App Note #5

www.vibetech.com

4/19/2004

MEscopeVES Application Note #5

Impact Testing Question
INTRODUCTION
There are two ways to perform a modal test on a structure
using an impact hammer and a tri-axial accelerometer.
One test is called a roving impact test and the other is
called a roving accelerometer test.
THE QUESTION
If both of these methods are applied to a structure with
nine measurement points of interest, do the two test methods yield the same modal information?
Roving Impact Test
In a roving impact test, Frequency Response Function
(FRF) measurements are made by attaching the tri-axial
accelerometer at a fixed point in the structure (say Point 9),
and impacting the it at points 1 through 9 in the Z direction. Since three acceleration outputs are simultaneously
measured for each force input, a total of 27 FRF measurements are gathered.

Roving Accelerometer Test.

Even though the structure illustrated is a flat plate, this question really applies to testing any structure. The Z direction
can be any direction you choose, and it can be a different
direction at each point.

THE ANSWER
The two tests are not equivalent. The roving impact test will
provide mode shapes with nine degrees-of-freedom (DOFs).
Motions at the nine test points in the Z direction will be
contained in each mode shape.
In contrast, the roving accelerometer test will provide mode
shapes with 27 DOFs. The mode shapes will exhibit motion
in the X, Y and Z directions at each of the nine test point
locations.
On the surface, the roving impact test appears to offer little
merit. However, this is not true. This test provides three
measured estimates for each mode shape component; the
roving accelerometer test only provides one.

Roving Impact Test.

Roving Accelerometer Test
In a roving accelerometer test, FRF measurements are
made by consistently impacting the structure at the same
point and in the same direction (say 9Z, for example). The
tri-axial accelerometer is attached to point 1 for the first
measurement and is then moved to point 2, and so on.
Again, a total of 27 FRF measurements are collected.

Should the structure prove difficult to analyze, the data

measured by the roving impact test can be fitted using multiple reference methods that can separate repeated roots or
deal with structures that have groups of local modes, shapes
physically segregated by direction or location. Data from
the roving accelerometer test cannot be used in this manner.
In short, the roving tri-axial accelerometer test is a single
reference test measuring 3 DOFs at each point. The roving
impact test using a triaxial accelerometer is a three reference
test measuring 1 DOF at each point.

Page 1 of 7

App Note #5

www.vibetech.com

A MORE DETAILED EXPLANATION

4/19/2004

[H ( j )] = 21j ( j[R (kp)] ) ( j[R (kp)] )

M

An experimental modal analysis characterizes dynamics at

N degrees-of-freedom (DOFs), each a motion in a specific
direction at a particular point on the structure. Hence,
there are N2 possible FRFs that could be measured between pairs of the N DOFs. These measurements can be
arranged in an N by N square matrix, Hij, where Hij = displacement at DOF, i, per unit force applied to DOF, j, as a
function of frequency.

ij

*
k

(1)

where:
Hij(j) = the element in the ith row and jth column of the
FRF matrix, representing displacement of DOF, i, resulting from a force applied to DOF, j, with typical
units of m/N or in/lb.

In general, it is unnecessary to measure all N2 FRFs. Most

structures can be completely characterized by measuring
just a single row or a single column of the FRF matrix.
That is, only N FRFs normally need to be measured. The
reason for this is explained later.

pk = the kth pole of the structure.

=

k + j k

(rad/sec)

k = damped natural frequency of the kth mode.

Modes of vibration are basically defined by three parameters:

1.
2.
3.

k =1

*
ij

ij

k = damping of kth mode.

Modal frequency
Modal damping
Mode Shape

(rad/sec)

Rij(k)= Residue matrix element relating DOFs i and j

in the kth mode.
(m/Ns or in/lb-sec)

= the forcing frequency

When the modes are used to predict the effects of a Structural Modification or combined with other modal models
through Substructuring, a fourth modal parameter, termed
modal mass, is also needed.
The test will capture data needed to identify many modes
(say M of them) from the one set of measurements. Every
FRF will exhibit characteristics reflecting the frequency
and damping of mode shapes. The set of N FRFs is needed
to identify mode shapes. Each mode shape vector has N
elements, one for each DOF measured.
Often, a structure will exhibit strong spatial separation of
modes, where a single DOF cannot be found that reflects
all M modes. These modes are referred to as local modes.
In this instance, either a (fixed) reference DOF must be
chosen which exhibits all of the modes, or two or more columns or rows must be measured to gather enough data for
a complete modal analysis.
More than N FRFs may also be required from structures
with strong spatial symmetry. Such structures often exhibit
repeated roots (two or modes at the same frequency, each
with a different mode shape). Again, two or more rows or
columns from the Hij matrix must be measured to complete
the modal analysis.

Lets examine the analytical formula for the FRF matrix to

see how these parameters are contained in a set of measurements. The FRF matrix can be written in partial fraction expansion form as a summation of pairs of terms, each
pair containing the contribution of a single mode.

(rad/sec)

j = 1 and * indicates complex conjugation.

Each FRF is the sum of 2M terms. Each term contains a
residue divided by a pole. The denominator of all FRFs is
the same; each term contains the same pair of poles for each
mode, k.
The FRFs differ only by numerators. Each numerator contains a specific residue, Rij(k), which is dependent upon the
response (i) and excitation (j) DOFs and each mode, k.
Residues are physical properties of a structure and have the
units of velocity/force (m/Ns or in/lb-sec). A residue matrix
is an N by N collection of the residues associated with a
single mode. When FRFs are measured, we actually measure the elements of M residue matrices simultaneously, one
for each mode encountered.
Normally, the FRF matrix is assumed to be symmetric. This
follows from the assumption of structural reciprocity. Consequently, the residue matrix is also symmetric.
In single-reference testing, we measure a single row or column of each residue matrix. This is (normally) sufficient
information to represent the entire residue matrix.

Page 2 of 7

App Note #5

www.vibetech.com

4/19/2004

It is shown later that each residue matrix contains one

mode shape in each of its rows and columns. Each residue
matrix element is the product of two mode shape elements
divided by the modes modal mass.
GLOBAL MODES
Since the same poles are contained in all FRF measurements, we can (in principle) identify all M modal frequencies
and damping from a single measurement. We gather a larger number of measurements, N, solely to collect the residues needed to determine the mode shapes.

The Global Properties (frequency and damping), contained in the poles are present in every measured FRF.
However, they are easier to detect from some measurements than from others. In general, a particular pole is best
identified from measurements that have a large resonance
peak (large associated residue) for that pole.

Local mode behavior in a two-plate structure.

Local modes are often the result of deliberate design intent.
In this instance, the four mounting springs were placed on
the centerlines of the small plate. These locations are Z-axis
nodal points for many of the small plates modes, including
all torsional modes.

Overlaid FRF Magnitudes show common poles.

The figure above shows the magnitudes of a number of
FRFs overlaid on one another. Notice that the resonance
peaks all appear at the same frequencies. This shows that
the modal frequencies of all of the modes are the same no
matter where the FRFs are measured on the structure. This
shows the global nature of modal frequencies.

LOCAL MODES
In structures comprised of multiple components, it is not
uncommon to encounter a local mode phenomenon. The
combination of two plates joined by four springs shown at
right illustrates such a situation. The small plate exhibits
some modes in which the base plate does not participate
and vice versa. Driving-point FRFs gathered on the small
plate (red) show some different resonances from those
gathered on the base plate (blue).

Page 3 of 7

Driving-point FRFs show local mode behavior.

App Note #5

www.vibetech.com

4/19/2004

As another example, automotive exhaust systems are attached to the car chassis by hangers deliberately placed at
nodal points of the exhaust piping.

MULIPLE REFERENCE TEST

When local modes are encountered, a multiple reference
analysis is required so that no modes will be missed. As an
example, using a single reference on the base plate of the
previously illustrated two-plate structure would fail to capture the local modes of the small mounted plate.
Using a single reference on the small attached plate would
correct this, but then some local modes of the base plate
would be missed. Either single reference test would capture the global modes common to both plates, but miss
some of the local modes.
Multiple reference testing and analysis is also required
when the structure is very simple and geometrically symmetrical. Consider the first five modes of a free-free square
plate illustrated at right. While the first three modes are
unique roots amenable to single reference testing, the
fourth and fifth modes are a repeated-root pair.
A square plate will exhibit many pairs of repeated roots.
The mode shape for one of these modes is always identical
to the shape for the other, but is rotated by 90 about the
axis of symmetry (Z in this instance). Repeated roots of
higher multiplicity than two are also possiblebut are rarely
encountered.

A free-free square plate exhibits repeated roots.

When repeated poles are encountered, special processing of
multiple rows or columns of FRFs is required to separate
the different shapes that occur at the same frequency.
Driving-point FRFs from square plate.
Measured FRFs give no direct indication of a repeated
root. The square plate measurements above include two
pairs of repeated poles (at red dots) and six distinct roots.

Suggestion: you can use two accelerometers and the Coherence Function to test your structure for a suspected repeated
root. Measure the coherence between accelerometers at two
different locations while impacting the structure at many different locations. Read the Coherence at the resonance frequency in question. A value greater than 0.9 indicates a
unique pole; a low value indicates a repeated root.

Page 4 of 7

App Note #5

www.vibetech.com

{x} = [U ]{q}

MORE ABOUT MODE SHAPES

The measured motion of a structure at N DOFs may be listed in a

{x}such that:

vector of N elements,

(5)

Where {q} is a vector of M modal coordinates.

Experimental versus Analytical Modes

x1
x
2
{x} = :
x
i
x N

(2)

Where xi is the displacement of the ith degree-of-freedom.

This vector is also called an Operating Deflection Shape (ODS)
or simply a deflection shape.
The first principle of Modal analysis is that any measured overall
motion vector can be expressed as a linear combination of vectors having known and fixed patterns called mode shapes. Each
mode shape can be written as a vector with N elements. For example, the kth mode shape,{uk} , may be stated:

u1k
u
2k
{u k } = :
u
ik
u Nk

In experimental modal analysis, we start with a series of

FRF measurements and work toward a model of the structure in terms of its modal parameters. In finite element
analysis (FEA) and other analytic vibration studies, we seek
exactly the same solution, but from a different starting
point: a set of differential equations.
It is useful for the experimentalist to appreciate how mode
shapes solve the analytical problem. This will also show the
relationship between mode shapes and residues.
Finite element analysis is used to develop three N by N matrices, [M], [C] and [K], that describe the dynamics of a
structure in terms of N degrees-of-freedom. A vector of N
forces, {F} to be applied to the structure completes the
model. These three matrices and force vector are arranged
as the set of second order differential equations, shown
below.

[M ]{&x&}+ [C ]{x&}+ [K ]{x} = {F }

(6)

(3)

In general, these equations are coupled meaning that the offdiagonal terms in the matrices are non-zero. They are typically solved to yield a set of N modal vectors typified in
form by (4).

Where uik is the motion of the ith DOF in the kth mode.
M mode shape vectors may be arranged as the columns of
a mode shape matrix,[U] , of the form:

u11
u
21
[U ] = u31

:
u N 1

4/19/2004

u12

.. u1k

u12
u12
:

.. u 2 k
.. u3k
.. :

u12

.. u Nk

u1M
u 2 M
u3 M

:
u NM

(4)

Two physical properties are assumed in the matrices, [M],

[C] and [K]. They are symmetric because the structure exhibits measurement reciprocity (this also implies that Hij =
Hji). Second, [K] is assumed to be positive definite because
the structure is statically determinant (does not require external forces to hold it in position).
Because of the symmetric and positive definite matrix properties, the solution vectors exhibit generalized orthogonality
with respect to [M], that is:

{u k }T [M ]{u k } = mk

Therefore, any motion the structure may exhibit can be

stated in terms of modal parameters:

and:

{u k }T [M ]{u m } = {u m }T [M ]{u k } = 0
Where mk is the modal mass of the kth mode.

Page 5 of 7

(7a)
(7b)

App Note #5

www.vibetech.com

The importance of generalized orthogonality may be better

appreciated by returning to the original problem statement
(6) (with [C] assumed equal to [0]). All of the modal vectors may be grouped together as a transformation matrix in
the manner of (5) and substituted in (6). Then premultiplying these equations by the transpose of the transformation results in:

[U ] [M ][U ]{q&&}+ [U ] [K ][U ]{q} = [U ] {F }

T
= [O mk ]{q&&} + [O mk k2 ]{q} = [U ] {F } = {Q}
T

4/19/2004

RETURNING TO RESIDUES
The residues retain the physical units of an experimental
modal analysis. Unlike modal mass, a residue is a physical
constant. Its value is independent of any scaling applied to
the mode shapes.
A residue is equal to two DOF elements of a mode shape
vector divided by its modal mass. Specifically:

(8)

Rij (k ) =

Where [U] is the matrix transpose of [U] and the vector, {Q}, is called a generalized force.
Note in (8) that the generalized orthogonality applies to
both the mass and stiffness matrices, and that both resulting matrices are diagonal. The diagonal elements, mk, are
the modal masses (sometimes called generalized masses)
of the system.
For lightly damped structures, even those exhibiting complex modes, it can also be shown that the damping matrix
is diagonalized.

[U ]T [M ][U ] [O mk ]

(9)

[U ]T [C ][U ] [O 2mk k ]

(10)

[U ]T [K ][U ] [O mk ( k2 + k2 )]

(11)

uik u jk
mk

u jk uik
mk

= R ji (k )

(12)

Where uik = displacement of DOF, i, in the kth mode,

ujk = displacement of DOF, j, in the kth mode and
mk = modal mass of the kth mode.
It is clear then, that there is a symmetric matrix of residues,
[Rij(k)], for each mode, k, measured from the structure. The
symmetry of the residue matrices properly reflects structural reciprocity.
It is also clear that the residues reflect the mode shapes redundantly. Every row contains the complete mode shape
multiplied the response DOF element of the vector. Likewise, every column contains the complete modal vector
multiplied by the shape element for the excitation DOF.

WITH REFERENCE TO THE INITIAL QUESTION

It is often convenient to scale modal vectors so that all of

their associated modal masses are equal to 1 mass unit
(kg, lbm, etc.) Vectors scaled in this manner are said to be
scaled to unit modal mass (UMM).

It is clear from equations (8) and (9) that modal mass is

proportional to the square-root of the length (magnitude)
of the mode shape vectors in both the normal mode and
complex mode cases.
Another important property of modal vectors is that while
their shape is unique, their values are not. Equations (9)
to (11) show that modal mass and the mode shapes are
related to one another. Both are arbitrary, but if a value is
chosen for one, the other becomes fixed also.

For the roving accelerometer test, the FRFs measured, and

consequently the residues and mode shape components obtained from them, are tabulated below.
FRFs
H1x,9z
H1y,9z
H1z,9z
H2x,9z
H2y,9z
H2z,9z

H9x,9z
H9y,9z
H9z,9z

Page 6 of 7

Residues
R1x,9z = u1x u9z
R1y,9z = u1y u9z
R1z,9z = u1z u9z
R2x,9z = u2x u9z
R2y,9z = u2y u9z
R2z,9z = u2z u9z

R9x,9z = u9x u9z

R9y,9z = u9y u9z
R9z,9z = u9z u9z

Mode Shape Components

u1x

u1y

u1z

u2x

u2y

U2z

9x
u9y

u9z

App Note #5

www.vibetech.com

When a roving accelerometer test is performed, the excitation DOF is fixed. That is, one component (the reference
ujk component) of each Rij(k) is the same. Each time we
position the tri-axial accelerometer, we gain three new
response uik components. Hence, this test measures a single
column of the M residue matrices. In nine positionings, we
measure 27 DOFs.
This column of FRFs will yield one estimate of each mode
shape in 3 directions (X, Y, and Z) at each of the 9 points.
The residues are also arranged in the one column of the
residue matrix as it shown below.

FRFs
H9x,1z
H9y,1z
H9z,1z
H9x,2z
H9y,2z
H9z,2z

H9x,9z
H9y,9z
H9z,9z

4/19/2004

Residues
R9x,1z = u9x u1z
R9y,1z = u9y u1z
R9z,1z = u9z u1z
R9x,2z = u9x u2z
R9y,2z = u9y u2z
R9z,2z = u9z u2z

R9x,9z = u9x u9z

R9y,9z = u9y u9z
R9z,9z = u9z u9z

Mode Shape Components

u1z

u1z

u1z

u2z

u2z

u2z

u9z

u9z

u9z

Residue Matrix
1Z

2Z

3Z

It is clear that the residues yield three estimates of each

mode shape, each shape with 9 DOFs 1Z, 2Z, 9Z. The
residues are arranged in three rows of the residue matrix, as
shown below.

9Z

R1 X , 9 Z 1 X
R1Y ,9 Z 1Y
R1 Z , 9 Z 1 Z

R 2 X ,9 Z 2 X
R2Y ,9 Z 2Y

R 2 Z ,9 Z 2 Z

R9 X , 9 Z 9 X

R9Y ,9 Z 9Y
R9 Z ,9 Z 9 Z

Residue Matrix
1Z

R
9 X ,1 Z
R9 Y , 1 Z
R
9 Z ,1 Z

For the roving impact test, the FRFs measured, and consequently the residues and mode shape components obtained
from them, are tabulated at right above.
When we perform a roving impact test, the response sensor
remains at a fixed location, oriented in a fixed direction.
Hence, for every impact measurement, the reference uik
component of each Rijk is the same. At each new strike
point of the hammer, a different ujk component is encountered. Hence this type of test measures one row of the residue matrix. When a tri-axial accelerometer is used, three
rows are measured simultaneously as there are three reference DOFs at the fixed accelerometer site.

2Z

3Z

R9 X , 2 Z

R9 X ,3 Z

R9Y ,2 Z

R9Y ,3 Z

R9 Z ,2 Z

R9 Z ,3 Z

9Z

1X

1Y
1Z

R9 X , 9 Z 9 X

R9Y ,9 Z 9Y
R9 Z ,9 Z 9 Z

Note, however, that the three rows contain only a single

driving-point measurement (9Z:9Z). Any single row contains sufficient information to calculate the modal frequency, modal damping and mode shape. However, modal
mass can only be calculated if the 9Z row is included in the
analysis.
With the driving point measurement, UMM scaled mode
shapes can be determined. This is a matter of concern only
if the results are to be used in a Structural Dynamics Modification (SDM).
REFERENCES
1.

Richardson, M., Modal Mass, Stiffness and Damping,

Vibrant Technology, Inc., January 2000.

1.

Frazer, R. A., Elementary Matrices, The Macmillan

Company, New York, NY, 1946.

Page 7 of 7