IOMAC'19
8th International Operational Modal Analysis Conference
2019 May12-14 Copenhagen
BASIC CONCEPTS OF MODAL SCALING
M. Aenlle1, and R. Brincker2
1
Dr, Dept. of Construction and Manufacturing Engineering, University of Oviedo, aenlle@uniovi.es
2
Prof., Department of Civil Engineering, Department of Civil Engineering, runeb@byg.dtu.dk
ABSTRACT
Mode shapes can be normalized in many different ways, the most common techniques being mass
normalization, normalization to the unit length of the mode shape (length scaling) and normalization to
a component (usually to the largest component) equal to unity (DOF scaling). For this reason, the modal
mass of a mode shape is not unique but it depends on the normalization method used to define the mode
shape. On one hand, the modal mass can be a mass or an inertia. On the other hand, the units of the
modal mass depend on the scaling technique and its magnitude depends on the number of degree of
freedoms (DOFs) used to discretize the model. In this paper, all these drawbacks are studied and a new
and better definition of the modal mass is presented, which leads to a better engineering interpretation
of this modal parameter
Keywords: Modal mass, Structural Dynamics, Modal Analysis
1. INTRODUCTION
When the modal model is used to define the dynamic behavior of a mechanical system, the modal
parameters (natural frequencies, mode shapes and damping ratios) are needed [1,2,3,4]. Mass
normalized mode shapes (hereafter denoted as 𝝓), contain information of both the mode shape and the
modal mass. A mode shape is defined un-scaled if it is not mass normalized. If un-scaled mode shapes,
hereafter denoted as 𝝍, are used, a new modal parameter for each mode, known as modal mass, is
needed to define the dynamic behavior of a mechanical system. When classical modal analysis (CMA),
also known as experimental modal analysis (EMA), is used to estimate the modal parameters of a
system, mass normalized mode shapes are obtained in the estimation process [1,2,3]. On the contrary,
in operational modal analysis (OMA) the force is not measured and only un-scaled mode shapes can be
estimated [4].
The equation of motion of a structural dynamic system with no damping subjected to a force 𝒑 is given
by [4,5,6]:
𝑲𝒖 + 𝑴𝒖̈ = 𝒑 (1)
Where 𝑲 is the stiffness matrix, 𝑴 the mass matrix and 𝒖 and 𝒖̈ are the displacement and the
acceleration vector, respectively. Eq. (1) provides the eigenvalue equation:
(𝑲 − 𝑴𝜔2 )𝝍 = 𝟎 (2)
Where 𝝍 and 𝜔 are the unscaled mode shape and the natural frequency, respectively.
Mode shapes can be normalized in many different ways but when working with experimental mode
shapes, the most common types of normalization are:
Mass normalization (or normalization to the mass matrix of the system).
Normalization to the unit length of the vector
Normalization to a component equal to unity (usually the largest component).
On the other hand, numerical mode shapes are generally mass normalized or normalized to a component
equal to unity.
The un-scaled (𝝍) and the scaled or mass normalized (𝝓) mode-shape vectors are related by the
expression:
1
𝝓=𝝍 (3)
√𝑚
Where 𝑚 is the modal mass and which can be obtained from the expression [4,5,6]:
𝑚 = 𝝍𝑇 𝑴𝝍 (4)
The length of a mode shape 𝝍 (also denoted in algebra as Euclidean norm, Euclidean length or 𝐿2 norm)
is given by:
𝐿𝜓 = √𝝍𝑇 ∙ 𝝍 (5)
A mode shape is said to be mass normalized if the modal mass is dimensionless unity, i..e:
𝑚𝜙 = 𝝓𝑻 𝑴𝝓 = 1 (6)
Mass normalized mode shapes 𝝓 can be obtained from the unscaled mode shapes 𝝍 by:
𝝍
{𝝓} = (7)
√𝝍𝑇 𝑴𝝍
A mode shape {𝝍} is normalized to the unit length, hereafter denoted 𝝍𝑳, if its length is unity, and it
can be obtained by:
𝝍 𝝍
𝝍𝑳 = = (8)
√𝝍𝑇 ∙𝝍 𝐿𝜓
Where the length of the vector 𝝍𝑳 is unity, i.e:
𝐿𝜓𝐿 = √𝝍𝑇𝑳 ∙ 𝝍𝑳 = 1 (9)
Another common type of normalization consists of assigning a magnitude equal to one to a component
of the vector (usually to the largest component). A mode shape normalized in this way is hereafter
denoted as 𝝍𝑼 . If the r-th DOF is chosen to be equal to unity, the mode shape 𝝍𝑼 is related to 𝝍 by:
𝝍
𝝍𝑼 = (10)
𝜓𝑟
Where 𝜓𝑟 is an scalar corresponding to the r-th component of the vector 𝝍.
If eq. (10) is particularized to mode shapes normalized to the unit length and to mode shapes normalized
to a component equal to unity, it results in:
𝟏 𝟏
𝝓 = 𝝍𝑳 = 𝝍𝑼 (11)
√𝑚𝜓𝐿 √𝑚𝜓𝑈
Where 𝑚𝜓𝐿 and 𝑚𝜓𝑈 are the modal masses corresponding to the mode shapes 𝝍𝑳 and 𝝍𝑼 , respectively.
Pre-multiplication of eq. (11) by 𝝓𝑻 , gives:
𝑚𝜙 = 1 𝑚𝜓𝐿 𝑚𝜓𝑈
= = (12)
𝐿2𝜙 𝐿2𝜓𝐿 =1 𝐿2𝜓𝑈
2. UNITS
The modal mass 𝑚𝜙 corresponding to the mass normalized mode shape 𝝓 is dimensionless unity. From
eq. (6) it is easily inferred that the translational components 𝝓𝑻 of the mode shape 𝝓 have the units
1 1
in the international system, whereas the units of the rotational components 𝝓𝑻 are . On the
√𝑘𝑔 𝑚√𝑘𝑔
other hand, the translational components of the mode shapes 𝝍𝑳 and 𝝍𝑼 are dimensionless and the
modal masses 𝑚𝜓𝐿 and 𝑚𝜓𝑈 are given in 𝑘𝑔. This has been summarized in Table 1.
Table 1. Units of modal parameters in the international system (SI). ----- indicates dimensionless.
Normalization to Normalization to
Mass normalization
unit length a component equal to unity
Variable Unit Variable Unit Variable unit
1
𝝓𝑻 𝝍𝑳 𝑻 ----- 𝝍𝑼 𝑻 ------
√𝑘𝑔
1 1 1
𝝓𝑹 𝝍𝑳 𝑹 𝝍𝑼 𝑹
𝑚√𝑘𝑔 𝑚 𝑚
𝑚𝜙 = 1 ----- 𝑚𝜓𝐿 𝑘𝑔 𝑚𝜓𝑈 𝑘𝑔
1 𝐿𝜓𝐿 = 1
𝐿𝜙 = √{𝜙}𝑇 ∙ {𝜙} ----- 𝐿𝜓𝑈 = √{𝜓𝑈 }𝑇 ∙ {𝜓𝑈 } -----
√𝑘𝑔 𝐿𝜓𝐿 = √{𝜓𝐿 }𝑇 ∙ {𝜓𝐿 }
𝑞𝜙 𝑚√𝑘𝑔 𝑞𝜓𝐿 𝑚 𝑞𝜓𝑈 𝑚
With respect to the length of mode shapes, 𝐿𝜓𝐿 and 𝐿𝜓𝑈 are dimensionless whereas 𝐿𝜙 have units of
1
.
√𝑘𝑔
Experimental mode shapes usually only contain translational DOF’s and they can be scaled with any of
the methods presented in section 1. On the other hand, some numerical mode shapes only contain
translational DOF’s (3D models, 2D planar elasticity models), whereas beams, plates and shells contain
both translational and rotational DOF’s. When a mode shape contains both translational and rotational
DOF’s, it cannot be normalized to the unit length (the length cannot be calculated because the units are
not consistent).
𝝍
The modal mass can also be a modal inertia. If the mode shape 𝝍 = { 𝝍 𝑻 }, containing both translational
𝑹
and rotational components, is normalized to a translational component equal to unity, the modal mass,
is given in 𝑘𝑔 (see table 1 ). However, if we normalize the vector 𝝍 to a rotational component equal to
unity, the translational DOF’s are given in meters and the rotational DOF’s are dimensionless. The
modal mass in this case is given in 𝑘𝑔𝑚 2 ,i.e. it is a modal inertia.
In analytical models both modal mass and modal inertia can be modal parameters of a mode shape.
However, mass normalization and normalization to a component equal to unity (usually the largest
translational component) are the normalization methods commonly available in numerical programs,
i.e. the modal mass (𝑘𝑔) is the parameter normally consider in finite element programs.
If the vector of displacements 𝒖 is decomposed in modal coordinates using mass-normalized mode
shapes:
𝒖 = 𝝓 ∙ 𝒒𝝓 (13)
the modal coordinates 𝒒𝝓 have the units 𝑚√𝑘𝑔.
If mode shapes 𝝍𝑳 are used in the modal decomposition:
𝒖 = 𝝓 ∙ 𝒒𝝍 𝑳 (14)
the translational components of the mode shapes are dimensionless, and the units of the modal
coordinates 𝒒𝝍𝑳 coincide with the units of vector 𝒖. The modal coordinates 𝒒𝝓 and 𝒒𝝍𝑳 are related by:
𝒒𝝓 = 𝒒𝝍𝑳 ∙ √𝑚𝜓𝐿 (15)
The same can be said for mode shapes 𝝍𝑼.
3. EXPANSION AND REDUCTION OF MODELS
In structural dynamics model reduction means to reduce a finite element model to one with fewer
degrees of freedom while the dynamic characteristics of the system are maintained [7]. In order that the
reduced model have the same modal parameters as the full model, the modal masses must remain
unchanged with all kinds of normalization, i.e.:
𝑚𝜙 = 𝑚𝜙𝑎 = 1, 𝑚𝜓𝐿 = 𝑚𝜓𝐿𝑎 𝑚𝜓𝑈 = 𝑚𝜓𝑈𝑎 (16)
Where the subindex ‘a’ indicates model with active DOF’s.
However, the following properties are modified:
The length of mode shapes 𝝓𝒂 , 𝝍𝑳𝒂 𝒂𝒏𝒅 𝝍𝑼𝒂 are different to that of the corresponding full
mode shapes (see table 2).
the reduced mode shape 𝝍𝑳𝒂 is no longer normalized to the unit length. If the reduced mode
shape is normalized to the unit length(𝝍∗𝑳𝒂 in table 2), the modal mass has to be modified
accordingly.
With mode shapes 𝝍𝑼 and 𝝍𝑼𝒂 , the DOF with the component equal to unity must be the same
in both the full and the reduced models. Otherwise, the modal mass has to be modified
accordingly.
All the aforementioned properties are showed in table 2 by means an example.
Table 2. Example of mode shapes and modal masses in reduction
Normalization to Normalization to
Mass normalization
unit length a component equal to unity
𝝓𝒂 { 0.9 } {0.7272} {1}
𝝓={ } = { 0.72 𝑎 } 𝝍𝑳 = { 0.5818 𝑎 } 𝝍𝑼 = { 0.8 𝑎 }
𝝓𝒅
{0.45}𝑑 {0.3636}𝑑 {0.5}𝑑
𝑚𝜙 = 1 𝑚𝜓𝐿 = 0.6535 𝑚𝜓𝑈 = 1.234
𝐿𝜙 = 1.1597 𝐿𝜓𝐿 =1 𝐿𝜓𝑈 = 1.3748
𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛
⇓ ⇓ ⇓
0.9 0.7272 0.7808 1
𝝓𝒂 = { } 𝝍𝑳𝒂 = { } 𝝍∗𝑳𝒂 = { } 𝝍𝑼𝒂 = { }
0.72 0.5818 0.6247 0.8
∗
𝑚𝜙𝑎 = 1 𝑚𝜓𝐿𝑎 = 0.6535 𝑚𝜓 𝐿𝑎
= 0.7535 𝑚𝜓𝑈𝑎 = 1.234
𝐿𝜙𝑎 = 1.0689 𝐿𝜓𝐿𝑎 = 0.9313 𝐿∗𝜓𝐿𝑎 = 1 𝐿𝜓𝑈𝑎 =1.2806
∗
The modal masses 𝑚𝜓𝐿 and 𝑚𝜓 𝐿𝑎
are related by means of the equation:
∗
𝑚𝜓𝐿 = 𝑚𝜓 ∙ 𝐿2𝜓
𝐿𝑎 𝐿𝑎 (17)
Where 𝐿2𝜓 is the square length of the vector 𝝍𝑳𝒂 .
𝐿𝑎
The same procedure must be followed with expansion of mode shapes.
4. MODAL ASSURANCE CRITERIA (MAC)
The modal assurance criteria (MAC) is the most used criteria to compare two mode shapes. We consider by
convenience the mode shapes 𝝓𝒙 and 𝝓𝑭𝑬, the subindices ‘x’ and’FE’ denoting experimental and numerical
models, respectively. The modal assurance criteria (MAC) is given as the square of the dot product between two
vectors with unit length [1,2,3,4], i.e.:
2
𝑀𝐴𝐶 = |𝝍𝑻𝒙𝑳 ∙ 𝝍𝑭𝑬𝑳 | = cos2 𝜃 (18)
From eq. (18) it is inferred that we cannot calculate the MAC between two vectors containing both translational
and rotational DOF’s, because these vectors cannot be normalized to the unit length.
In case of other kinds of normalization, the following expression must be used [1,2,3,4]:
𝟐 𝟐
|𝝓𝑻𝒙 ∙ 𝝓𝑭𝑬 | |𝝍𝑻𝒙𝑼 ∙ 𝝍𝑭𝑬𝑼 |
𝑀𝐴𝐶 = = (19)
|𝝓𝑻𝒙 ∙ 𝝓𝒙 ||𝝓𝑻𝑭𝑬 ∙ 𝝓𝑭𝑬 | |𝝍𝑻𝒙𝑼 ∙ 𝝍𝒙𝑼 ||𝝍𝑻𝑭𝑬𝑼 ∙ 𝝍𝑭𝑬𝑼 |
5. COMPARISON OF MODAL MASSES
Let's consider two mode shapes 𝝍𝒙𝑼 and 𝝍𝑭𝑬𝑼 with modal masses 𝑚𝑥𝑈 and 𝑚𝜓𝐹𝐸 , respectively. The
𝑈
modal assurance criteria (MAC) can be used to compare both mode shapes and the MAC is unique
because both mode shapes have to be normalized to the unit length of the vector. However, as it can be
seen in the next example, if we compare the modal masses of two modes, the discrepancies are different
depending on the normalization used for the mode shapes.
Example:
0.32 0.4577 0.5714
𝝓𝒙 = {0.56} → 𝝍𝒙𝑳 = {0.8009} → 𝝍𝒙𝑼 = { 1.0 }
0.27 0.3861 0.4821
𝑚𝑥𝐿 = 2.045 𝑚𝑥𝑈 = 3.189
0.30 0.4128 0.5085
𝝓𝑭𝑬 = {0.59} → 𝝍𝑭𝑬𝑳 = {0.8119} → 𝝍𝑭𝑬𝑼 = { 1.0 }
0.30 0.4128 0.5085
𝑚𝐹𝐸𝐿 = 1.894 𝑚𝐹𝐸𝑈 = 2.873
𝑚𝐹𝐸𝐿 𝑚𝐹𝐸𝑈
𝜖𝐿 = 100 ∙ (1 − ) = 7.38% 𝜖𝑈 = 100 ∙ (1 − ) = 9.909%
𝑚𝑥𝐿 𝑚𝑥𝐿
We do not obtain the same error comparing the modal masses of mode shapes normalized to the length
or comparing those corresponding to mode shapes normalized with the maximum component equal to
unity. This is because the mode shapes 𝝍𝒙𝑼 and 𝝍𝑭𝑬𝑼 do not have the same length (figure 1).
𝝓𝒙
1 𝝓𝑭𝑬
𝝍𝒙𝑼
𝝍𝒙𝑳
𝜃 𝝍𝑭𝑬𝑼
𝝍𝑭𝑬𝑳
1
Figure 1. Representation of two-dimensional vectors with different kinds of normalization
In operational modal analysis, the force is not measured and the modal mass cannot be estimated [4].
Some authors [8] have proposed to normalize experimental un-scaled mode shapes using the modal
masses of a finite element model. However, two aspects must be considered:
Mode shapes 𝝍𝒙𝑼 and 𝝍𝑭𝑬𝑼 have different length (see figure 1), which means that the modal
mass 𝑚𝐹𝐸𝑈 must not be used to scale 𝝍𝒙𝑼.
The vector 𝝍𝑭𝑬𝑼 cannot be normalized to the unit length if both rotational and translational
DOF’s are present in the mode shape.
If the experimental mode shape only contains translational components (as it is common in modal
analysis) the following procedure is recommended:
The modal mass 𝑚𝐹𝐸𝑈 and the mode shape 𝝍𝑭𝑬𝑼 are known from the FE model.
The mode shape 𝝍𝑭𝑬𝑼 is reduced to the active DOF’s (vector 𝝍𝑭𝑬𝑼𝒂 ) whose modal mass is
𝑚𝐹𝐸𝑈𝑎 = 𝑚𝐹𝐸𝑈
The modal mass 𝑚𝐹𝐸𝐿𝑎 of the vector 𝝍𝑭𝑬𝑳𝒂 can be obtained by:
𝑚𝐹𝐸𝑈𝑎
𝑚𝐹𝐸𝐿𝑎 =
𝐿2𝜓𝐹𝐸 (1)
𝑈𝑎
The experimental mode shape 𝝍𝒙𝑳𝒂 is normalized by:
𝝍𝒙 𝑳𝒂
̂ =
𝝓𝒙 𝒂
√𝑚𝐹𝐸𝐿𝑎 (1)
Where ‘ ̂ ′indicates approximation.
Example:
0.4577 0.5085 0.5085
𝝍𝒙𝑳𝒂 = {0.8009} 𝝍𝑭𝑬𝑼 = { 1.0 } → 𝝍𝑭𝑬𝑼𝒂 = { 1.0 }
0.5085
𝑚𝐹𝐸𝑈 = 2.873 𝑚𝐹𝐸𝑈𝑎 = 2.873
𝑚𝐹𝐸𝑈𝑎 2.873
𝑚𝐹𝐸𝐿𝑎 = = = 2.287
𝐿2𝜓𝐹𝐸 1.2586
𝑈𝑎
{
0.4577
} 0.303 0.32
̂ =
𝝓 0.800
={ } 𝝓𝒙 = {0.56}
𝒙 𝒂 √2.287 0.530
0.27
6. CONCLUSIONS
Modal mass is one of the modal parameters needed to define the dynamic behavior of a structure. The
values that the modal mass can take are dependent on the type of normalization. In this paper, the most
common types of normalization have been considered: mass normalization, normalization to the unit
length and normalization to the largest component equal to unit.
The units of the modal mass are also dependent on the normalization. The modal mass of mass
normalized mode shape is dimensionless unity. If the modal masses of two modes are compared, the
mode shapes have to be normalized to the unit length.
Although the length of a vector given by eq. (5) is usually considered in algebra and also in the field of
structural dynamics and modal analysis, it presents two main inconveniences:
Mode shapes with translational and rotational components cannot be normalized to length. As
a result, the modal assurance criteria (MAC) cannot be calculated when a mode shape have
both translational and rotational DOF’s.
The length of the mode shapes is dependent on the number of DOF’s consider in the mode
shape vector.
For the aforementioned reasons, mass normalization and normalization to a component equal to unity
(usually the largest translational component) are the normalization methods commonly available in
numerical programs.
Both modal mass and modal inertia can be modal parameters of a mode shape. The modal mass of a
mode shape normalized to a displacement equal to unity, is given in 𝑘𝑔. On the other hand, if the
normalization is to a rotation equal to unity, then modal mass is given in 𝑘𝑔𝑚 2 , i.e. it is a modal inertia.
The same can be said for mode shapes normalized to the unit length: modal mass in 𝑘𝑔 for a mode
shape with translational components and modal inertia in 𝑘𝑔𝑚 2 for mode shapes with rotational
components.
ACKNOWLEDGEMENTS
The financing support given by Banco Santander through the Mobility Grants for Researchers of
Professors of the University of Oviedo, is gratefully appreciated. The authors acknowledge the funding
received from Centre for Oil and Gas - DTU/Danish Hydrocarbon Research and Technology Centre
(DHRTC).
REFERENCES
[1] Ewins DJ (2000) Modal Testing: Theory, Practice and Application, Second Ed. London: Research
Studies Press LTD
[2] Maia NMM. and Montalvao e Silva JM (2003) Theoretical and Experimental Modal Analysis.
England: Research Studies Press LTD
[3] Heylen W, Lammens S and Sas P (2007) Modal Analysis theory and testing. Belgium: Katholieke
Universiteit Leuven, Faculty of Engenieering
[4] Brincker R., Ventura C. (2015). Introduction to Operational Modal Analysis. Chichester : John
Wiley & Sons Ltd.
[5] Clough RW and Penzien J (1993) Dynamics of structures, 2nd edition. New York: McGraw-Hill-
[6] Chopra AK (1995) Dynamics of Structures. New Jersey: Prentice Hall
[7] O’Callahan J.C., Avitabile P. and Riemer R. (1989). System Equivalent Reduction Expansion
Process. In: Proc. of the 7th Intl Modal Analysis Conference (IMAC), (pp.29-37),Las vegas, USA.
[8] Schwarz B, Richardson M. (2006). Using FEA modes to scale experimental mode shapes. In:
International modal analysis conference (IMAC XXIV), (paper 46), St Louis, USA.
