Presented at IMAC XXI February 3-6, 2003

Page 1 of 7
OBTAINING STRESSES AND STRAINS FROM ODS DATA

Brian Schwarz & Mark Richar dson
Vibrant Technology, Inc.
Jamestown, California

ABSTRACT
During an Operating Deflection Shape (ODS) test, there is
often concern not only about the displacements that occur
during operation, but also about the stress or strain levels
that are being encountered. If in-plane vibration were
measured, then strain could be estimated as the change in
displacement between two transducers divided by the dis-
tance between them. Unfortunately, vibration is usually
measured normal to the surface instead of in-plane, so stress
or strain cannot be calculated from this data.
However, if a finite element model is used in conjunction
with ODS test data, the model can be deformed using the
measured data and the appropriate stresses, strains and even
applied forces can be calculated. Details of this method and
several practical examples of its use are included in this
paper.
OBTAINING STRAINS FROM ODSs
When performing an ODS test, we can measure ODS FRFs
(magnitudes & phases) at several DOFs and obtain a set of
displacements that describe the deformation of a structure,
at any frequency [6]. The question then becomes, For a
given ODS of structural deformations, what are the associ-
ated stress and strain levels being experienced by the struc-
ture?
Since strain is defined as the rate of change in the deforma-
tion, it can always be calculated fromdisplacements. For
the simple case of a rod being extended and compressed
along its length, its strain would be calculated with the for-
mula,
L
x
(1)
where:
x = the amount of deflection of the rod.
L = the length of the rod.
Stress is simply related to strain by the modulus of elastic-
ity,
E (2)
In most practical cases however, the strain field is more
complex and includes the effects of structural bending and
torsion. For these cases, the equations relating displacement
and strain are much more complex, but finite element analy-
sis can be used to obtain a solution.
Finite element analysis is commonly used to determine
stress or strain levels due to certain static loading conditions
on a structure. Using this same approach, a finite element
model can also be used with experimental ODS data to cal-
culate the stresses and strains being experienced within a
structure.
BACKGROUND THEORY
The simplest model of a vibrating structure is a single-
degree-of-freedom (SDOF) system consisting of a mass
connected by a spring and damper to ground.

Figure 1. Mass-Spring-Damper (SDOF) Structure.
The motion of this structure is governed by Newtons Sec-
ond Law,
( ) ( ) ( ) ( ) t x K t x C t x M t f + + & & & (3)
where:
M = the distributed mass of the structure.
C = the damping within the structure.
K = the stiffness of the structure.
f(t) = the force applied to the structure over time.
x(t) = the displacement of the structure over time.
) t ( x& = the velocity of the structure over time.
) t ( x& & = the acceleration of the structure over time .
Presented at IMAC XXI February 3-6, 2003


Page 2 of 7
In the frequency domain, the equivalent equation of motion
is,
) s ( X ] K s C s M [ ) s ( F
2
+ + (4)
where:
X(s) = Laplace transform of the displacement.
F(s) = Laplace transform of the force.
+ j s complex Laplace variable
Assuming a quasi-static solution and removing the time
variable from equation (3), a simplified form of Newtons
Second Law relates the amount of deformation directly to
the applied forces,
x K f (5)
This is also known as Hookes Law. Since the force and
displacement are assumed to be constant (quasi-static) with
respect to time, the mass and damping terms (involving in
the velocity and acceleration) are negligible and are there-
fore not required.
Complex Structures
Newtons Second Law still governs the motion of more
complex structures, but instead of a single equation like (3)
or (4), multiple equations are written. The mass, stiffness,
and damping are replaced with matrices of constants, and
the force and motion terms become vectors instead of sca-
lars.
Solid Mechanics has been used to derive closed form solu-
tions to equations (3) or (4) for a variety of simplified ge-
ometries. Formulas to obtain stresses and strains in beams
and plates are compiled in Reference [1]. Formulas to obtain
the dynamic characteristics of beams and plates are com-
piled in Reference [2].
In general, it is impossible to derive a closed form solution
for a complex structure such as a machine or a bridge.
However, finite element modeling can be used to sub-divide
a complex structure into a series of simple elements such as
beams or plates. The stiffnesses of these simple elements
can be represented using known analytical formulae. Many
interconnected finite elements can then be used to form a set
of matrix equations that define the dynamics of complex
structures.
Each finite element provides a description of how the struc-
ture behaves in its local region. The stiffness matrix for the
overall structure is then constructed by calculating element
stiffnesses and also taking into account how each of the
elements interacts with its surrounding neighbors (i.e. its
boundary conditions).
A typical finite element is defined by points and straight line
boundaries connecting the points. A set of equations can
then be defined for the displacement anywhere inside the
boundaries of the element, based on the displacements of its
endpoints . Since displacements can be defined anywhere in
an element, stresses and strains can also be defined any-
where. The general equation for the stiffness matrix of an
element is ,

V
T
dV ] B ][ D [ ] B [ ] K [ (6)
where:
[D] = elasticity matrix.
[B] = strain matrix.
The strain matrix [B] transforms the displacements into
strains and the elasticity matrix [D] transforms the strains
into stresses. The form of the strain matrix and the elasticity
matrix will vary based upon the type of finite element used.
The elasticity matrix will also vary based upon the material
model being used (Isotropic, Orthotropic, etc.). Strain ma-
trices for different element types as well as numerical
integration techniques can be found in most finite element
text books [3], [4].
The finite element method generates stiffness matrices for
each element, and then sums them together to create a
global stiffness matrix. For complex structures, equation (5)
becomes,
[ ]
[ ] ( )
global local
global global global
x K
x K F

(7)
To solve equation (7), it is assumed that either the displace-
ment of a degree-of-freedom (direction at a point) is known,
or the applied force at the DOF is known. This allows the
solution of equation (7) to be partitioned as follows,

'

1
]
1

'

u
k
22 21
12 11
k
u
x
x
K K
K K
f
f
(8)
where:
u
f = unknown forces.
k
f = known forces.
u
x = unknown displacements.
k
x = known displacements.
Equation (8) can be solved for the unknown displacements
as follows,
[ ] { }
k 21 k
1
22 u
k 21 k u 22
k u 22 k 21
x K f K x
x K f x K
f x K x K


+

(9)
Presented at IMAC XXI February 3-6, 2003


Page 3 of 7
Equations (9) shows that the unknown displacements
u
x can be calculated from the known forces
k
f and known
displacements
k
x . Once all of the displacements in the
structure are known, equation (8) can be solved to determine
the unknown applied forces
u
f . Or we can go back to the
element matrices in (6) and compute stresses and strains
within the elements.
SHAPE EXPANSION USING FINITE ELE-
MENTS
From equation (9) it is clear that the deformations at un-
known DOFs
u
x are based not only on any known applied
forces, but also on known (measured) deformations.
In general, since only displacements are measured in an
ODS test, all of the known forces are assumed to be zero.
This assumption further simplifies equation (9) to,
[ ] [ ]{ }
k 21
1
22 u
x K K x

(10)
Equation (10) can therefore be used to extend (interpolate or
extrapolate) shape data to all unmeasured DOFs using data
from the measured DOFs.
Illustrative Example No. 1
Consider a 25 in. long 1 in. square steel cantilever beam
deflected 2 in. at its free end, as shown in Figure 2. Apply-
ing the 2 in. deflection as a prescribed (measured) displace-
ment and solving this problem using the NASTRAN finite
element program, shows that a 320 lb. force must be applied
at the free end of the beam to deflect it 2 in.
X
Y
Z
320.
8000.
320.
123456
2
2. 1.E-6 V1
L1
C1
Output Set: MSC/NASTRANCase1
Deformed(2.): Total Translation

Figure 2. Cantilever Beam Deflected by an End Load.
The moments in the beam start at zero at the free end, and
increase linearly to 8000 in-lbs at the fixed end.
The maximum bending stress at the fixed end of beam is ,
psi 48000
5 . 0 * 8000
I
Mc
12
) 1 (
4

where:
M = moment of inertia.
c = distance from neutral axis to t he top or bottom sur-
face.
I = cross sectional inertia.
With 2 in. of prescribed displacement at the free end, the
displacement at the midpoint of the beam is calculated as
0.625 in.
If this displacement is applied to the finite element model
instead of the 2 in. displacement at the free end, a different
solution will result, as shown in Figure 3.
X
Y
Z
840.
10080.
840. 840.
123456
2
0.5806 1.E-10
V1
L2
C2
Output Set: MSC/NASTRANCase1
Deformed(1.524): Total Translation

Figure 3. Cantilever Beam Deflected by a Center Load.
Moving the prescribed displacement also moves the location
of the applied force. Instead of applying a force at the free
end, it would be applied at the same location as the pre-
scribed displacement, at the midpoint.
Notice that a 0.625 in. prescribed displacement at the mid-
point displaces the free end only 1.524 in. instead of 2 in.
Since the load is applied at the midpoint and the beam is not
subjected to any other forces, it therefore remains straight to
the right of the midpoint.
When the beam is displaced from its midpoint, the forces,
moments and stresses in the beam will clearly be different
than when the free end is dis placed. To displace the mid-
point by 0.625 in., an 800 lb. force must be applied at the
midpoint. The moment at the fixed end of the beam in-
creases to 10000 in-lbs and the peak bending stress i n-
creases to 60000 psi.
This example shows that a finite element solution is possi-
ble if displacements are measured at all of the DOFs where
forces are applied to the structure. Otherwise, the load will
be redistributed and the calculated deformations, stresses
and strains will be incorrect.
Nevertheless, this method provides improvements over
purely geometrical interpolation methods that determine the
motions of unmeasured DOFs using the motions of nearby
measured DOFs. For the above case, most geometric inter-
polation methods would yield a straight line deformation
between the free and fixed end points. Furthermore, once
the motions of the unknown DOFs have been calculated, the
finite elements can be used again to calculate stresses and
strains.
Presented at IMAC XXI February 3-6, 2003


Page 4 of 7
IMPROVED SHAPE EXPANSION
The standard finite element solution (using the example
above) assumes that either the force or the displacement is
known at each DOF. However, this does not necessarily
need to be true to obtain a solution using equation (8). All
that is required is that the total number of unknown dis-
placements and forces must equal the total number of equa-
tions. This restriction is still satisfied if a measured dis-
placement is applied to one DOF, and a force is specified at
a different DOF.
To clarify this, the stiffness matrix can be re-written using 4
separate partitions.
1. DOFs with known displacements and unknown
forces. This includes all DOFs that act as boundary
conditions, and any DOFs where ODS data is meas-
ured, and forces are applied but unknown.
2. DOFs with known displacements and known forces.
For most vibration tests, these are the DOFs where
ODS data is measured, but where no external forces are
applied.
3. DOFs with unknown displacements and unknown
forces. These are the DOFs at which displacements are
not measured, and forces are applied but also unknown.
The number of DOFs of this type will equal the number
of DOFs in partition 2.
4. DOFs with unknown displacements and known
forces. These are the DOFs for which displacements
are not measured, and no forces are applied.
Equation (8) now expands into,

'

1
1
1
1
]
1

'

4
3
2
1
44 34 42 41
34 33 32 31
24 23 22 21
14 13 12 11
3
1
x
x
x
x
K K K K
K K K K
K K K K
K K K K
0
f
0
f
(11)
Partitioning this matrix and using the known forces and dis-
placements to solve for the unknown displacements gives,

'

1
]
1

1
]
1

'

'

1
]
1

'

1
]
1

'

1
]
1

'

1
]
1

'

2
1
42 41
22 21
1
44 34
24 23
4
3
2
1
42 41
22 21
4
3
44 34
24 23
4
3
44 34
24 23
2
1
42 41
22 21
x
x
K K
K K
K K
K K
x
x
x
x
K K
K K
x
x
K K
K K
x
x
K K
K K
x
x
K K
K K
0
0
(12)
In order to solve equation (12), the inverse of a non-
symmetric matrix that is the size of the number of unknown
DOFs must be calculated. This matrix is also not sparsely
populated, so the usual finite element solution methods that
apply to symmetric, sparse matrices cannot be used. Conse-
quently, our ability to solve problems with large numbers of
unknown DOFs is limited.
It is also possible to create a singular matrix that cannot be
inverted. This can be done by clustering the unknown
forces together, and isolating the unknown DOFs from the
forces.
Illustrative Example No. 2
Consider a stepped aluminum rod subjected to an end load
with a measured deflection of 5 mils at point 4 (the mid-
point).

Figure 4. Cantilever Beam with Variable Cross Section.
For a Rod with fixed cross section, the stiffness matrix is ,
1
]
1

1 1
1 1
L
EA
K (13)
where:
L = the Rod length.
A = the Rod cross sectional area.
E = Modulus of elasticity.
The stresses and strains in the Rod are

E
L
) u u (
L
u
2 1
(14)
Consider the six cross sections of the beam in Figure 4 to
have the following properties,
L=10in
E=10
7
psi
2
7 6
2
6 5
2
5 4
2
4 3
2
3 2
2
2 1
in 1 A , in 2 A , in 4 A
in 6 A , in 8 A , in 10 A





These parameters were used to assemble the partitioned
stiffness matrix shown in equation (11). Rows 1 & 7 of the
stiffness matrix are not used because they correspond to the
Presented at IMAC XXI February 3-6, 2003


Page 5 of 7
unknown forces,
1
F &
7
F . Columns 1 & 4 are partitioned
to be mult iplied by the known displacements,
1
u = 0 &
4
u = 0.005. The remaining rows & columns form the parti-
tion of the stiffness matrix to be inverted.

'

1
1
1
1
1
1
1
1
1
]
1

'

7
6
5
3
2
7
7
1
u
u
u
005 . 0
u
u
0
1 1 0 0 0 0 0
1 3 2 0 0 0 0
0 2 6 4 0 0 0
0 0 4 10 6 0 0
0 0 0 6 14 8 0
0 0 0 0 8 18 10
0 0 0 0 0 10 10
10
10
F
0
0
0
0
0
F
(15)
Solving (15) for the unknown displacements gives,

'

'

'

1
1
1
1
1
1
]
1

1
1
1
1
1
1
]
1

'

'

1
1
1
1
1
1
]
1

'

1
1
1
1
1
1
]
1

,
_

'

1
1
1
1
1
1
]
1

'

1
1
1
1
1
1
]
1

'

0273404 . 0
0145745 . 0
10 x 1915 . 8
10 x 8723 . 2
10 x 2766 . 1
u
u
u
u
u
005 . 0
0
0 0
4 0
10 0
6 0
0 10
1 3 2 0 0
0 2 6 0 0
0 0 4 6 0
0 0 0 14 8
0 0 0 8 18
u
u
u
u
u
u
u
u
u
u
1 3 2 0 0
0 2 6 0 0
0 0 4 6 0
0 0 0 14 8
0 0 0 8 18
005 . 0
0
0 0
4 0
10 0
6 0
0 10
u
u
u
u
u
1 3 2 0 0
0 2 6 0 0
0 0 4 6 0
0 0 0 14 8
0 0 0 8 18
005 . 0
0
0 0
4 0
10 0
6 0
0 10
10
0
0
0
0
0
3
3
3
7
6
5
3
2
1
7
6
5
3
2
7
6
5
3
2
7
6
5
3
2
6
(16)
The forces
1
F &
7
F can now be calculated from the dis-
placements.
( )
( )
lb 12766
0273404 . 0 * 1 0145745 . 0 * 1
10
10
u K F
lb 12766
10 x 2766 . 1 * 10 0 * 10
10
10
u K u K
u K F
7
7
1 i
i i 7 7
3
7
2 12 1 11
7
1 i
i i 1 1

+
+

As expected for a static solution, the applied forces are
equal and opposite. This is known as static equilibrium.

Point u(in) u (in) (in/in) (psi)

1 0.0000
0.0012766 1.2766E-4 1277
2 0.0012766
0.0015957 1.596E-4 1596
3 0.0028723
0.0021277 2.128E-4 2128
4 0.005
0.0031915 3.192E-4 3192
5 0.0081915
0.006383 6.383E-4 6383
6 0.0145745
0.012766 1.277E-3 12766
7 0.0273404

Table 1. Calculated Results for Cantilever Beam.
For this solution, a 5x5 non -symmetric matrix was inverted.
A finite element solution would have partitioned the stiff-
ness matrix as follows and stored the data in banded form
allowing for a much more efficient storage of data.
Presented at IMAC XXI February 3-6, 2003


Page 6 of 7
1
]
1

1
1
1
1
1
1
]
1

1
1
1
1
1
1
1
1
1
]
1

0 1 2 0 8
1 3 6 14 18
10
10
K
1 1 0 0 0
1 3 2 0 0
0 2 6 0 0
0 0 0 14 8
0 0 0 8 18
10
10
K
1 1 0 0 0 0 0
1 3 2 0 0 0 0
0 2 6 4 0 0 0
0 0 4 10 6 0 0
0 0 0 6 14 8 0
0 0 0 0 8 18 10
0 0 0 0 0 10 10
10
10
K
7
banded
7
22
7

Storing the data in band form would have reduced the re-
quired memory by 60%. Finite element solution methods
also employ bandwidth optimizers that allow them to opti-
mize data storage. Memory reductions of 10 to 1 are often
achieved. Solutions that require over 1,000MB of computer
memory can be solved using less than 100MB of memory.
As previously stated, this method may not always provide a
solution. For instance, in Illustrative Example No.2 above,
if a force were also been applied to point 6, the matrix to be
inverted would have a column of zeroes and would be sin-
gular. Of course this could be alleviated by specifying the
displacement of point 7.
The sum of the forces applied at points 6 & 7 could then be
calculated, along with the displacements of the other points.
However, the displacement of point 7 and the distribution of
forces between points 6 & 7 still could not be determined.
USING VELOCITY OR ACCELERATION
DATA
Accelerations or velocities are more commonly measured
than displacements in an ODS test. This is not a problem
however, since it is straightforward to convert from one set
of motions to another in the frequency domain.
To convert from dis placement to velocity in the frequency
domain, the Laplace transform is used for differentiation
[5].
{ }
) j ( X
) j ( X ) j (
) j ( X j ) j ( X
) j ( X j ) j ( X
) s ( sX ) s ( X
) 0 ( F ) s ( sf ) t ( ' F L
2
2

& & &
&
&
(17)
Multiplying the left and right hand sides of equation (17) by
j results in the following equations for using velocity
measure ments,

'

1
]
1

1
]
1

'

'

1
]
1

1
]
1

'

2
1
42 41
22 21
1
44 34
24 23
4
3
2
1
42 41
22 21
1
44 34
24 23
4
3
x
x
K K
K K
K K
K K
x
x
x
x
K K
K K
K K
K K
j
x
x
j
&
&
&
&
(18)
For using acceleration measurements,

'

1
]
1

1
]
1

'

'

1
]
1

1
]
1

'

2
1
42 41
22 21
1
44 34
24 23
4
3
2
1
42 41
22 21
1
44 34
24 23
4
3
x
x
K K
K K
K K
K K
x
x
x
x
K K
K K
K K
K K
j
x
x
j
& &
& &
& &
& &
&
&
&
&
(19)
Equations (18) and (19) show that Shape Expansion is done
the same way using displacement, velocity, or acceleration
frequency domain data, or using displacement time domain
data. Of course, to compute stresses and strains, displace-
ment data is still required. Frequency domain velocity or
acceleration data can be used directly, but the results must
be integrated to displacements in order to calculate stresses
and strains.
CONCLUSIONS
A method for calculating structural stresses and strains from
experimental ODS data was introduced. The method relies
on finite element analysis to calculate the stiffness proper-
ties of the structure, from which all unmeasured displace-
ments are then calculated, including in-plane displacements
which are not typically measured. In-plane displacements
are required to calculate stresses and strains.
Two simple examples were included to illustrate the calcu-
lations required to implement this method. Standard finite
element analysis generally assumes that the forces are ap-
plied at the same DOFs where the experimental ODS data is
taken. It was shown that this assumption can be relaxed. By
modifying the assump tion regarding the known and un-
known forces, it was shown by example that a valid solution
can still be obtained using the same finite element equa-
tions.
Presented at IMAC XXI February 3-6, 2003


Page 7 of 7
The new solution is somewhat restricted in that it requires
the inversion of a non-sparse non-symmetric matrix. There-
fore, large model sizes may be computationally prohibitive.
This new approach offers two significant advantages. First,
it yields more realistic shape interpolations than geometri-
cally based interpolation methods, and it can also be used to
extrapolate shapes, which is not possible with geometric
interpolation except in the simplest cases. Secondly, by
combining experimental ODS data with finite element mo d-
eling, this approach provides an alternative for determining
structural stress and strain than either finite element analysis
alone or the use of experimental strain gages.
REFERENCES
[1] Young, W.C. Roarks Formulas for Stress & Strain,
6
th
Edition, McGraw-Hill, 1989.
[2] Blevens, R. Formulas for Natural Frequency and Mode
Shape Robert E. Krieger Publis hing Co., Malabar, FL,
1979.
[3] Zienkiewicz, O.C. & Taylor, R.L. The Finite Element
Method, Volumes 1 & 2, 5
th
Edition, Butterworth-
Heinemann, 2000.
[4] Bathe, K.J. & Wilson, E.L. Numerical Methods in Fi-
nitel Element Analysis, Prentice-Hall, 1976.
[5] Richardson, M. "Is It A Mode Shape Or An Operating
Deflection Shape?" Sound and Vibration Magazine, Febru-
ary, 1997. (Vibrant Tech. Paper No. 10 www.vibetech.com)