured time series of the force signal of an impulse hammer blow shown in
Figure 15.7 has a total of 10 usable measured values for the impulse at a sampling rate
fs = 51,200 Hz. Measurements of impulse hammer blows therefore require sufficiently
high sampling rates. The 51,200 Hz used in this example as the sampling rate of the
transducer must be regarded as borderline low.
Figure 15.8 shows the usable frequency bandwidth of the measured impulse hammer
blow. The frequency range up to 2 kHz (max. 2.5 kHz) is read off as usable. Up to this
cut-off frequency the frequency response is sufficiently linear and the amplitude or
normalized amplitude is not reduced by more than 20%, but it is not optimal. In the very
low frequency range (0 to 10 Hz) this measurement is unusable. The reason for this is also
to be found in the measuring technique used, since in this case it only has no influence on
the measurement result above 10 Hz.
The required impulse height, i.e. the strength of the impulse hammer blow, depends on
the energy required to excite the structure to be examined. For this purpose, the trade offers
various heavy impulse hammers. In addition, different impulse hammers can be equipped
with different hammer tips. Depending on the material properties of the hammer tip, the
same impulse hammer can be used for different frequency ranges.
In the frequency range up to an amplitude drop of 3 . . . 5 dB, the force excitation of the
impulse hammer can be used. For the example in Fig. 15.9 this means a usable frequency
range up to

• 4200 Hz when using a metal tip

• 1400 Hz when using a plastic tip
• 380 Hz when using a rubber tip
15.3 Operational Performance of the Experimental Modal Analysis 525

Spectral evaluation of pulse widths

Pulse width
1
0.2 ms
2 ms
Pulse height (normalized)

5 ms
0.8

0.6

0.4

0.2

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency [Hz]

Fig. 15.6 Frequency spectrum of simulated impulse hammer blows of different widths

Measured impulse hammer blow

200
Pulse width 0.135 ms

150
Pulse height (N)

100

50

3.023 3.024 3.025 3.026 3.027 3.028 3.029 3.03

Pulse width [s]

Fig. 15.7 Measured pulse hammer impact with 0.135 ms pulse width

With a different impulse hammer, with a different structure to be examined and with other
contact partners, these cut-off frequencies (and excitation amplitudes) will be different. It is
recommended to determine this individually for the respective measuring task.
The quality of the measured impulse is additionally strongly dependent on how the
structure is struck with the impulse hammer. An evaluation of the impact quality is
therefore mandatory and some test impacts before the measurement are recommended.
The evaluation is made in pulse height and pulse width. In addition, it must be checked
whether a blow with several excitation pulses has not been made by mistake (compare
Fig. 15.10).
For the excitation response point measurement, a measuring point is required for which
the condition m = n applies. This is achieved by striking the response sensor, e.g. an
526 15 Experimental Modal Analysis

Frequency spectrum of the measured impulse hammer blow

0.8
normal pulse height

0.6

0.4

0.2

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency [Hz] Page 17

Fig. 15.8 Amplitude spectrum of the measured impulse hammer blow

Impulse hammer blows with different tips

1
Metal tip
0 Faux fabric tip
Rubber tip
-1
Normalized amplitude in dB

-2

-3

-4

-5

-6

-7

-8

-9

-10
0 1000 2000 3000 4000 5000 6000 7000
Frequency in Hz

Fig. 15.9 Amplitude spectrum of different impulse hammer tips with the same attachment point

acceleration sensor, with an impulse hammer in the immediate vicinity. Of course, care
must be taken to ensure that the response sensor itself is not struck by the impulse hammer
blow. In the flywheel example, this was achieved by placing the accelerometer on the
inside. The impulse excitation was achieved by striking the flywheel from the outside (see
Fig. 15.3).

15.3.3 Object Excitation by Means of a Shaker

Shakers (electrodynamic or hydraulic) allow the excitation of structures with arbitrary

signals. Thanks to a wide power range, even large, thus very complex, strongly damped
15.3 Operational Performance of the Experimental Modal Analysis 527

Impulse hammer blow

100

90

80

70

60
Force in N

50

40

30

20

10

-10
0 0.5 1 1.5 2 2.5
Measured values 4
10

Fig. 15.10 Impulse hammer blow with several excitation pulses

and very heavy structures can be excited. It should be noted that shakers with high power
values have a lower highest excitation frequency. However, this is often not problematic
because heavy structures, which require high excitation powers, have low modal
frequencies. This circumstance only has to be taken into account when designing the
excitation signal.
The excitation force is measured between the shaker and the structure under investiga-
tion, at best directly at the initiation point. This force signal is used for the calculation of the
transfer functions. Alternatively, an impedance measuring head can also be used at this
measuring point, which supplies the acceleration signal in addition to the force signal.
Otherwise, an accelerometer must also be placed at the force application point.
Accelerations are usually measured at all specified measuring points.
Noise, sine signals, sine sweep (chirp) or other arbitrary signal forms can be used as
excitation signals. It is also possible to use measured signals from operation (referred to as
road load).
Care must be taken to ensure that the shaker is securely positioned and carefully aligned.
No static forces may be introduced into the structure to be examined and the introduction
mechanism must not have any play under any circumstances. Both would lead to consid-
erable deviations in the calculated transfer functions. Reactions from the test object to the
coil of the shaker must be prevented as far as possible. Both are achieved by connecting the
shaker to the excitation point via a thin rod (called a stinger), often designed like an
expansion shaft screw.
528 15 Experimental Modal Analysis

Fig. 15.11 Measuring object

flywheel

15.4 Evaluation of the Experimental Modal Analysis in MATLB®

15.4.1 Evaluation of Measurements with Impulse Hammer Excitation

The procedure for evaluating the experimental modal analysis with impulse hammer
excitation is to be illustrated using the example of a flywheel of an internal combustion
engine (Fig. 15.11).
The flywheel was excited from the outside with the impulse hammer at 10 positions,
i.e. at a distance of 36 degrees each. An accelerometer was attached to the inside at position
3. This results in the excitation response point measurement, i.e. to determine the excitation
deflection ϕn, the measurements from position 3 with which this example begins.
Measurements were taken without any further programmed interactive assistance. The
measuring technique used was simply created and the measurements were carried out
individually by startForeground. Up to 16 individual impulse hammer blows per position
are available for further evaluation.
Figure 15.12 shows the measured time series of the impulse hammer and the accelera-
tion response, but the diagram also shows that the individual time series have a time offset
from each other due to the measurement procedure.
The MATLAB® function modalfrf used to determine the transfer functions allows
several measurements belonging to each other to be evaluated. For this purpose, the
individual measurements are to be brought to the same measurement length. From the
observation of the impulse responses, it can be seen that a decay time of 3 s is completely
sufficient. In addition, the impulse hammer blow must always occur at the same time,
which is defined as 2 ms. These and other data required for modalfrf are defined by
parameterization.
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 529

Measured impulse hammer blows

400
350
Pulse height (N)

300
250
200
150
100
50
0

Measured impulse responses

3000
Impulse response (m/s2 )

2000

1000

-1000

-2000

-3000
0 1 2 3 4 5 6 7 8 9 10
Time [s]

Fig. 15.12 Representation of the time series of the impulse hammer blows measured without
triggering as well as the impulse responses by excitation at position 3

ð15:18Þ

The “trimming” of the time series is realized via a loop. This determines the data position of
a threshold value overrun of the pulse hammer time series. The matrix index position stands
for the measurement position, while inc stands for the number of the executed blow.
Finally, the script must evaluate all blow positions with the totality of the blows.

ð15:19Þ

The threshold value for the impulse hammer blow is defined here as 50 N. The instruction
*1000/kaliHammer converts the voltage measured in volts into the required physical
quantity Newton. The calibration factors are usually available in mV per physical quantity.
The determined data position is now corrected forward by the number of measured
values of the pre-trigger, while the end is located blocksize-1 further back in the time series.
530 15 Experimental Modal Analysis

ð15:20Þ

The respective raw data of response and excitation can now be formed from the measure-
ment time series. This is done with simultaneous conversion into physical values.

ð15:21Þ

For the selection of measurements suitable for modal analysis, it is suitable in the first step
to consider the pulse widths in milliseconds and pulse heights in newtons of the pulse
hammer time series. To determine the pulse width, MATLAB® provides the function
pulsewidth(data, fs). The pulse height can be determined using max(data).

ð15:22Þ

The pulse heights and pulse widths determined in this way are first evaluated for plausibil-
ity and usability using an X-Y plot.
Not all of the impulse hammer blows shown in Fig. 15.13 are suitable for further
evaluation. A suitable pulse width seems to be in the range of 0.12 to 0.13 ms, a usable
pulse height between 200 and 300 N. These values are not generally valid and differ from
test object to test object. The subjective decision is confirmed by calculating the coherence
(compare Sect. 14.5, Eq. 14.117) of the signals to each other. This is one of the result
vectors of the MATLAB® function modalfrf. By means of.

ð15:23Þ

the transfer functions as well as the coherence are determined. Before this, however, the
excitation and response signals must be combined (Fig. 15.15). This is done by simply
joining the signals selected by pulse height and pulse width. Modalfrf is based on the
Fourier transform (compare Sect. 14.4). To reduce leakage, the signals must be weighted
with a window function (compare Sect. 14.4.5). However, the problem shown in Fig. 14.44
must be considered. A window function is required which forces the signal to decay to zero
at the end of the signal. No weighting should be carried out at the start of the signal, as the
system is at rest here and the measured values are therefore zero.
The RectExpo function is provided for this purpose in the additional materials. This
window function consists of a rectangular (rect) portion at the beginning of the window and
a decaying portion towards the end of the window.

ð15:24Þ

As with all window functions, the length (window length) of the window function is
required, this must be identical to the specification N or nfft in the Fourier transform and
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 531

Pulse heights / pulse widths

350

300

250
Pulse height [N]

200

150

100

50
0.1 0.11 0.12 0.13 0.14 0.15
Pulse width [ms]

Fig. 15.13 X-Y plot of the pulse heights and pulse widths for the evaluated impact position

must be an integer. The parameter fs stands for the sampling rate used for the measurement
performed. Rect specifies how the specification in nrect, the length of the rect component,
is to be interpreted. The following parameters are available

• %: as a percentage
• time: as time specification in ms
• n: as the number of measured values

Depending on the parameter used, the number of measured values of the rect component is
calculated differently. The parameter weighting specifies the weighting of the decay
function. The last parameter type specifies the form of the decay function. This contains

• 1 for linear decay. The parameter weighting has no function here, it only has to contain a
value.
• 2 for a decay curve e-t  weighting. This window function does not necessarily decay to
0 in the given window length!
• 3 for a decay curve sin(x)2. weighting specifies the data range in % over which the decay
function is to extend. If the addition of the rect-part and the length of the decay function
is smaller than the window length, the defined window length is extended with 0-values.
This window function also does not necessarily decay to 0 within the window length.
532 15 Experimental Modal Analysis

Excitation signal
400
Force [N]

200

-200

Response signal
Acceleration [m/s2]

2000

-2000

-4000
0 2 4 6 8 10 12
Measured values 105

Fig. 15.14 History of the window function of antWindow

Figure 15.14 shows the course of the weighting using the example of the window function
for the response signal.
The window function for suggestion and response do not have to be identical.

ð15:25Þ

provides different window functions for the excitation signal and the response signals
respectively. For the window function of the excitation pulse hammer signal, a fast decay to
zero is recommended. This causes the signal component to be forced to zero after the
excitation pulse. This means that no interfering signal components, e.g. due to the handling
of the impulse hammer after the impact, are included in the calculation of the transfer
function.
The excitation and response signals shown in Fig. 15.15 were used to calculate the
transfer function. The MATLAB® function modalfrf uses long individual segments from
the blocksize signals to average the result. The evaluation of the calculated transfer function
j H xF ðf Þ j (Fig. 15.16, blue line) can be done using the coherence (Fig. 15.16, red line). If
this lies in the range 0.8 to 1 for the frequency ranges in question, the result is trustworthy.
Dips in the coherence in the frequency ranges with transfer function values close to zero
(is called antiresonance) are common.
For the use of the MATLAB® function modalfrf it is to be noted that the excitation
signal must be a time series of the force. No other physical quantity is provided here. Time
series as acceleration, velocity or displacement are possible as response signal. The
measured object acceleration is assumed in the standard setting. If this is not the case,
the interpretation of the response signal can be changed by an additional parameter pair.
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 533

Window function RectExpo

0.8
Weighting

0.6

0.4

0.2

0
0 5 10 15
Measured value 4
10

Fig. 15.15 Composite excitation and response signal for transfer function calculation using modalfrf

FRF / Coherence
-2
10
1

10-4 0.8
Receptance [m / N]

Coherence [---]
0.6
10-6

0.4

-8
10
0.2
Receptance
Coherence
-10 0
10
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency [Hz]

Fig. 15.16 Representation of the transfer function and the coherence in a diagram, facilitating a
qualitative evaluation of the transfer function

ð15:26Þ

By extending the instruction with the parameter pair ‘sensor’, ‘dis’ a response signal is now
assumed as displacement. Possible values for ‘Sensor’ are ‘acc’ for the acceleration sensor
(acceleration, default), ‘dis’ for the displacement and ‘vel’ for the velocity.
There is another special feature to note in the blocksize parameter (referred to as window
in the MATLAB® help). If this is an integer value, it is considered as the length of each
signal segment. If the blocksize parameter is a vector to be considered as a window
function, then both the excitation and response signals are weighted by this vector and
the length of the vector is used for segmentation. However, it does not make sense to
534 15 Experimental Modal Analysis

weight the excitation signal and the response signal with the same weighting vectors for
impulse hammer measurements (see instruction 15.25).
The parameter noverlap has a similar effect in the function modalfrf as in the function
spectrogram (see Table 14.13 overlap). For impulse hammer measurements this must be
0, as they are contiguous single impulse hammer measurements.
Thanks to the weakly damped structure, the frequencies of three modes can be read
directly from the transfer function. The frequencies found are stored in the vector.

ð15:27Þ

for further analysis.

Accurate modal frequencies and the associated damping ratios are required for the
calculation of the vibration mode. The MATLAB® function modalfit is used for this
purpose.

ð15:28Þ

For the calculation, this requires the transfer function frf, the associated frequency vector f,
the sampling rate fs and a specification for the maximum mode number up to which the
calculation is to be performed. Without the other parameters, a “standard” calculation is
performed. This gives an initial overview of the existing modes and is of particular interest
if no clearly identifiable peaks are to be found in the transfer functions. If peaks in the
transfer function can be identified, then modal fit around the specification PhysFreq
followed by a vector of frequencies is performed a default. The frequency range to be
considered can be limited by appropriate parameterization. This should be used, since with
impulse hammer excitations the sampling rates must always be very high and thereby the
analysis frequency range lies above the frequency range of interest.
Different fitting methods are offered to determine the damping:

• lsce (default) stands for least-squares complex exponential method. The lsce method
uses the least squares method. It calculates the impulse response of the FRF and fits it by
summing the complex damped (sine) oscillations. It uses an algorithm described as
Prony analysis or Prony’s method (see [8]), which was developed in 1795 by the French
mathematician and engineer Gaspard Riche de Prony.
• lsrf stands for least-squares rational function and comes from the MATLAB® System
Identification Toolbox. This algorithm requires less data than the other two methods for
determining attenuation. It is the only applicable method when a frequency vector f with
non-uniform frequency spacings is present.
• pp stands for peak picking method. The peak picking method assumes that exactly one
structural mode exists for each significant peak in the transfer function. This represents a
single-mass oscillatory system. The peak picking method provides the most trustworthy
result for transfer functions with significant peaks.
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 535

Table 15.2 Comparison of results of fitting methods for determining modal frequencies and
attenuations
Comparison of modal frequency and attenuation
lsce result lsrf result pp-result
Fn Dn Fn Dn Fn Dn
1221.9 0.0005 1222.0 0.0004750 1222.1 0.0005
3222.0 0.0004 3222.9 0.0004605 3218.7 0.0006
3700.3 0.0010 3222.9 0.0004605 3702.9 0.0018

In the subjective comparison by examining the transfer function in Fig. 15.16, the
attenuations of the first mode (1222 Hz) and the second mode (3220 Hz) should be
approximately the same, while the attenuation of the third mode (3707 Hz) must be
much larger than that of the first two modes. This is confirmed by the lsce and the
pp. results. With the lsrf method the third mode is not detected at all (see Table 15.2).
In general, the determination of the modal parameters is subject to large scatter. The
background to this is the complexity of the method, which relies on Fourier transforms.
In the further progress of the analysis, it is now necessary to determine the deflection
values ϕ, which are required for the description of the oscillating shape. The oscillation
shape itself can only be determined after the analysis of all measured positions.
From the frf determined in instruction 15.24, the magnitude maximum is determined in
the span around the respective pole location (idx).

ð15:29Þ

idx is the index of the pole position which is given by.

ð15:30Þ

is determined. The span is determined with.

ð15:31Þ

Analytically, the maximum of the transfer function HxF ±1 index must be found around the
pole location. This deviation results from the condition of integer values of blocksize in
modalfrf (instruction 15.26).
From.

ð15:32Þ

the deflection direction is determined. To calculate the deflection itself, the evaluation of
the excitation response point measurement is required first. The deflection ϕn results from
Eq. 15.16 as MATLAB® -statement from the transfer function H xF,nm to.
536 15 Experimental Modal Analysis

ð15:33Þ

and the deflection ϕm from Eq. 15.17 to.

ð15:34Þ

The numerical results are stored in a data structure for archiving and further use.

ð15:35Þ

Visualization can be done as a diagram for simple structures, such as Fig. 15.17, or via an
animated 3D visualization.
The determination of the vibration modes of the experimental modal analysis provides a
qualitative result. The determined deflection amounts are a relative deflection to the
excitation deflection. All variable parameters of the vibration measurement and the
performed analysis and calculation steps have an influence on the numerical result.
A further problem in the calculation of the oscillation form is due to the fact that the
values must be taken from the resonance point. However, the determination accuracy is
lowest at the resonance point itself. The numerical result is therefore subject to considerable
scatter.
A total of six steps are required for the evaluation of the experimental modal analysis.

• Examine raw data (example script: Mod01_ImpulsRohdaten.m).

• Trimming of the individual measurements (example script: Mod02_ImpulsBewerten.
m). The pulse beat must always be positioned at the same time in the time series. This is
achieved by triggering the data. From the evaluation of the raw data, an estimate can be
made as to when the vibration of the object has decayed and therefore what the time
length of the evaluation data should be.
• The first selection of the pulse hammer measurements used for the evaluation is then
made via the evaluation of the pulse widths and pulse heights (example script:
Mod03_ImpulsAufbereiten.m).
• The coherence of the impulse hammer blows used so far is determined. Only if this is
sufficiently good, the further steps in the evaluation take place, if not, the first selection
is refined. In the same step the FRF was calculated (example script: Mod04_FRF.m).
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 537

Deflection diagram mode 1

90
1.1
120 60

1.05

150 30
1

0.95

180 0.9 0

210 330

240 300

270

Fig. 15.17 Representation of the vibration mode of the first mode of the flywheel in a polar plot

• If the data quality is found to be sufficiently good, the last calculation step is the
determination of the deflection values (example script: Mod05_BerDat.m).
• In the last step the swinging shapes are shown.

15.4.2 Evaluation of Measurements with Shaker Excitation

The evaluation of the experimental modal analysis with shaker excitation is basically
carried out with the same working steps as for the impulse hammer excitation. As an
example, an experimental modal analysis with shaker excitation was performed on an
exhaust system (Fig. 15.18).
In the example, the exhaust system was excited with a frequency sweep from 15 to
250 Hz over 2 s. The measurement was repeated several times. The measurement was
538 15 Experimental Modal Analysis

Fig. 15.18 CAD model of the exhaust system under consideration. Acceleration sensors were
installed at the positions marked with red dots. The blue elements are absorbers, which were installed
by the manufacturer of the exhaust system to reduce vibrations. These absorbers are also equipped
with acceleration sensors (green dots)

repeated several times. The comparison of the measurements with each other via the
coherence function showed that averaging over several measurements is not possible in
this case. The sequence of the frequency sweep is very difficult to keep equal.
Figure 15.19 shows the transfer functions from the measurement with shaker excitation.
In contrast to the impulse hammer measurement on the flywheel, no clearly pronounced
resonance points can be found. The reason for this is not the measurement method, but the
damping in the structure, which leads to a higher modal coupling. With real test objects,
this quality of the determined transfer functions is to be expected rather than those from the
flywheel measurement.
Natural frequencies can be assumed at the frequencies 30 Hz and 78 Hz.
The stability diagram shown in Fig. 15.20 is now helpful, as it shows the.

ð15:36Þ

is generated. The input parameter for this function is the FRF matrix, in which all transfer
functions from the excitation point to the respective response points are located. In the
example, these are six response points. From this, a mean value is formed, which is shown
as a red line in Fig. 15.20.
In addition, the “stable” frequencies determined by mathematical approximations are
represented by an o-sign, the “stable” frequencies and attenuations by a + sign and the
instabilities by a . - sign.
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 539

FRF / Coherence at measuring point B11

Receptance [m / N]

10-5

0 50 100 150 200 250

Frequency [Hz]
FRF / Coherence at measuring point B10
Receptance [m / N]

-5
10

0 50 100 150 200 250

Frequency [Hz]
FRF / Coherence at measuring point B08
Receptance [m / N]

-5
10

0 50 100 150 200 250

Frequency [Hz]
FRF / Coherence at measuring point B07
Receptance [m / N]

10-5

0 50 100 150 200 250

Frequency [Hz]
FRF / Coherence at measuring point B03
Receptance [m / N]

-5
10

0 50 100 150 200 250

Frequency [Hz]
FRF / Coherence at measuring point B01
Receptance [m / N]

-5
10

0 50 100 150 200 250

Frequency [Hz]

Fig. 15.19 Transfer functions of the measuring points on the exhaust system relevant for the
experimental modal analysis

The interpretation of the diagram is carried out independently of the type of measure-
ment via the averaged transfer functions. Poles are only present if a local peak can be seen
in the transfer function. In mathematical approximations of higher order, however, very
many poles are found in a transfer function, even if these produce hardly visible peaks in
the line course of the transfer function. A further criterion for the presence of a pole is that
the frequency and the attenuation must be the same across all model orders (Model Order in
Fig. 15.20) – the mathematical approximations.
540 15 Experimental Modal Analysis

Stabilization Diagram
10 2
Stable in frequency
Stable in frequency and damping
45 Not stable in frequency
Averaged response function
0
10
40

35 10 -2

30
Model Order

10 -4

Magnitude
25

-6
20 10

15
10 -8

10

10 -10
5

0 10 -12
0 50 100 105 200

Frequency [Hz]

Fig. 15.20 The stability diagram with the transfer function averaged over all transmission paths and
markers for the stability criteria “stable in frequency”, “stable in frequency and attenuation” and
“unstable in frequency”

The lines from + and o must therefore be perpendicular and cross with the peak of the
transfer function. Only then is there a pronounced natural frequency in the structure under
investigation.
In the example, the modal frequencies 30.34 and 77.8 Hz with an average D of 2.4% and
1.9%, respectively, are taken from the instruction.

ð15:37Þ

determined.
The calculation of the deflections is done via two interleaved for-loops from the FRF
matrix the required values are determined as described in the impulse hammer evaluation.
15.4 Evaluation of the Experimental Modal Analysis in MATLB® 541

ð15:38Þ

The exhaust system is mounted at a total of three positions. The driving point is located in
front of the first bearing when viewed in the X-direction, which is located at approx.
X = 200 mm. Figure 15.21 shows the vibration mode of the second mode at 77.8 Hz.
542 15 Experimental Modal Analysis

Oscillating shape of the exhaust duct in Z-direction

10-7
5
Mode 2: 77,8 Hz
4 Swing mould Mode 2

2
Deflection [mm]

-1

-2

-3

-4

-5
0 200 400 600 800 1000 1200
X-coordinate of the exhaust system [mm]

Fig. 15.21 Illustration of the vibration shapes of the second mode on the exhaust system

References

1. Ewins, D. J.: Modal Testing: Theory, Practice and Application, Aufl. 2. Research Studies Press,
Baldock (2003)
2. Døssing, O.: Structural Testing Part I: Mechanical Mobility Measurements. Brüel & Kjær, Nærum
(Revision April 1988)
3. Døssing, O.: Structural Testing Part II: Modal Analysis and Simulation. Brüel & Kjær, Nærum
(March 1988)
4. Kokavecz, J.: Messtechnik der Akustik, Kapitel 8. Modalanalyse. Springer, Berlin (2010)
5. Natke, H. G.: Einführung in Theorie und Praxis der Zeitreihen- und Modalanalyse – Identifikation
schwingungsfähiger elastomechanischer Systeme, Aufl. 3. Vieweg + Teubner, Wiesbaden (1992)
6. Brandt, A.: Noise and Vibration Analysis: Signal Analysis and Experimental Procedures. Wiley,
Chichester (2011)
7. Strohschein, D.: Experimentelle Modalanalyse und aktive Schwingungsdämpfung eines
biegeelastischen Rotors. Dissertation. (2011)
8. Peter, T.: Generalized Prony Method. Dissertation,. Göttingen (2013)