FE-Fatigue Rel.

5
1st Page Back Main Menu
Theory Page 1

FE-Fatigue theory – Table of contents

1. Introduction to fatigue analysis from finite element models ................. 3

2. Overview of new features in FE-Fatigue Rel. 5........................................ 6

2.1 Auto elimination by Group ............................................................................................ 6
2.2 Virtual strain gauges...................................................................................................... 6
2.3 Seam weld analysis........................................................................................................ 6
2.4 High temperature fatigue analysis................................................................................ 6
2.5 Vibration fatigue analysis.............................................................................................. 6

3. Fatigue Theory............................................................................................ 7
3.1. The S-N (High Cycle Fatigue) Approach ..................................................................... 7
3.2. The Local Strain theory of fatigue ............................................................................... 8
3.3. Multiaxial strain-life ...................................................................................................... 9
3.3.1 Introduction ............................................................................................................ 9
3.3.2. Co-ordinate System Rotation............................................................................... 9
3.3.3. Assessing Multiaxiality....................................................................................... 10
3.3.4. Critical Plane Analysis........................................................................................ 11
3.3.5. The Hoffmann-Seeger Method ........................................................................... 12
3.3.6. The Klann-Tipton-Cordes Method ..................................................................... 13
3.3.7 Non-proportional multi-axial methods ............................................................... 15
3.3.8 Multiaxial notch correction procedure ............................................................... 16
3.3.9 Cyclic Plasticity Modelling .................................................................................. 17
3.3.10 Multiaxial Rainflow Counting ............................................................................ 20
3.3.11 Fatigue Damage Calculation ............................................................................. 21
3.3.12 Critical Plane Methods....................................................................................... 22
3.3.13 Types of Cracks ................................................................................................. 23
3.3.14 Critical Plane Rainflow Counting...................................................................... 24
3.3.15 References.......................................................................................................... 25
3.4. Safety factor................................................................................................................. 26
3.4.1 Life based factor of safety................................................................................... 26
3.4.2 Stress based factor of safety .............................................................................. 26
3.4.3. Multiaxial safety factor ....................................................................................... 26
3.5. Multiaxial Safety Factor .............................................................................................. 27
3.5.1. The Definition of High-Cycle Multiaxial Fatigue............................................... 27
3.5.2. High-Cycle Multiaxial Fatigue Theories ............................................................ 27
3.5.3. McDiarmid Criterion............................................................................................ 28
3.5.4. Dang Van Criterion.............................................................................................. 30
3.6. Theoretical background to Spot Weld analysis ....................................................... 34
3.6.1. Introduction ......................................................................................................... 34
3.6.2. The fatigue analysis of spot welds - general description ............................... 35
3.6.3. Structural stress calculation.............................................................................. 35
3.6.4. Material properties .............................................................................................. 38
3.6.5. Damage Calculation............................................................................................ 40
3.6.6. A note on the subject of modelling spotwelds................................................. 41
3.6.7. References........................................................................................................... 42

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 2
3.7 FE-based fatigue analysis of welded structures ....................................................... 44
3.7.1 Introduction .......................................................................................................... 44
3.7.2 Background .......................................................................................................... 44
3.7.3 BS7608 and other simple S-N methods ............................................................. 46
3.7.4 The “Volvo” approach for thin-sheet structures............................................... 53
3.7.5 References............................................................................................................ 62
3.8 Multiple mean stress curve analysis .......................................................................... 63
3.8.1 Interpolating life from a set of multiple mean stress life curves ..................... 63
3.9 Vibration fatigue ........................................................................................................... 66
3.9.1 Introduction .......................................................................................................... 66
3.9.2 Review of S-N analysis in the time domain ....................................................... 67
3.9.3 Simple fatigue analysis using time history recreation ..................................... 69
3.9.4 Fast fatigue analysis methods in the frequency domain ................................. 71
3.9.5 Comparison between fatigue analysis techniques ........................................... 76
3.9.6 FE based vibration analysis in the frequency domain ..................................... 77
3.9.7 Conclusions.......................................................................................................... 82
3.9.8 References............................................................................................................ 82

4. Time Histories........................................................................................... 83
4.1 Why time histories are required.................................................................................. 83
4.2 Constant amplitude versus variable amplitude......................................................... 83
4.3 Obtaining and creating variable amplitude time histories ....................................... 85
4.4 Multiple load cases ...................................................................................................... 86
4.5 Using nSoft time history processing to increase calculation speed ...................... 86
4.6 Transient or time-step analysis .................................................................................. 89

5. Materials Data ........................................................................................... 90

5.1 Materials database ....................................................................................................... 90
5.2 S-N data......................................................................................................................... 90
5.3 E-N data......................................................................................................................... 91
5.4 Multiaxial data............................................................................................................... 91
5.5 Auto-generated data .................................................................................................... 92
5.6. Surface finish correction ............................................................................................ 94
5.7. Fatigue strength reduction factor.............................................................................. 95

6. Principles of Linear Superposition ......................................................... 96

7. Calculation of fatigue life from rainflow matrices ................................. 98

7.1. Strain-life Fatigue Analysis from a Rainflow Matrix................................................. 98
7.2. Stress-life Fatigue Analysis from a Rainflow Matrix.............................................. 104

8. Auto elimination ..................................................................................... 105

9. Virtual strain gauges .............................................................................. 107

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 3

1. Introduction to fatigue analysis from finite element models

FE-Fatigue performs fatigue analysis from FE results. CAE environments include, for example,
HyperMesh, which interfaces to FE-Fatigue by exporting stress/strain information in nCode's FES
file format. This is comparable to exporting data for other solvers, such as NASTRAN.
FE-Fatigue provides stress-life and strain-life analysis, the theoretical basis of which is described
in this document.
FE-Fatigue is part of the nSoft suite of products and a basic nSoft-E license provides the
functionality for data display and manipulation. FATIMAS products are not essential to use FE-
Fatigue however they provide additional useful functionality. Time histories of stress/strain (DAC
files) can be exported from FE-Fatigue for use in FATIMAS modules.
An FE-Fatigue installation comprises:
• fatfe (FE-Fatigue solver)
• fatres (FE-Fatigue results listing and sorting)
• mdm (Materials Database)
• fatduty (Duty Cycle Analyser)
• pvxmul (Multiple Channel Peak Valley Slicing)
• remdac, dacrem (RPC translators)
• sdrc2fes (.UNV to .FES conversion)
• fe2fes (Generic FE results to .FES conversion)
• Additional files include utilities and example files.
• On-line documentation is also available electronically in Adobe Acrobat PDF format along
with a printed user guide of worked examples. The documentation includes documents
which detail the internal structure of FES and DAC files.

The following pages show the process of analysing FE data with FE-Fatigue – the important files
are as follows:

• Partial FES file – created by the translator

• Full FES file – file required by nCode analysis package
• Results file – file in a format suitable for direct import into the appropriate
pre and post processor.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 4

Figure 1 FE Fatigue’s basic system diagram

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 5

Where the pre-and post-processor does not support the FES file, the following system diagram
applies:

Figure 2 System diagram with translation (if a translator is required it must be run first to create the
partial FES file).

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 6

2. Overview of new features in FE-Fatigue Rel. 5

2.1 Auto elimination by Group
The auto elimination feature of FATFE has been enhanced to allow the user to retain the top n
percent of results for each material group present in the model.
See Auto elimination for more details.

2.2 Virtual strain gauges

This feature allows the user to place a virtual strain gauge on the part being analysed. Element or
node strains that are calculated in the FE solver are exported such that the strain output will
emulate a strain gauge rosette at the same location on the actual part. This provides an improved
facility for measurement to CAE correlation and calibration to occur and improves the confidence
of FE models by a more direct comparison of strain results.
See Virtual strain gauges for more information.

2.3 Seam weld analysis

Release 5 supports seam weld analysis from finite element stresses. See Seam weld analysis for
more information.

2.4 High temperature fatigue analysis

Release 5 of FE-Fatigue supports the fatigue analysis of components at constant high
temperature.

2.5 Vibration fatigue analysis

Fatigue calculations may be done based on the results of a random vibration FE analysis, which is
in the form of Power Spectral Density (PSD) stress output. See Vibration fatigue analysis for more
details.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 7

3. Fatigue Theory

3.1. The S-N (High Cycle Fatigue) Approach

It has been recognised since 1830 that a metal subjected to a repetitive or fluctuating load will fail
at a stress level lower than that required to cause fracture on a single application of the load. The
nominal stress method was the first approach developed to try to understand this failure process
and is still widely used in applications where the applied stress is nominally within the elastic
range of the material and the number of cycles to failure is large. From this point of view, the
nominal stress approach, is best suited to that area of the fatigue process known as high cycle
fatigue. The nominal stress method does not work well in the low cycle region where the applied
strains have a significant plastic component. In this region a strain based methodology must be
used.
For more information, see the nCode book of Fatigue Theory.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 8

3.2. The Local Strain theory of fatigue

The nominal stress approach has been used extensively in the study of premature failures of
components subjected to fluctuating loads. Traditionally, the magnitude of the observed cyclic
stresses were observed to be less than the tensile elastic limit and the lives long, i.e. greater than
about 105 cycles. This pattern of behaviour has classically been referred to as high-cycle fatigue.
As duty cycles have became more severe and components more complicated, another pattern of
fatigue behaviour has emerged. In this regime, the cyclic loads are relatively large and have
significant amounts of plastic deformation associated with them together with relatively short lives.
This type of behaviour has been commonly referred to as low-cycle fatigue or more recently
strain-controlled fatigue. The transition from low-cycle to high-cycle fatigue behaviour generally
occurs in the range 104 to 105 cycles.
The analytical procedure evolved to deal with strain-controlled fatigue is called the strain-life, local
stress-strain or critical location approach.
For more information, see the nCode book of Fatigue Theory.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 9

3.3. Multiaxial strain-life

3.3.1 Introduction
The essential elements of any FE-based fatigue analysis are:
• Identification of the loading environment
• Establishment of the relationship between applied loads and local stresses and/or strains
• The relationship between local stresses/strains and fatigue damage
For multiaxial analysis, these three elements can be analysed at various levels of detail, and with
the increasing detail comes a cost of analysis time and interpretation of results. At any node or
element in an FE-Fatigue analysis, the local stress state can be assessed to determine the level
of multiaxiality. If we assume that fatigue damage initiates at the surface of a structure then any
direct or shear stresses normal to the surface are zero, and there are only two principal stresses
which may be non-zero.
Hence there are three basic cases:
• Case 1 Uniaxial stress state – there is one principal stress which is significantly larger
than the second for the whole of the load history and whose angle does not
change.
• Case 2 Proportional biaxial stress state – the ratio of the two principal stresses is non-
zero, but remains constant for the duration of loading. The angle remains
constant also.
• Case 3 Non-proportional stress state. Either the biaxiality ratio or the angle of the
maximum principal changes significantly through the time history.
For case 1, no special algorithm corrections need to be applied to convert elastic stresses and
strains to elastic-plastic. The Neuber correction, or similar methods, alone are sufficient.
For case 2, procedures should be used to take into account the fact that the loading is non-
uniaxial. Two such procedures are due to Hoffmann and Seeger [25] and Klann-Tipton-Cordes
[26]. These are described briefly in the following pages. They apply only to the local strain
approach.
For case 3, a full multiaxial notch correction procedure should be used. In the current release of
FE-Fatigue there is no multiaxial notch correction procedure available. Results from highly
multiaxial elements with high levels of strain should be treated with caution.
Damage models have been developed which can account for non-proportional and multiaxial
loading conditions. Examples are Fatemi-Socie, Bannantine and Wang-Brown. These are
implemented in FE-Fatigue. A simple critical plane approach is also included which can take into
account the mobility of the stress tensor – this is described on page 11. In conjunction with regular
uniaxial stress-strain predictions and damage models it can provide a life estimate in non-uniaxial
cases and can be applied in both strain-life and stress-life damage models.

3.3.2. Co-ordinate System Rotation

In order to do any multiaxial assessment or analysis, FE-Fatigue requires stress or strain data in
the plane of the surface. In particular, it requires that the X-Y axis pair defines the plane tangential
to the surface, with the Z axis normal to that plane. This can be done by the software creating the
partial FES file, in which case a header flag will be set to indicate that only 2D stresses are
present. However, most translators do not support this feature and therefore most data files will

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 10

contain stresses oriented in a co-ordinate system not compatible with the above requirement.
Since the FES file does not contain co-ordinate system information, a method is applied which can
derive co-ordinate transformation information from the stress data itself. This method relies upon
the integrity of the data and will not work if any of the assumptions are violated.
In order to derive a transformation matrix, it is assumed that the two largest principal stress
tensors lie in the plane of the surface and that the third principal is normal to the surface and has a
value of zero, or near zero. There are cases where this is not true, for example where an external
or internal pressure is applied or where the coarseness of the FE model leads to inaccuracies in
the extrapolation of results from element centroid or Gauss point location to surface nodes.
If we have a surface stress tensor of 6 components in a co-ordinate system X-Y-Z, we need to
define a system X’-Y’-Z’ where X’-Y’ defines the plane tangential to the surface and Z’ is normal to
the surface. To obtain principal stress values and directions, we can calculate the eigenvalues and
eigenvectors of the following matrix:
S = [ Sxx Sxy Sxz
Syx Syy Syz
Szx Szy Szz ]

where the cross-terms are identical ( e.g. Syx= Sxy ). The eigenvalues give the principal stresses
and the eigenvectors the corresponding direction cosines. The eigenvectors can be stored in a
3x3 matrix T, which is the transformation matrix from the original co-ordinate system X-Y-Z to the
co-ordinate system defined by the principals. If we re-order the columns of T to make the first
column correspond to the absolute maximum principal, the second column the second principal
and the third the principal closest to zero, then we have a transformation matrix which will
transform any stress tensor to a co-ordinate system X’-Y’-Z’ in which X’ is aligned with the
maximum principal stress. To transform a tensor in X-Y-Z to X’-Y’-Z’, calculate
S’ = TSTT

where TT is the transpose of T. If the third principal is not zero, then the results of the
transformation may not lie in the surface tangential plane and the software will issue a warning.

3.3.3. Assessing Multiaxiality

FE-Fatigue provides tools to assess whether uniaxial techniques are valid at any part of the
model, especially where stress levels are high. The biaxiality ratio, defined as the ratio of the
smaller absolute in-plane principal to the larger, can be calculated and plotted, either as a mean
and spread over the whole model or on a detailed, element by element basis.
The biaxiality ratio lies between –1 and +1, where zero indicates a uniaxial condition. The angle
calculated by FE-Fatigue describes the angle of the absolute maximum principal stress vector
with the X-axis of the planar co-ordinate system. In practice this value is in itself of little practical
use, but the spread of the angle for all significant stress tensors shows the mobility of the principal
vector. Because the value of the biaxiality ratio and the angle can change for each point in a time
history, global measures of mean biaxiality ratio, standard deviation of biaxiality ratio, dominant
angle and angle spread are calculated for the entire model and can be contour plotted. To remove
the effect of low stress values, points whose maximum principal stress is less than a specified
gate are excluded from the calculations.
The Node/Element option on the Utilities menu in FE-Fatigue can be used to look at the state of
stress at any particular node or element in detail. This allows plots of the time histories of stress,
strain, biaxiality ratio and angle as well as cross-plots of biaxiality ratio and angle vs. absolute

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 11

principal stress. The dominant value and the degree of scatter in the plot clearly show whether the
data is uniaxial, proportional or non-proportional.
This information can be used to determine the best choice of analysis route for a particular
component.

3.3.4. Critical Plane Analysis

When the stress tensor is mobile it becomes necessary to use a critical plane approach, i.e. one
which references specific planes rather than using combined stress or strain parameters such as
Von Mises or maximum principal strain. The critical plane should not be confused with the crack
plane. Cracks typically start in shear mode and after a transition period grow in opening mode.
The period we call crack initiation life may include growth in a number of planes. The local strain
approach does not model cracks explicitly but attempts to predict initiation life on the basis of bulk
stress and strain parameters. The critical plane is simply the plane on which the stress and strain
parameters are calculated. The method used to rotate the co-ordinate system of the stresses and
strains by means of a tensor rotation such that the x-y plane of the new co-ordinate system lies in
the critical plane.
For planes with an angle of intersection at 90 degrees, the equations for the strain and stress

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 12

normal to the plane and at a rotated angle φ (phi) is as follows:

2 2
ε n = ε x cos φ + ε y sin φ + γ xy cosφsinφ

2 2
σ n = σ x cos φ + σy sin φ + 2τ xy cosφsinφ

FE-Fatigue calculates the damage using the strain (E-N calculations) or stress (S-N calculations)
at 18 angles from 0 to 170 degrees inclusive. The largest damage is then used as the result. Only
the rainflow matrix for the worst case angle can be extracted, but time histories for all angles can
be exported for preview, or additional calculation in FATIMAS.

3.3.5. The Hoffmann-Seeger Method

This method applies when loading is proportional. If the loading is non-proportional then most of
the assumptions implicit in the method are violated; hardening is in reality kinematic and not
isotropic, and the principal axes not only rotate, but the principal stress and strain axes are no
longer necessarily aligned with each other. Hoffmann-Seeger is used in conjunction with Neuber
to correct the elastic stresses and strains to elastic-plastic. If loading is biaxial, Neuber alone is
insufficient as the cyclic stress-strain curve is from uniaxial data. Hoffmann and Seeger suggest a
method for extending use of the Neuber correction to multiaxial loading, subject to the following
assumptions:
• The principal stress and strain axes are fixed in orientation
• The ratio of the in-plane principal strains is constant
• The uniaxial stress-strain curve can be extended for use with suitable equivalent stress
and strain parameters such as the Von Mises parameters.
Hencky’s flow rules and Masing material memory behaviour are also assumed.
First the elastic values of the signed Von Mises stress and strain (from the FE analysis) are
computed:

σ eq
σ e1 2 2 and ε eq = -------
-
σ eq = ----------- σ e1 – σ e1 σ e2 + σ e2 E
σ e1

together with the strain biaxiality ratio e2/e1, which is assumed to be constant and equal to the
elastic value. Then the Neuber correction is made, by solving the equations:

1 2
---
σ σq n σ eq
ε q = -----q- + æè ------′ öø And σ q ε q = σ eq ε eq = ----------
E
E K

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 13

Then the principal strains and stresses can be calculated as follows:

σq
ν' = 1--- – æ 1--- – ν eö ---------
2 è2 ø Eε q

1 – ν' a
ε 1 = ε ----------------------------
q 2
1–a+a

′ 1+a ö
ε 3 = -ε 1 v æè ---------------
-
′ ø
1–va
1
σ 1 = σ q ------------------------------
2
1 – ae + ae

ε
----2- + ν'
σ ε1
a = -----2- = ------------------
-
σ1 ε2
1 + ν' -----
ε1

3.3.6. The Klann-Tipton-Cordes Method

Whilst the Hoffmann-Seeger method adjusts the stresses and strains cycle by cycle, the method
proposed by Klann, Tipton and Cordes makes a one-time modification of the cyclic stress strain
curve according to the biaxiality ratio and then performs a normal Neuber calculation using the
modified parameters.
This method is also known as the parameter modification method.
Essentially, the equation:

1
----
σ σ q n'
ε q = -----q- + æè ------öø
E K'

is used to calculate a series of corresponding values of σq and εq. For each pair of values, ν’ is
calculated from:
σq
ν' = 1--- – æè 1--- – ν eöø ---------
2 2 Eε q
and a from:
ε
----2- + ν'
ε1
a = ------------------
-
ε2
1 + ν' -----
ε1

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 14

where:

ε2 ae – νe
----- = -------------------
ε1 1 – ae νe

Pairs of values of ε1 and σ1 are then computed as follows:

1 – ν'a
ε 1 = ε q ------------------------------
2
1 – ae + a e

1
σ 1 = σ q ------------------------------
2
1 – ae + ae

These pairs of values represent a cyclic stress strain curve for a particular biaxiality ratio ae. A
new set of Ramberg Osgood parameters K” and n” is obtained by fitting the equation:
1
-----
σ σ 1 n''
ε 1 = -----1- + æè ------öø
E'' K''

to the calculated points. E” is calculated explicitly from:

E -
E'' = ----------------
1 – νe a

The modified parameters are used to carry out the Neuber correction in the normal way.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 15

3.3.7 Non-proportional multi-axial methods

Uniaxial methods for life prediction using the local strain approach have been in use for some time, hav-
ing their roots in the work of Basquin [8] Manson [9] and Coffin [10,11], incorporating rainflow cycle
counting and material memory [12] and Miners rule [13]. Such methods are available in FATIMAS [14]
and FE-Fatigue. Within the well known limitations of these methods they work quite well for a variety of
components where the local loading in the critical area is uniaxial or near-uniaxial. This class of com-
ponents includes many that are subject to complex multiaxial loading environments [5,6]. However,
there are many other components where a combination of loads and geometric effects generates local
loadings which are proportional or non-proportional multiaxial.
The life prediction process from measured strains, or from elastic strain inputs, can be divided into
two steps. The first step is to determine the relationship between the strains and all the stress and
strain components required for the damage calculation, through application of a cyclic plasticity
model. For elastic inputs this also involves a notch correction procedure. The second step is to
carry out cycle and damage accumulation. FATIMAS [14] addresses problems where the strains
can be measured with a rosette, i.e. biaxial loading on a free surface. FE-Fatigue uses the same
model to calculate lives from elastic-plastic FE strain tensors resolved to the surface of the model.
It uses a notch correction procedure when the strains or stresses are linear elastic.
For these problems the process can be summarised by the flow chart in:

3 strain histories from cyclic

an elastic-plastic FE plasticity
run in the plane of modelling
the surface OR

3 strain histories cyclic cycle

plasticity 3 stress component counting
from a strain LIFE
modelling histories and 4 strains and damage
gauge rosette OR accumulation

3 elastic strain
histories from cyclic plasticity
linear modelling +
notch correction
superposition

Figure 3 Outline of strain-based life prediction process.

The essential calculations made by the software are as follows:
1. It takes three components of strain εx, εy and εxy, either directly from FE or via linear
superposition.
2. It feeds these 3 strain channels into the Mròz-Garud cyclic plasticity model, the outputs of
which are the remaining non-zero strain component εz and the in-plane stresses σx, σy and σxy.
The model includes a notch correction procedure when strains are elastic.
3. It processes the resulting 7 components of stress and strain either by the conventional critical
plane methods, or by multiaxial rainflow counting and then accumulating damage using the

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 16

Wang-Brown methods.
These calculations are described in more detail in the following sections.

3.3.8 Multiaxial notch correction procedure

Transient elastic-plastic FE analyses can be prohibitively expensive for realistic FE models and
input loading histories. Multiaxial notch correction procedure, which utilizes results from simple
static linear-elastic FE analyses and the stress-superposition principle, is necessary. Inputs to the
procedure are “pseudo” stresses (from the linear-elastic computations), from which the elastic-
plastic strains and related stresses are estimated using a Neuber-type analysis, i.e. the overall
strain energy density equivalence between linear-elastic and elastic plastic stress-strain states,
figure 4. The cyclic plasticity model, necessary for the notch correction procedure, is the Mroz-
Garud model. The procedure is confined to the free surface conditions, i.e. plane stress
assumption are utilized, figure 5.

Figure 4 Overall strain energy density equivalence.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 17

Figure 5 Coordinate system definition.

A powerful feature of the implemented notch correction procedure is that a set of additional
conditions, necessary for the complete formulation of a multiaxial stress state problem, is based
on ratios of strain energy density increments contributed by each pair of corresponding stress and
strain increments (“proportional work path”), Equation 1. This assumption allows for significantly
reduced dependency of the results on geometry and constraint conditions at the notch tip. Other
often used conditions, like ratios of principal strain or stress increments, can make the results
strongly dependent on the notch constraint conditions, and the user is forced to make arbitrary
choice between them. Uncontrolled ratcheting, sometimes observed with the Mroz-Garud cyclic
plasticity model, is dramatically reduced or disappears altogether with the implemented multiaxial
notch correction routine, since the solution process is essentially “energy bound”.

e e e e a a a
S 22 ∆e22 + e22 ∆S 22 = S 22 ∆e22 + e22 ∆S 22a
e e e e
S 23 ∆e23 + e23 ∆S 23 = S 23a ∆e23
a a
+ e23 ∆S 23a
S 33e ∆e33
e e
+ e33 ∆S 33e = S 33a ∆e33
a a
+ e33 ∆S 33a (1)

The predictions of elastic-plastic strains and related stresses from linear-elastic inputs compare
very favorably with transient elastic-plastic FEA results for a wide range of loads, from
proportional to significantly non-proportional.

3.3.9 Cyclic Plasticity Modelling

The stresses and strains required by the damage models can be calculated if the relation between
the equivalent plastic strain increment Deeq and the equivalent stress increment ∆σ eq is known
during the application of a given load increment. However, it is known that the current relation
depends on the previous load path and therefore the plasticity model must deal with loading path
dependent material constitutive behaviour.
Several models are available in the literature [1,2,16,17] of which the model proposed by Mròz [1]
and recently modified by Garud [2] are the most popular. Mròz [1] has proposed that the uniaxial
stress-strain material curve be represented by a set of plasticity surfaces in three dimensional
stress space. In the case of a two dimensional stress state, the plasticity surfaces reduce to
ellipses on the plane of principal stresses described by:

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 18

2 2
σ eq = σ1 – σ1 ⋅ σ2 + σ2 (1)

and illustrated in figure 4.

σ2

2 f
3
1 f
2
0 f
1

ε σ1

Figure 6 Linearisation of the material σ-e curve and corresponding plasticity surfaces.
The load path dependent memory effects are modelled by prescribing a translation rule for the
ellipses moving with respect to each other over distances given by the stress increments. It is
also assumed that the ellipses move inside each other and they do not intersect. If the ellipses
come in contact with one another they move together as a rigid body.
The translation rule proposed by Garud [2] avoids the intersection of the ellipses that could occur
in some cases in the original Mròz [1] model. The Garud translation rule is illustrated in and can
be described by a model consisting, for simplicity, of only two plastic surfaces (ellipses).

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 19

σ2

dσ

B2 1

2 4 B1 f2
3

O1
5
O1

σ1
f’1
f1

Figure 7 Geometrical interpretation of the Mròz-Garud incremental plasticity model.

In order to predict material response due to the stress increment ds, the following steps are made:
1. Extend the stress increment dσ to intersect the first external non-active plastic surface f2 at
point B2
2. Connect point B2 and the centre of the intersected plastic surface f2
3. Find point B1 on the active plastic surface f1 by drawing a line parallel to the line O2B2 through
the centre O1 of the surface f1
4. Connect the conjugate points B1 and B2 by the line B1B2
5. Translate the ellipse f1 in the direction of B1B2 from point O1 until the end of the vector dσ lands
on the moving ellipse f1
The translation rule assures that the two ellipses are tangential with the common point B1B2
without intersecting each other. Two or more tangential ellipses translate as a rigid body and the
largest moving ellipse (figure 7) indicates the proper constitutive relation (linear segment) to be
used for a given stress increment. The principal of the method is described here for two surfaces.
The FATIMAS module MLF and FE-Fatigue use 40, with the corresponding linear sequences
being fitted to the Ramberg-Osgood equation to give a more or less smooth curve. More details of

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 20

this method can be found in reference 3.

3.3.10 Multiaxial Rainflow Counting

Wang and Brown [18] proposed a multiaxial cycle counting method on the basis of strain
hardening behaviour under non-proportional variable amplitude loading. Relative stresses and
strains were introduced so that a pair of turning points define the start and end points of a reversal,
where the equivalent relative strain rises monotonically to a peak value. Since plastic deformation
generates the driving force for small fatigue cracks, hysteresis hardening provides a physical
parameter for cycle counting, analogous to rainflow counting in the uniaxial case.
Each reversal commences with elastic unloading, which is followed by reloading and plastic strain
hardening up to the next turning point. The most significant turning point occurs at the highest
value of equivalent strain. This is illustrated at time 0 in figure 8, which shows a repeating block of
a combined tension/torsion non-proportional load history. The equivalent strain is defined as the
von Mises strain.
1

0.5
STRAIN (%)

epsilon

0 gamma
equivalent
-0.5

-1
0 20 40 60 80

TIME (secs)

Figure 8 A variable amplitude non-proportional strain history, showing applied tensile (epsilon) and tor-
sional (gamma) strains with the absolute equivalent strain.
The cycle counting method is illustrated by the following example. Starting from the most
significant turning point, a graph is drawn for the loading block of relative equivalent strain, where
relative strain
* A
ε = ε ij – ε
ij ij

represents the change of strain since time A; figure 9 shows the relative equivalent strain, with
respect to times 0, 10 and 20 seconds. Using the relative strain, a reversal can be defined starting
from 0, up to the maximum value 10 seconds. To obtain the second reversal the relative strain is
re-plotted starting from the next turning point where unloading commences (at 10 seconds), and
the portions of the strain hardening curve for the reversal are selected by a traditional rainflow
procedure [3]. The region of unloading and reloading within that reversal is counted in the next
step.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 21

1.5
1
STRAIN (%)

0.5 epsilon

0 gamma
relative - 0
-0.5
relative - 10
-1
relative - 20
-1.5
0 20 40 60 80

TIME (secs)

Figure 9 The variable amplitude history, showing relative equivalent strains plotted with respect to times 0,
10 and 20 seconds respectively.
Using the next turning point, relative strain is re-plotted with respect to 20 seconds for the
subsequent continuous fragment of strain history, yielding the third reversal in figure 9. This
procedure is repeated for each turning point in chronological order, until every fragment of strain
history has been counted. The recursive nature of this process makes this module significantly
slower than other analysis modules such as CLF and uniaxial methods in FE-Fatigue.

3.3.11 Fatigue Damage Calculation

The counting method described above is independent of fatigue damage parameters, being
based on hysteresis deformation behaviour. Being unrelated to material properties, it can be inte-
grated with any multiaxial fatigue damage model. If the counted reversals are non-proportional, a
fatigue damage parameter that accounts for non-proportional straining effects is required. The
path-independent damage parameter proposed by Wang and Brown [19] has been shown to pro-
vide good correlation for several materials under proportional and non-proportional loading,
′
σ – 2σ n,mean
γ max + S ⋅ δε n f b ′ c
ε̂ ≡ ----------------------------------------
′ ′
- = ------------------------------ ( 2N f ) + ε ( 2Nf ) (2)
1 + v + S(1 – v ) E f

where γmax is the maximum shear strain amplitude on a critical plane (proportional or non-
proportional), δεn is the normal strain excursion between the two turning points of the maximum
shear strain (that is the range of normal strain experienced on the maximum shear plane over the
interval from start to end of the reversal), and σn,mean is the mean stress normal to the maximum
shear plane. This can be omitted if no mean stress correlation is required. If mean stress is
included, this is equivalent to the morrow method. The term S is a material constant determined
from a multiaxial test (typically between 1 and 2 for Case A and around 0 for Case B) and v‘ is the
effective Poisson’s ratio. The right hand side of the equation is the same as the uniaxial strain life
equation, with a Morrow mean stress correction [20]. Mean stress is measured as the average of
the maximum and minimum stress values over the reversal. The total damage induced by a
loading history is calculated using Miner’s rule [13].
The plot in figure 10 is a plot of predicted life against experimental life for a variety of proportional
and non-proportional tests on laboratory specimens, from Reference [4].

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 22

10,000

Predicted lifetime
(block)

1000 non proportional

proportional
ideal
factor 2
100
100 1,000 10,000

measured lifetime (blocks)

Figure 10 Comparison of experimental and predicted results for the Wang-Brown method.

3.3.12 Critical Plane Methods

The other multiaxial damage parameters considered are the more conventional critical plane
parameters. In these methods, stresses and strains are resolved onto a particular plane, inclined
at an angle θ = 90 degrees (Case A) and/or θ = 45 degrees (Case B) to the free surface. Cycle
counting (uniaxial) and damage parameter calculation is carried out on the critical plane and the
damage accumulated. The orientation φ of the projection of the normal to the damage plane is
increased by 10 degree increments from 0 to 170 degrees. The plane with the largest
accumulated damage is said to be the critical plane. The models are:

1. Normal strain (θ = 90 degrees only):

∆ε n σ′ f b c
--------- = ------ ( 2Nf ) + ε′ f ( 2N f ) (3)
2 E

is the strain amplitude normal to the critical plane. Otherwise this is the usual Coffin-Manson-
∆ε
Basquin equation, where --------n- .
2

2. Shear strain:

∆γ ( 1 + V e )σ′ f b c
------ = --------------------------- ( 2N f ) + ( 1 + V p )ε′ f ( 2N f ) (4)
2 E

3. Smith-Watson-Topper/Bannantine (θ = 90 degrees only) [21]:

2
∆ε n σ′ (5)
--------- ⋅ σ n, max = -------f- ( 2N f )2b + σ′ f ⋅ ε′ f ( 2Nf ) b + c
2 E

where σn, max is the maximum normal strain on the critical plane which occurs during each
rainflow cycle. Otherwise this is the Smith-Topper-Watson method [22].

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 23

4. Fatemi-Socie [23]:

∆γ æ σ n, maxö ( 1 + V e )σ′ b
n ( 1 + V e )σ′ 2b c n ( 1 + v p )ε′ f σ′ f b+c
- = ---------------------------f ( 2Nf ) + ------------------------------f ( 2Nf ) + ( 1 + V p )ε′ ( 2N f ) + -----------------------------------
------ 1 + n ---------------- - ( 2N f ) (6)
2è σy ø E 2Eσ y f 2σ y

3.3.13 Types of Cracks

Figure 11 Types of Cracks

In multiaxial fatigue conditions cracks can grow in different directions into the surface, depending
on the biaxiality ratio, as shown in Figure 9. For negative biaxiality ratios, e.g. torsion loading
when a = -1, cracks tend to grow at 90 degrees into the surface and tend to be shallow and long.
These are named type A cracks. For positive biaxiality ratios, e.g. biaxial tension when a > 0,
cracks tend to grow at 45 degrees into the surface and tend to be deep and short. These are
named type B cracks. In uniaxial loading conditions, when a = 0, cracks can grow either way, i.e.
they can be both type A and B.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 24

3.3.14 Critical Plane Rainflow Counting

Figure 12 Uniaxial Rainflow Counting

The method of identifying cycles in variable amplitude multiaxial fatigue conditions is crucial for
obtaining correct life predictions. Recommended procedure for critical plane methods is to rain-
flow count the primary damage parameter, e.g. shear strain parallel to the crack plane, and "track"
the secondary damage parameter, e.g. component of stress perpendicular to the crack plane.
Common rainflow counting algorithms are not able to do this correctly. For example, common rain-
flow counters would identify cycle E-F in the above diagram with associated maximum stress at
point F. In multiaxial fatigue conditions, however, the maximum stress during the cycle would not
necessarily occur at point F, but anywhere from E to E'. Thus, while identifying strain cycle E-F, the
stress has to be "tracked" between points E-F-E' in order to properly determine its maximum
value. A related issue is worth noting: while peak-valley editing (extracting peaks and valleys
from a time history) is perfectly valid in uniaxial case, it should not be performed in multiaxial
cases, because important information about stress magnitudes can be lost.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 25

3.3.15 References
(1) MRÒZ Z., (1967), On the Description of Anisotropic Work Hardening, Journal of Mechanics and
Physics of Solids, vol. 15, pp. 163-175
(2) GARUD Y. S., (1981), A New Approach to the Evaluation of Fatigue under Multiaxial Loading, Journal of Engi-
neering Materials and Technology, vol. 103, pp. 118-125
(3) WANG C. H. and BROWN M. W., (1996), Life Prediction Techniques for Variable Amplitude Multiaxial Fatigue -
Part 1: Theories, Journal of Engineering Materials and Technology, vol. 118, pp. 367-370
(4) WANG C. H. and BROWN M. W., (1996), Life Prediction Techniques for Variable Amplitude Multiaxial Fatigue -
Part 2: Comparison with Experimental Results, Journal of Engineering Materials and Technology, vol. 118, pp.
371-374
(5) HEYES P. J., MILSTED M. G. and DAKIN J., (1996), Multiaxial Fatigue Assessment of Automotive Chassis
Components on the basis of Finite-Element Models, Multiaxial Fatigue and Design, ESIS 21 (Edited by A.
Pineau, G. Cailletaud and T. C. Lindley) MEP London, pp. 461-475
(6) HEYES P., DAKIN J. and ST.JOHN C., (1995), The Assessment and Use of Linear Static FE Stress Analyses for
Durability Calculations, Proceedings of the Ninth International Conference on Vehicle Structural Mechanics and
CAE, pp. 189-199
(7) HEYES P. and FERMÉR M., (1996), A Program for the Fatigue Analysis of Automotive Spot-Welds Based on
Finite Element Calculations, Proceedings of the Symposium on International Automotive Technology, SAE
Technical Paper 962507
(8) BASQUIN O. H., (1910), The Exponential Law of Endurance Tests, Proceedings of the American Society for Test-
ing Materials, vol. 10, pp. 625-630
(9) MANSON S. S., (1953), Behaviour of Materials under Conditions of Thermal Stress, Heat Transfer Symposium,
University of Michigan Engineering Research Institute, pp. 9-75
(10) COFFIN L. F., (1954), The Problem of Thermal Stress Fatigue in Austenitic Steels at Elevated Temperatures,
ASTM STP No. 165, p.31
(11) COFFIN L. F., (1954), A Study of the Effects of Cyclic Thermal Stresses on a Ductile Metal, Trans. American
Society for Testing and Materials, vol. 76, pp. 931-950
(12) MATSUISHI M. and ENDO T., (1968), Fatigue of Metals Subjected to Varying Stress, Presented to Kyushu Dis-
trict Meeting, JSME.
(13) MINER M. A., (1945), Cumulative Damage in Fatigue, Journal of Applied Mechanics, vol. 12, pp. A159-A164
(14) nCode International Ltd., (1997), nSoft-E FATIMAS software manual
(15) GLINKA G. and BUCZYNSKI A., (1997), Elastic-Plastic Stress-Strain Analysis of Notches under Non-Propor-
tional Cyclic Loading Paths, Proceedings of the 5th International Conference on Biaxial/Multiaxial Fatigue and
Fracture, Krakow, Poland
(16) CHU C. C., (1989), A Three-Dimensional Model of Anisotropic Hardening in Metals and its Application to the
Analysis of Sheet Metal Formability, Journal of Mechanics and Physics of Solids, vol. 22, no. 3, pp. 197-212
(17) ARMSTRONG P. J. and FREDERIC C. O., (1966), A Mathematical Representation of the Multiaxial Bausch-
inger Effect, CEGB Report RD/B/M731, Berkeley Nuclear Laboratories
(18) WANG C. H. and BROWN M. W., (1993), Inelastic Deformation and Fatigue under Complex Loading, Proceed-
ings of the 12th International Conference on Structural Mechanics in Reactor Technology, vol. L, pp. 159-170
(19) WANG C. H. and BROWN M. W., (1993), A Path-Independent Parameter for Fatigue under Proportional and
Non-Proportional Loading, Fatigue and Fracture of Engineering Materials and Structures, vol. 16, pp. 1285-1298
(20) BROWN M. W., SUKER D. K. and WANG C. H., (1996), An Analysis of Mean Stress in Multiaxial Random
Fatigue, Fatigue and Fracture of Engineering Materials and Structures, vol. 19, no. 2/3, pp. 323-333
(21) BANNANTINE J. A., (1989), A Variable Amplitude Multiaxial Fatigue Life Prediction Method, Ph.D. thesis,
University of Illinois at Urbana-Champaign
(22) SMITH K. N., WATSON P. and TOPPER T. H., (1970), A Stress-Strain function for the Fatigue of Metals, Jour-
nal of Materials, vol. 5, no. 4, pp. 767-778
(23) FATEMI A. and SOCIE D. F., (1988), A Critical Plane Approach to Multiaxial Fatigue Damage Including Out-of-
Phase Loading, Fatigue and Fracture of Engineering Materials and Structures, vol. 11, no. 3, pp.149-165
(24) DREßLER K., KÖTTGEN V. B. and KÖTZLE H., (1995), Tools for Fatigue Evaluation of Non-Proportional
Loading, Proceedings of Fatigue Design 1995 (Edited by Gary Marquis and Jussi Solin), vol. 1, pp. 261-277
(25) HOFFMANN, M. and SEEGER, T. “Estimating multiaxial elastic-plastic notch stresses and strains in combined
loading.” Biaxial and Multiaxial Fatigue, EGF3 (Edited by M.W.Brown and K.J.Miller), 1989, Mechanical Engi-
neering Publications, London, pp 3-24.
(26) KLANN, D.A., TIPTON, S.M., CORDES, T.S., (1993) “Notch stress and strain estimation considering multi-
axial constraint.” SAE technical Paper 930401

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 26

3.4. Safety factor

Safety factor calculations are designed to provide a single scalar quantity which is used to
determine a level of confidence that a component will achieve a given life. The life is often an
endurance limit, but may also be a finite life, or a level of stress above the endurance limit which
can be tolerated by the structure.
In FE-Fatigue, there are three ways in which a safety factor can be calculated, life based, stress
based and multiaxial.
3.4.1 Life based factor of safety
In this method, applicable for strain-life and stress-life, a target life is specified. The program then
performs iterative calculations, adjusting the linear scaling factor on stress until the life is within a
tolerance factor of the specified life. User preferences determine the maximum factor to be
attempted and the tolerance for ending the iteration process.
3.4.2 Stress based factor of safety
The stress-based safety factor method calculates a safety factor relative to a reference stress
which will typically be the fatigue limit of the material, taking into account the mean stress. It
considers only the largest stress cycle developed during the loading history. The definition of
safety factor using the Goodman correction is illustrated below.

Safety factor using Goodman

Reference Stress
Alternating Stress

A
A’
σa

σm UTS
Mean stress

Only the largest stress cycle is considered, with amplitude σa and mean σm. The safety factor
calculated in FE-Fatigue is defined as A/σa. Factors < 1 are considered safe and >1 unsafe. An
alternative definition is A'/σa which will give more pessimistic predictions in the "safe" regime. FE-
Fatigue does not use this method. The reasoning behind this is that the mean stress and
alternating stress often have different origins. The method used gives a safety factor on
alternating stress for a given mean stress.
3.4.3. Multiaxial safety factor
FE-Fatigue includes options to apply the Dang Van and McDiarmid methods to create a multiaxial
safety factor plot. These methods are discussed separately.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 27

3.5. Multiaxial Safety Factor

3.5.1. The Definition of High-Cycle Multiaxial Fatigue
Multiaxial fatigue is a term referring to the fatigue of components under complex stress states.
High-cycle multiaxial fatigue deals with components that experience large number of cycles and
are therefore designed for infinite life. Such components include crank-shafts, propeller shafts,
and rear axles. In these components, it is not necessary to quantify the amount of fatigue dam-
age, but just to consider if any fatigue damage will occur during the load history.
For high-cycle multiaxial fatigue, actual quantities of fatigue damage are not considered. Instead,
the loading history is examined to determine only if damage occurs at any moment. If damage
does occur, then the component does not have infinite life. The high-cycle criteria considered can
also be used to calculate a safety factor of a component if infinite life is determined.
In cases of high-cycle multiaxial fatigue, the complex stress states in the loading history are
important to consider in order to accurately predict fatigue life. A simple uniaxial fatigue method is
not sufficient and can produce non-conservative results.

3.5.2. High-Cycle Multiaxial Fatigue Theories

Several methods have been developed to calculate a fatigue limit in the high-cycle fatigue (HCF)
regime with significant multiaxial loadings. Diboine (3) classifies these methods according to the
basis of calculation:
• Empirical
• Von Mises
• Tresca
• Microstructural
The empirically-based fatigue criteria require testing of a specific material in a loading
configuration similar to that of interest in order to produce a numerical fit to the data and make
further predictions. This defeats the purpose of using fatigue predictions, and therefore this type
of method will not be considered further.
Von Mises-based criteria were prevalent early in the studies of high-cycle fatigue, the Sines and
Crossland criteria being quite well-known. However, Von Mises is a deformation criterion, which
does not consider the plane of maximum shear stress range. Fatigue cracks have been observed
to initiate and grow initially along these planes of maximum shear stress.
The above observations lead to the more intuitive Tresca-based high-cycle fatigue criteria.
Findley reported early (1959) on Tresca-based predictions, but McDiarmid’s Tresca-Based
criterion introduced in 1987 appears to be widely accepted and well-known.
Microstructural approaches to high-cycle fatigue have also become more common in recent
years. These approaches consider the stabilised, microscopic stresses in calculations of fatigue
damage. Dang Van first introduced this type of approach in 1973, but his modified approach
proposed in 1987 is easier to use and provides better correlation.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 28

3.5.3. McDiarmid Criterion

Description

The McDiarmid method is a critical plane based fatigue limit criterion which is applicable to cases
where crack initiation is at free surfaces. The critical plane is defined as the plane of maximum
shear stress range. The fatigue criterion is based on two loading parameters, the shear stress
amplitude (half of the maximum range) on the critical plane and the maximum normal stress (for
the entire loading history) on the plane of maximum shear stress range. The McDiarmid method
considers Case A and Case B cracking separately in calculating the fatigue damage. Case A
cracking when cracks initiate by shear along the surface, e.g. when there is a torsional loading.
Case B cracking occurs when the maximum shear plane is inclined at 45 degrees to the free sur-
face with shear cracks being driven into the surface.

The fatigue limit according to McDiarmid can be expressed in the following form:
τa / tA,B + σn, max / 2σT = 1 (1)
where:
τa= shear stress amplitude
tA,B= reversed shear fatigue strengths (amplitude) for Case A and B crack growth
σn, max= maximum normal stress for the entire loading history on the
plane of maximum shear stress amplitude
σT = tensile strength

If the value on the left hand side of this equation exceeds 1, damage is predicted to occur. This
fatigue limit can be displayed graphically in a plot of
τa / tA,B vs. σn, max / 2σT (see below).

Figure 13
Any loading that goes above the line is considered to cause damage and the fatigue life will not be
infinite. As with many calculation methods, one problem is to obtain suitable materials properties.
In this case we require the fully reversed shear fatigue strength for Case A and Case B cracking.
According to the McDiarmid theory, the Case A value should simply be the shear fatigue stress
amplitude at the fatigue limit for a pure torsion case. The case B value would require the fatigue
limit to be determined under conditions that would generate Case B cracking.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 29

In the absence of this information, the values of tA,B may be estimated from the uniaxial fatigue
limit σL as follows. In the case of uniaxial loading, Case A or B cracking may occur, so we can re-
write the McDiarmid criterion at the fatigue limit under uniaxial loading conditions as:
(0.5*σL)/(tA,B ) + (0.5*σL)/(2*σT) = 1
So the necessary case A and B fatigue limits are given by:
tA,B = σL/ (2 - (σL/2*σT))
In practice, both the case A and case B planes are searched to find the plane with the maximum
value of shear stress amplitude. Case A planes intersect the surface at 90 degrees and case B at
45 degrees. The software requires that the stresses are resolved to the plane of the surface such
that the X-Y plane is the plane of the surface. Rather than searching infinite planes, the orientation
of the case A and B planes with respect to the X axis is varied in 10 degree increments. The
stresses at these angles (angle indicated as φ) are calculated as follows:
Case A Cracking (θ =90 degrees):

σn = σxcos2φ + σysin2φ + 2τxycosφsinφ (2)

σxz = 2σxsinφcosφ + 2σysinφcosφ + τxy (-sin2φ + cos2φ) (3)

Case B Cracking (θ = 45 Degrees):

σn = 0.5σx cos2φ + 0.5σysin2φ + τxycosφsinφ (4)

τyz = -σxcos2φ - σysin2φ - τxycosφsinφ (5)

Because the fatigue limit criterion is expressed as

τa / tA,B + σn, max / 2σT = 1 (6)
a safety factor can be defined as:
Safety factor = 1 / ((τa / tA,B)+ (σn, max / 2σT)) (7)
which is reported in the results file for contour plotting.
The maximum shear range for each plane is calculated by the difference of the maximum and
minimum values of shear stress for the loading history. The McDiarmid criterion calculations are
then performed for the planes of maximum shear stress range.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 30

3.5.4. Dang Van Criterion

Description
The Dang Van Criterion is a multiaxial fatigue limit criterion which can be classified as a
‘microstructural’ method because it is based on the concepts of stress and crack initiation at a
microscopic level.
Essentially, the Dang Van method assumes that around the fatigue limit, cyclic plasticity will occur
on a microscopic scale due to the inhomogeneity of the material on the scale of individual grains.
This very localised cyclic plasticity will lead to relaxation of mean stresses on the microscopic
scale, or another way of putting this is the development of microscopic residual stresses. Once
this stabilised state (elastic shakedown) has occurred, the Dang Van criterion (1,2) states that
crack initiation will occur whenever a function f of the microscopic stress, is greater than or equal
to zero at any time in a stabilised cycle.
f{σ(P,t)} ≥ 0 for P∈V(M) (8)
where:
M = a point of interest/stress concentration on a component (notch, fillet)
V(M)= an elementary representative volume around M
P= a point in V(M) where a grain(s) is critically oriented
σ(P,t)= a microscopic stress tensor at location P at time t

Dang Van proposes that f is a function of the microscopic shear stress (which can cause plasticity)
and the hydrostatic tension (which can open microscopic cracks, whatever their orientation). The
function is a simple linear one:
f(σ) = τ + a*ph – b and f(σ) = τ + a*ph + b (9)
where:
τ = microscopic shear stress at a given time
ph = microscopic hydrostatic tension at a given time
b = shear stress amplitude at the fatigue limit in pure torsion
a = hydrostatic stress sensitivity

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 31

This function establishes the Dang Van Endurance Domain shown below:

Figure 14
A loading path Γ is illustrated in Figure 14 and damage is predicted to occur only if the loading
path crosses the lines delimiting the endurance domain. Note that if the microscopic shear stress
is replaced by the microscopic Tresca or maximum shear stress, only the upper part of the
diagram need be considered, and the part of the loading path below the abscissa is reflected
upwards ( Γ’ ).
Considering again equation (9), b is taken to be the shear stress amplitude at the fatigue limit
under conditions of pure torsion, and the parameter a is known as the hydrostatic stress
sensitivity. To determine the hydrostatic stress sensitivity factor, the fatigue limit must be known
under at least two loading conditions. For instance, if the fatigue limit is known under fully
reversed torsion, b, and bending, f, a can be estimated from
a = 3(b- f/2)/f (10)
If the fatigue limit is known under only one condition, some assumptions could be made. For
instance if you only had the bending fatigue limit f, you could assume that the Von Mises stress
was a reasonable criterion to apply which gives b=0.5774f and a=0.23.
All this is relatively simple. The complexity of the Dang Van method lies in the calculation of the
microscopic shear stress.
The macroscopic stress Eij can be divided into a hydrostatic part, ph and the
deviatoric part, Sij
Sij = Eij – ph.δij [where δij is the Kroneker delta]
The microscopic stress in a stabilised condition is calculated from:
σij = Eij + ρij*
where ρij* is the stabilised residual stress.
The microscopic deviatoric stress is calculated from:
dev σij = Sij + dev ρij*
Then the microscopic maximum shear stress is calculated from:
τmax = 1/2(Tresca(dev σij))
The microscopic residual stress is calculated using an iterative procedure which attempts to
simulate the process of elastic shakedown.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 32

Its aim is to find the centre of the nine-dimensional hypersphere which can encompass the loading
path. The centre of this hypersphere defines the microscopic residual stress. Practically this is
achieved by simulating a combination of isotropic and kinematic hardening as illustrated (in two
dimensions!) below:

Figure 15 2D representation of hypersphere

A small initial yield surface is defined, denoted C0. When the loading path contacts the yield
surface, the yield surface is allowed to move (kinematic hardening) and expand (isotropic
hardening) and the process continues, cycling through the loading history until there is no further
expansion or movement of the yield surface. The displacement of the centre of the yield surface
from the origin then defines the microscoopic residual stress. A hardening parameter controls the
rate of isotropic hardening. Using a larger value of this parameter causes the stabilised condition
to be approached more rapidly, but there may be overshoot leading to loss of accuracy. The
default value of this parameter is 0.05 which is a good compromise for most loadings, but a
smaller value will give greater accuracy for very short loading histories.
It is convenient when using the Dang Van criterion to define a safety factor which may be used to
compare the worst loading condition to the fatigue limit criterion.
The definition of safety factor used in FE-Fatigue is:
Safety factor = b/(Max(τmax) + a.ph)
except when the denominator is less than or equal to zero, when the life is predicted to be infinite
and the safety factor is effectively infinite. For the purposes of contour plotting, an upper limit for
safety factors is used in results files, with the default being 100, but which can be set as a
preference.
References
1] DANG VAN, K., CAILLETAUD, G., FLAVENOT, J.F., LE DOUARON, A., LIEURADE, H.P. (1989) Criterion
for High Cycle Fatigue Failure Under Multiaxial Loading, Biaxial and Multiaxial Fatigue, EGF 3, Mechanical
Engineering Publications, London, pp. 459-478.
2] DANG VAN, K., GRIVEAU, B., MESSAGE, O. (1989) On a New Multiaxial Fatigue Limit Criterion: Theory
and Application, Biaxial and Multiaxial Fatigue, EGF 3, Mechanical Engineering Publications, London, pp. 479-
496.
3] DIBOINE, A. (1996) Fatigue Assessment of Components under Complex Loadings, Multiaxial Fatigue and
Design, ESIS 21, Mechanical Engineering Publications, London, pp. 425-443.
4] HEYES, P., MILSTED, M., STJOHN, C. (1993) Application of Finite-Element Based Life Prediction to Chassis
Engineering, Report for Ford Motor Company.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 33
5] MCDIARMID, D.L. (1991) A General Criterion for High Cycle Multiaxial Fatigue Failure, Fatigue and Fracture
of Engineering Materials and Structures. Vol. 14, No. 4, pp. 429-453.
6] MCDIARMID, D.L. (1994) A Shear Stress Based Critical Plane Criterion of Multiaxial Fatigue Failure for Design
and Life Prediction, Fatigue and Fracture of Engineering Materials and Structures. Vol. 17, No. 12, pp. 1475-1484.
7] PAPADOPOULOS, I.V., DAVOLI, P., GORLA, C., FILIPPINI, M., and BERNASCONI, A. (1997) A
Comparative Study of Multiaxial High-Cycle Fatigue Criteria for Metals, International Journal of Fatigue. Vol. 19,
No 3, pp.219-235.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 34

3.6. Theoretical background to Spot Weld analysis

3.6.1. Introduction
A key feature of the modern durability design process is the use of computer-based finite element
methods to predict durability at an early stage in the design cycle. This process is driven by the
need for designs with low weight, low cost of manufacture, short development cycles, and good
durability.
Calculations based on fatigue life and realistic loading histories permit structures and components
to be optimised for durability without the need for the expensive and time-consuming testing of a
series of prototypes. Design analysis based on fatigue life calculations results in designs that are
less conservative (i.e. better optimised) than those based on traditional criteria such as maximum
load or stress for a series of standard load cases.
Resistance spot welds are very commonly used in the automotive industry in the fabrication of all
manner of components and structures, and the durability of such structures is very often controlled
by the strength of the spot welds. The cost of tooling up for a single weld spot as part of an
automated manufacturing process may be as much as $30,000, and this can more than double if
a weld spot has to be added during production to remedy a problem [1]. These costs may be
minimised if the life of spot welds can be predicted at an early stage in the design process, though
the reduction in development time and improvement in quality is likely to be more significant.
Smith and Cooper [2] addressed the problem of life prediction of shear spot welds using a fracture
mechanics approach. They noted that a spot weld could be "....considered to be a circular solid
surrounded by a deep circumferential crack, which when loaded in a combination of Mode I and
Mode II, would grow a branch crack in the direction of maximum local Mode I". They showed that
good predictions of life could be made on the basis of calculated crack growth rates, and used
their calculations to generate some simple design curves. The method was based on detailed
finite element modelling of simple spot-welded lap-joints loaded in shear. This method would need
further development in order to cover all the possible weld configurations used in automotive
structures and to deal with the variable amplitude out-of-phase loadings to which they are subject.
The results of such a development might be a simple design code for spot-welds along the lines of
the British Standard BS 7608 [3] with families of load-life curves for different classes of spot-weld;
in practical FE models of automotive structures detailed modelling of individual spot welds is not a
desirable option.
In fact, load is a rather poor parameter for correlating the fatigue strength of spot-welds under
different loading conditions. Radaj [4] and Sheppard [5,6] note that durability of spot welds of a
variety of configurations and loadings can be better understood through numerical analysis of the
local stresses at the weld spot edge on the inside of the plate - the structural stresses around the
weld.
Rupp, Störzel and Grubisic [1] describe the calculation of these structural stresses, and also carry
out fatigue life predictions based on maximum and minimum stresses and a load spectrum. The
method used in FE-Fatigue is more or less identical to that described by Rupp et al [1], and
combines their method for structural stress calculation with the usual techniques for calculation of
dynamic stresses. It is therefore possible to make spotweld fatigue calculations in FE-Fatigue
based on static, transient and modal (using modal superposition approach) structural analysis
results [7].

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 35

3.6.2. The fatigue analysis of spot welds - general description

The method described below has been typically applied using bar elements to represent spot
welds joining two sheets of shell elements. The method can be used with other element types and
FE analysis codes, but this discussion will primarily focus on using CBAR elements with
NASTRAN [8]. The forces transmitted through these CBAR elements are used to calculate the
structural (nominal or "hot-spot") stresses in the weld "nugget" and the adjoining sheet metal at
intervals around the perimeter of the nugget. These stresses can then be used to make fatigue life
predictions on the spot weld using a S-N (total life) method.
In the FE model, the spot-welds should be represented by stiff CBAR elements joining 2 sheets of
shell elements, and perpendicular to both. The CBAR elements should be sufficiently stiff that the
stiffness of the joint is not sensitive to the CBAR properties. In practice this can readily be
achieved by giving the CBAR element the same modulus of elasticity as steel, and dimensions of
the cross section approximately equal to the weld nugget it represents. The shells are positioned
at the mid-planes of the sheet metal and the length of the CBAR element and the separation of the
shells should therefore be half the sum of the sheet thicknesses. There is no need for any
refinement of the mesh around the spot-welds. The only requirement for the shell elements used
to model the sheets is that they transmit the correct loads to the bar elements. In fact it seems that
best results (i.e. most realistic joint stiffnesses) are achieved when the dimensions of the shell
elements are quite large - more than twice the diameter of the weld nuggets.
The steps in the fatigue calculation are now described in more detail.

3.6.3. Structural stress calculation

Figure 16 Typical spot weld

The shaded part is the spot weld "nugget". In a finite element analysis, the weld is modelled as a
stiff bar element joining the mid-planes of 2 sheets of thin shell elements.The length of the bar
element is 0.5(s1+s2) where s1 and s2 are the thicknesses of sheets 1 and 2 respectively. Point 3
is on the axis of the weld nugget and at the interface of the 2 sheets, i.e. 0.5 s1 from Point 1. All
forces and moments are taken to be in the co-ordinate system illustrated in Figure 16. This is a
Cartesian system with the Z axis going from Point 1 to Point 2. This is different both from the
arrangement used by Rupp et al [1] and that used in NASTRAN for CBAR elements, but a little
simpler. The relationship between the FE-Fatigue and NASTRAN CBAR coordinate systems is
illustrated in Figure 17.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 36

Figure 17
In order to account for the local CBAR coordinate system in NASTRAN, the following
transformation is required:
Point 1
FE-FatigueNASTRAN
FX1= -Z Force (plane 2 shear)
FY1= Y Force (plane 1 shear)
FZ1=X Force (axial)
MX1= -Z Moment 1 (plane 1 bend A)
MY1= -Y Moment 1 (plane 2 bend A)
MZ1= X Moment (torque)
Point 2
FE-FatigueNASTRAN
FX2= Z Force (plane 2 shear)
FY2= -Y Force (plane 1 shear)
FZ2= -X Force (axial)
MX2= Z Moment 2 (plane 1 bend B)
MY2= Y Moment 2 (plane 2 bend B)
MZ2= -X Moment (torque)
Point 3 Nugget values
(using above FE-Fatigue values)
FX3 = FX1
FY3 = FY1
FZ3 = FZ1
MX3 = (MX1*t2 - MX2*t1) / (t1 + t2)
MY3 = (MY1*t2 - MY2*t1) / (t1 + t2)
MX3 = (MZ1*t2 - MZ2*t1) / (t1 + t2)
Where t1 is the shell thickness at point 1 and t2 is the shell thick at point 2
The CBAR element forces and moments Fx,y,z and Mx,y,z are written to the .FES file in FE-Fatigue
co-ordinate system, for each of the three specified points. These forces and moments (except Mz)
are used to calculate nominal stresses (structural stresses) on the inner surface of sheet 1 and

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 37

sheet 2, and in the weld nugget at the interface of the two sheets, at intervals around the
circumference of the spot weld (by default, θ = 0o to 360o by increments of 10o). The forces and
moments at points 1 and 2 are those applied by the spot welds on the sheets, and the forces and
moments at point 3 will be those applied by the upper section (between point 3 and point 2) on the
lower section (between point 1 and point 3) This statement defines the directions and signs of the
forces and moments.
The stresses are calculated as follows:
For point 1 the equivalent stress on the inner surface of the sheet as a function of angle θ around
the circumference of the spot weld is:
σ v1 = – σ max ( F x1 ) cos θ – σ max ( F y1 ) sin θ + σ max ( F z1 ) + σ max ( M x1 ) sin θ – σ max ( M y1 ) cos θ (1)
where:
Fx1
σ max ( Fx1 ) =
πds1 ............(2)

Fy1
σ max ( Fy1 ) =
πds1 ............(3)

æ 1.744F z1ö
σ ( F z1 ) = K 1 ç ---------------------
-÷ forF z1 >0 .........(4)
è s12 ø

σ( Fz1 ) = 0 for Fz1 ≤ 0 .............(5)

so that only the tensile component of the axial force in the nugget contributes to damage, and:

æ 1.872M x1ö
σ max ( M x1 ) = K 1 ç ----------------------
-÷ ............(6)
è ds 1 2 ø

æ 1.872M y1ö
σ max ( M y1 ) = K 1 ç ----------------------
-÷ ............(7)
è ds 1 1 ø

These equations are based on analytical expressions for stress, modified with empirical factors
taking into account the observed fatigue behaviour from a variety of spotwelded components and
loading conditions. Note that K1=0.6√s1 (compensating for the size/stress-gradient effect in
bending) and d is the diameter of the weld nugget, all dimensions being in mm. Forces are in N
and moments in Nmm. The equations for point 2 are similar.
The stresses in the nugget are treated differently. From the forces calculated for point 3, nominal
stresses are calculated at intervals around the circumference of the weld nugget, say at 10 degree
intervals. The method of Rupp et al then suggests that the direct stress be calculated on multiple
planes at 10 degree intervals, i.e. using a stress-based critical plane method. This would mean 36
x 18 = 648 calculations for each weld nugget. This is very computationally intensive, especially in
view of the fact that the normal failure mode is by crack growth through the sheet metal; spot
welds do not usually fail by cracking through the nugget unless the nugget diameter is too small
compared to the sheet thickness. For this reason, two faster options are included in the software:
to ignore the possibility of nugget failure altogether, and to use the absolute maximum principal
stress as the damage parameter, as commonly used in FE-Fatigue (only 36 calculations). This is

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 38

calculated as follows:

τ = τ max ( Fx 3 )sin 2 θ + τ max ( Fy 3 ) cos 2 θ .........(8)

σ = σ ( Fz 3 ) + σ max ( M x 3 ) sin θ − σ max ( M y 3 ) cos θ .........(9)

where:
16 Fx 3
τ max ( Fx 3 ) =
3 πd 2 ..........(10)
16 Fy 3
τ max ( Fy 3 ) =
3πd 2 ..........(11)

4 Fz 3
σ ( Fz 3 ) = when Fz 3 > 0
πd 2 ...........(12)
σ( Fz 3 ) = 0 when Fz 3 ≤ 0 ...........(13)

32 M x 3 32M
σ max ( M x 3 ) = σ (M y3-
) = -----------------
πd 3 ..........(14a) max y3 2 ........(14b)
πd

From the shear and direct stresses on the nugget, the in-plane principal stresses can be
calculated from:

2
σ σ 2
σ 1, 3 = --- ± ------ +τ ............(15)
2 2
The principal stress with the greatest magnitude is taken as the damage parameter.

3.6.4. Material properties

The system requires an S-N curve for each metal sheet and for the weld nugget at load ratio R=0,
plus a mean stress sensitivity factor and a standard error parameter. These S-N curves are
specifically spot-weld S-N curves and are quite distinct from the parent plate material S-N curves.
The formulation of the S-N curve is as normal, that is:

∆S = SRI 1( N f ) b1 ...........(16)

for number of cycles to failure Nf < Nc1 the transition life. For Nf > Nc1 a second slope b 2 is used.
The Stress Range Intercept SRI1, and slopes b1 and b2 are material constants. Each cycle with
amplitude S and mean stress Sm is corrected to calculate an equivalent stress amplitude S 0 at R=0
using the following equation:
S + MS m
S0 =
M + 1 ..........(17)

where M is the mean stress sensitivity (typically 0.1)

S-N curves for spot weld analysis must be obtained from tests on spot welded specimens. The
datasets included in the material database are taken from the work of Rupp et al [1] and represent
the results of tests on a variety of different specimen and loading configurations, with different
spotweld sizes and sheet diameters. Specimens that have been used for spotweld testing include

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 39

H-shear, H-peel, hat-profile (torsion), hat-profile (internal pressure), double cup, and so-on. One
such specimen – the H-shear specimen – so called because it has the profile of the letter H and
the spotwelds are loaded in shear, is illustrated in figure 18.

Figure 18 H Shear specimen

For more details of the testing and different specimen types, see reference [1]. The stress “S” in
the spot weld S-N curves is that defined in equations (1) to (15) above, and unless the specimen
and loading is extremely simple, its calculation requires that the specimen first be modelled in FE
to determine the spotweld cross-sectional forces and moments. To summarise, the procedure for
determining spotweld S-N curves is as follows:
Carry out constant amplitude fatigue tests with chosen specimen and loading configurations, with
a load ratio R=0
For each specimen and loading configuration, carry out finite element modelling and analysis to
determine the spotweld cross-sectional forces and moments.
Calculate the stresses using equations (1) to (15) – FE-Fatigue provides access to the structural
stress results, so in general, the easiest way to determine the structural stresses in the fatigue
tests is to simulate the test in FE-Fatigue.
Plot the data and carry out regression to determine the S-N curve parameters (equation (16)) as
normal.
In practice, the theory is not perfect, and S-N curves derived from specimens which load the spot
welds in a different way (e.g. H-shear and H-peel) may be slightly different in slope and intercept.
However the errors are not sufficiently large to be a problem in a design analysis environment.
The S-N curves in the database represent an average from a variety of different component types.
Rupp et al [1] noted that, for a range of automotive steels, there were actually modest differences
between the S-N curves and so the use of generic S-N curves for sheet steel and weld nuggets is
quite reasonable, especially when more precise S-N data, or even a material specification is not
available. Such S-N curves are provided in the material database. There is quite a wide scatter
band, which is partly a reflection of the fact that this data represents spot-welds in a variety of
steels, including mild and high strength. Better predictions may be possible if S-N data specific to
the materials being used is available.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 40

3.6.5. Damage Calculation

By default, damage calculations are carried out at 10 degree intervals around the spot weld in
both sheets and in the weld nugget (though the number of calculation points can be modified by
increasing the angle interval to 20 or 30 degrees and by omitting the weld nugget from the
calculation). By default, there are therefore 108 fatigue calculations per spot-weld. At each
calculation point the effective stress history is calculated either directly from the force and moment
results from a transient FE analysis, by scaling and superimposing the force and moment results
of a number of static load cases according to the quasi-static method [7], or using a modal
superposition method. The stress history at each calculation point is calculated from the force and
moment results using equations (1) to (15), then rainflow cycle counted to form a range-mean
histogram. Rainflow cycles are converted to equivalent stress amplitude for R=0 using equation
(17), then damage is calculated and summed using Miner's rule. Rupp et al [1] recommend that a
Miner’s sum of 0.5 be used when the loading is of variable amplitude, and 1 for constant amplitude
loadings. The shortest predicted life from all the calculation points predicts the life of the spotweld,
and also provides an indication as to which sheet the crack is likely to appear in, and its initial
direction.

Note: In FE-Fatigue the equivalent units can be used to simulate a Miner’s Sum value not equal
to 1.

The method for life prediction of spot-welds described here is somewhat computationally
intensive. Computation time is roughly proportional to the number of data points in the load
histories. In addition to reducing the number of calculation points, substantial reductions in
computation time can therefore be achieved by judicious filtering of the loading inputs, e.g. using
PVXMUL.
The method as originally developed by Rupp et al [1] only applied to spotwelds joining two sheets
together. In practice, some spotwelds are often created which join three sheets or even four
(though this is undesirable and likely to lead to poor welds). Further research was carried out to
extend the method to three sheet welds. The basic findings of this research were:
• Where three sheets are spotwelded together, it is unusual for the crack to grow in the
middle sheet, in fact, during tests on a variety of specimen shapes and loading
conditions, no cracks grew in the middle sheet.
• Where three sheets are spotwelded together, the structural stresses in the outer two
sheets, and the resulting fatigue lives can be reasonably predicted by treating the weld as
two two-sheet welds, and ignoring the possibility of failure at the middle sheet.
Therefore when modelling three sheet spotwelds for analysis in FE-Fatigue, the best approach is
to model them using two CBAR elements, as if they were separate spotwelds, and to check the
three-sheet correction option when running the software. This will detect the common node
shared by the two CBAR elements, ignore results calculated at this node, and report for both
CBAR elements the shortest life from either sheet.
The spotweld analysis option in FE-Fatigue uses the cross-sectional forces and moments in
CBAR elements to calculate the structural stresses around the spotweld. The structural stresses
are based on analytical expressions for stress, modified by empirical factors which adjust the
theoretical stress to take into account size and loading type effects. The factor K1=0.6√s in
equation (4) is such a factor. This approach can be generalised to include a stress factor applied
to components of the structural stress due to Fx,y, Mx,y and Fz, of the following general form:

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 41

SF = F x diametere1 x thicknesse2 ............(18)

Where F is a factor and e1 and e2 are the diameter and sheet thickness exponents. In general
there are therefore 9 parameters which describe the complete set of stress factors. Comparing
this with the original structural stress equations from Rupp et al, it can be seen that the
parameters for steel are as follows:

Component Factor F Diameter Exponent e1 Thickness Exponent e2

Fx,y SFFXY = 1.0 DEFXY = 0.0 TEFXY = 0.0

Mx,y SFMXY = 0.6 DEMXY = 0.0 TEMXY = 0.5

Fz SFFZ = 0.6 DEFZ = 0.0 TEFZ = 0.5

These are the default parameters used in the software for spotwelds in steel. If the material type
is an aluminium alloy, the size and thickness effects are somewhat different and the following set
of parameters are automatically selected.

Component Factor F Diameter Exponent e1 Thickness Exponent e2

Fx,y SFFXY = 0.4 DEFXY = 0.5 TEFXY = -0.25

Mx,y SFMXY = 0.4 DEMXY = 0.5 TEMXY = -0.25

Fz SFFZ = 1.0 DEFZ = 0.0 TEFZ = 1.0

It is also possible for the user to override these automatically selected values and define the
parameters. This is done by using ENM to set the environment keyword SPOTWPAR. The value
of the keyword should be the nine parameters, comma separated, i.e. SPOTWPAR = SFFXY,
DEFXY, TEFXY, SFMXY, DEMXY, TEMXY, SFFZ, DEFZ, TEFZ.
WARNING: This keyword should not be set unless the user has sufficient evidence to justify using
values different to those set as defaults in the software!

3.6.6. A note on the subject of modelling spotwelds

One criticism that has been levelled at the simple method of modelling spotwelds using CBAR
elements is that it tends to lead to a slight under-estimation of the global stiffness of spotwelded
structures. This lack of stiffness is due to the fact that each spotweld is modelled by CBAR
elements which are connected to the surrounding mesh at a single node. The bending stiffness of
this connection is a little too low compared with reality. The problem is minimised by using large
shell elements around the spotweld, and is more apparent as the sheet thickness increases
(leading to larger bending moments).
Another issue is that of the meshing requirement. The method requires that the CBAR elements
be perpendicular to the sheets of shells they join, and that they be of the correct length. This
requires that when meshing the flanges to be joined, nodes exist directly opposite each other at
the location of the spot weld. This is not always easy to achieve, especially if the parts to be joined
are meshed separately. Some FE pre-processors do however include automated spotweld

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 42

modelling capability to generate bar elements of the required type at appropriate locations in the
model.
For these reasons, a large number of different strategies for modelling spotwelds have been
explored, with the ultimate aim of achieving a strategy which will satisfy the requirements of
different groups (NVH, Durability, Crash) whilst allowing rapid meshing and easy positioning of
spotwelds. For NVH, the global stiffness and modal behaviour is most important, whereas for
durability, the ability to calculate reasonably accurate local stresses is paramount. Some of the
different methods that have been tried are listed below:

• Selective thickening of shell elements around the spotweld – this improves the stiffness of
the structure. If dynamic calculations are being carried out, negative mass elements have
to be added to compensate.
• Connecting the CBAR elements to the surround shells using a “spider” of rigid bar
elements to distribute the load – this can improve the stiffness but adds complexity.
• Using shell elements to represent each spotweld – this method adds 32 shell elements
for each spotweld, and so can effectively double the model size for a typical car body.
An alternative method, which is supported in FE-Fatigue, uses 8-noded CHEXA elements in
NASTRAN to connect the sheets of shell elements. The solid is connected to the sheets using
MPC equations (RBE3 elements) and congruent meshes on the two flanges are not required. The
MPCFORCES on the solid elements are used to calculate an equivalent set of forces and
moments at each end of the weld.
In summary, the simple CBAR method tends slightly to underestimate the global stiffness of a
spotwelded structure, but has proved its worth over some years as a practical and effective way of
modelling spotwelds for durability calculations. A number of other methods have been suggested
in order to improve global stiffness, ease modelling requirements and satisfy the requirements of
groups analysing other vehicle attributes. Support has been added for some of these and other
new methods are being reviewed.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 43

3.6.7. References
[1]Rupp, A., Störzel, K. and Grubisic, V. (1995) "Computer Aided Dimensioning of Spot-Welded
Automotive Structures". SAE Technical Paper 950711.
[2]Smith, R. A. and Cooper, J. F. (1988) "Theoretical predictions of the fatigue life of shear spot
welds." Fatigue of Welded Structures, Ed. S. J. Maddox, pp. 287 - 293, The Welding Institute.
[3]British Standards Institution. (1993) "Code of Practice for Fatigue Design and Assessment of
Steel Structures." BS 7608.
[4]Radaj, D. (1990) "Local Fatigue Strength Characteristic Values for Spot Welded Joints."
Engineering Fracture Mechanics, Vol. 37, No. 1, pp. 245 - 250.
[5]Sheppard, S. D. and Strange, M. E. (1992) "Fatigue Life Estimation in Resistance Spot Welds:
Initiation and Early Growth Phase." Fatigue and Fracture of Engineering Materials and Structures,
Vol. 15, No. 6, pp. 531 - 549.
[6]Sheppard, S. D. (1993) "Estimation of Fatigue Propagation Life in Resistance Spot Welds."
ASTM STP 1211, Advances in Fatigue Life Prediction Techniques, M. R. Mitchell and R. W.
Landgraf, Eds., pp. 169 - 185, ASTM Philadelphia.
[7]Heyes, P., Dakin, J. and StJohn, C. (1995) "The Assessment and Use of Linear Static FE
Stress Analyses for Durability Calculations." SAE Technical Paper 951101.
[8]MSC/NASTRAN Quick Reference Guide, Version 68, The MacNeal Schwendler Corporation,
Los Angeles, CA., 1994.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 44

3.7 FE-based fatigue analysis of welded structures

3.7.1 Introduction
Welding is used in many industries as an effective and economical method for making structural
joints between metal parts. However, the nature of the welding process means that welded joints
generally have a fatigue strength that is inferior to that of the parts being joined together. At the
same time, welds naturally tend to be made at geometric features or changes in section in the
structure. Even in a well-designed structure, it is often the welded joints that are most likely to fail
by fatigue, and a fatigue assessment of any such structure must place a high priority on the
evaluation of the durability of the welds.
This section provides a brief overview of:
• some of the issues concerning fatigue of welds
• some different methods
• implementation in FE-Fatigue

3.7.2 Background

WELD
WELD TOE

HAZ
PLATE
Figure 19 Weld cross section, showing fusion and heat affected zone (HAZ) and most likely failure location

The fatigue strength of welded joints, is in general much less than that of the parts which are
welded together, or of the “parent plate”. There are a number of reasons for this:
• Welding typically introduces geometric features, leading to a stress concentration. The
stress will typically be highest at the toe of the weld, and the shape in this area may not be
well controlled.
• The welding process will very often produce defects which can act as crack initiation sites
– slag inclusions, incomplete fusion, porosity etc.
• Around the fusion zone there is a heat affected zone (HAZ) where the parent material has
been heated to a high temperature and allowed to cool fairly rapidly. This may cause
major changes to the microstructure and properties in this region.
• Unless steps are taken to relieve them, welded structures will normally have residual
stresses, which may be of the order of the yield stress of the material.
All these factors cause the fatigue behaviour of a welded joint to be quite different from that of the
parent materials from which it is constructed. For this reason, it is not reasonable to model the
behaviour of the welded joint based on the parent plate properties, with modifications to take into
account the defects, residual stresses, notches, and material property changes, and expect to get
accurate predictions!

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 45

All the approaches described in this section are therefore based on characterisation of the fatigue
behaviour of the welded joints themselves, normally in the form of S-N curves. These curves
incorporate all the effects of defects, residual stresses, notches, and material property changes
which are introduced when the weld is made.

Consider the S-N curves shown below:

S-N Data Plot

classF
SRI1: 1.201E4 b1: -0.3333 b2: -0.2 E: 2.07E5 UTS: 500
BS4360-50D
SRI1: 1903 b1: -0.123 b2: 0 E: 1.914E5 UTS: 480

1E3

Str
ess
Ra
ng
e
(M
Pa)

1E2

1E1
1E3 1E4 1E5 1E6 1E7 1E8
Life (Cycles)

This figure compares the S-N curves for a typical structural steel (BS4360-50D) and one of the
weld classes from BS7608. Note that the fatigue strength of the weld is much less than that of the
parent plate, except at very short lives where the curves converge. Note also that the slope of the
weld S-N curve is much greater. This is typical of weld S-N curves and is indicative of a situation
where fatigue damage is dominated by crack growth. In fact many assessments of the structural
integrity of welds, particularly in large structures will use a crack growth approach, and life may be
well predicted by a linear elastic fracture mechanics (LEFM) approach.
There are a couple of other points to note about fatigue of welds.
• There may be large differences between the fatigue resistance of different grades of steel.
When welded however, these differences may become very small. This is because the
processes occurring in the fusion and heat affected zones completely change the
properties of the material, including negating the effect of any heat treatment.
• Fatigue in welds is comparatively insensitive to mean stresses. When you consider that
welds are often subject to large and unknown residual stresses, it is not surprising that the
effects of mean stress are often neglected when predicting the fatigue life of welds.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 46

3.7.3 BS7608 and other simple S-N methods

Introduction
One typical (and very useful) S-N based method for life prediction of welds is described in the
British Standard BS7608:1993 “Code of practice for fatigue design and assessment of steel
structures” [1]. This standard is very closely based on BS5400 Pt 10 “Code of practice for fatigue
design of steel concrete and composite bridges” [2]. From this name you can tell that the methods
were initially developed for application of civil engineering structures constructed from thick plates
and beams. However, it has been successfully applied to a variety of problems in the automotive
and other industries, on much thinner structures than envisaged by the writers of the standards,
with some surprisingly good results. A brief summary of the main points is given here, together
with some hints on application in FE-Fatigue. However, if the reader intends to use this method, it
is strongly recommended that he read the standard document first.
The standard is applicable to structures made from wrought structural steel with a yield strength of
less than 700 MPa, in air or seawater in the sub-creep regime (i.e. not at high temperatures). It
allows the fatigue assessment of plates, bolted and riveted joints, and welds. Here we will only
consider its application to welded joints.

The main steps

The main steps in a BS7608 fatigue analysis are as follows:
1. Choose a weld class for each joint, or part of a joint based on
a. joint geometry
b. the loading direction
c. the likely failure location
d. the type of weld (full penetration etc)
2. Determine the required stress history due to the loading
3. Calculate the life:
a. rainflow count the stress history
b. use S-N curve for the appropriate weld class
c. use Miner’s rule with small cycle correction to calculate damage and life
d. apply size effect correction (for plates > 16 mm only)
4. Assess the validity of the results

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 47

Weld classification
BS7608 includes a system for classifying welds as briefly described above. The figure below
shows some of the pictures and notes that are useful in classifying the welds. Because the
standard was designed essentially for civil engineering structures, some judgement and
imagination is called for when applying it to different situations such as automotive components,
where the configurations of welds may not closely match any of those illustrated in the standards,
and where the loading direction may not be simple. However, welds can normally be given a
reasonable classification, or considered to lie between two possibilities. Most welds tend to be
classified F, F2 or G.

Figure 20 Weld classification in BS7608

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 48

Determine the stress history

The location and nature of the stress that must be used to calculate the fatigue damage is
described in detail in the standard, and may vary for each case. However, usually:
• The stress required is the principal stress with the largest range. This corresponds in most
cases with the absolute maximum principal stress.
• The stress may be the nominal stress in a plate, but more often it is a “hot spot” or
structural stress, close to the weld toe.
The method of determining the structural stress is illustrated in Fig. 21.

Structural stress
at toe of weld
F Stress

t 0.4t

Figure 21 Determination of structural stress

The required stress is determined by extrapolating the stress in the plate to the weld toe, but
excluding stresses from locations closer than 0.4t to the weld toe, where t is the sheet thickness.
It is also allowable to determine the stress by experimental means. In that case a resistance strain
gauge of 3 to 5 mm in length should be applied to the plate, with its edge 0.1t from the weld toe.
See Figure 20.

Figure 22 Experimental determination of weld toe stresses

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 49

Clearly, in a design analysis environment, the first method is rather cumbersome, especially when
many welds must be considered in a rather simply meshed structure, and the experimental
method is not useful. A rather simpler strategy is required. The essence of the stress
determination is that we need to find the structural stress at the weld toe, but neglecting the
detailed effect of the stress concentration. One simple approach that gives satisfactory results for
welded sheet structures is to make a mesh from thin shell elements, where the elements
represent the mid-planes of the sheet metal, and the weld is not modelled in detail. The mesh
should be arranged so that it has a row of elements with their centroids at the location of the weld
toe as illustrated in Figure 22.

F Stress

t 0.4t

Element centroids
positioned at weld
toe

Figure 23 Simple mesh to support BS7608 calculation

So long as the mesh is of good quality, the stresses from these elements should give a good
approximation to the weld toe structural stress.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 50

Calculate the life

Figure 24 BS7608 S-N curve for Class F2 welds

Life predictions in BS7608 use rainflow counting of the largest principal stress (absolute maximum
principal) and Miner’s rule, in the normal way. No mean stress correction is used. The other
feature of BS7608 is the way that it considers the effect of small cycles.
In BS7608 welded analysis, S-N curves such as that illustrated above are used (the dashed lines
represent the +/- 2 standard deviation curves). At a life of 107 cycles there is a change in slope,
from -1/3 to -1/5. The stress level at this point represents the fatigue limit, and if all cycles are
below this stress level, no damage should be predicted, i.e. the second slope is changed from –1/
5 to 0. However, if one or more cycles is above this stress level, the second slope of –1/5 is used
for the whole calculation.
If you want to make a BS7608 calculation using FE-Fatigue, you can simply use the S-N curves as
defined in the database. However, in this case you will not be considering the case where all
cycles lie below the fatigue limit, and your calculation may be somewhat conservative.
For a more rigorous application of the standard, open the material database manager and copy
your chosen weld class data to a new dataset and edit the data to set the second slope b2 to zero.
Then, when carrying out the fatigue calculation in FE-Fatigue, on the Advanced Options form,
select the Small Cycle (Haibach) correction.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 51

Assess the results

Some typical fatigue analysis results are illustrated in Figure 24. Note that:
• results are only calculated for the strip of elements representing the weld toe.
• there was some doubt about the correct classification of the weld, so results were
calculated for Class G and Class F2.
• standard S-N curves were used with no further consideration of material variation
• the component was made of thin sheets no more than 3 mm thick.
• there is good correlation between analysis and test

Figure 25 Typical analysis results for an automotive suspension component subjected to a constant ampli-
tude fatigue sign-off test.

In the example illustrated above, the loading was quite simple. Where there is multi-axis loading, a
multiaxial assessment should be carried out. In particular, there is a requirement in BS7608 that
the principal stress directions should not vary by more than 45 degrees. If this limit is exceeded,
the analyst will have to exercise some judgement as to the most appropriate method.

Aluminium alloys
There is a very similar procedure available for use with welded aluminium structures. This is
described in BS8118 Part 1: 1991 – “Structural use of aluminium. Part 1:Code of practice for
design” [3] which has a short section on fatigue. This section is very closely based on the
methods described in BS5400 Pt 10. In fact the S-N curves look very similar except that the
fatigue strength is reduced by a factor of 3.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 52

Other simple approaches for welds

Of course, the general S-N based procedure described above may be used even if the analyst
does not want to use the weld classification and S-N curves found in the British or other
standards; there is nothing to prevent the use of user defined S-N curves.
There are a number of other approaches which might also be applied in FE-Fatigue. One such
method developed by a group of the JSAE is summarised in the diagram below.

Figure 26 JSAE method

This method is based on a parameter called εa,0.3t which is simply the strain amplitude determined
at a distance of 0.3t from the weld toe, which, with some modifications to take into account the
effect of thickness and flank angle variations, can be used to correlate fatigue lives from a variety
of different weld configurations.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 53

3.7.4 The “Volvo” approach for thin-sheet structures

Introduction
Approaches such as the BS7608 method are not universally popular for automotive applications,
for the following reasons:
• They were mainly developed for thick sheet structures, whereas the majority of automotive
welds join sheets of thickness 3mm or less.
• Weld classification systems designed for civil engineering structures such as bridges can
be difficult to apply to many automotive structures where many different weld geometries
may be seen.
• They do not always lend themselves to use in conjunction with a fairly unrefined finite
element model such as is typically used for automotive body analysis.
These criticisms were the motivation behind a new method developed at Chalmers University in
Gothenburg, at the behest of the Volvo Car Corporation. The method as originally developed and
described in Reference [5,6] was based on similar concepts to the LBF spot weld method. In this
original concept, structural stresses at the weld toe were calculated based on the nodal forces
acting on the weld toe elements. The method has been refined and modified to use node-at-
element stresses from the weld toe, and some practical examples of the use of the refined method
are described in Reference [7].
The method implemented in FE-Fatigue is based on that originally proposed by Volvo Car
Corporation and Chalmers University of Technology in Gothenburg and developed in co-operation
with nCode. This method has been developed specifically for automotive components welded
from thin sheets (1-3 mm in thickness). It can be summarised briefly as follows:
1. The method was originally developed for welds modelled and analysed in NASTRAN.
However, it should also work with any FE code, provided node-at-element stress results are
available.
2. There are some specific rules for meshing the welds, but essentially the structure must be
meshed predominantly with CQUAD4 elements (or equivalent) representing the mid-planes
of the metal sheets, and the weld bead represented by a single row of shells with thickness
corresponding to the thickness of the weld throat. The nodes of these elements should lie at
the weld toe. The mesh around the weld should be regular with elements of around 5 mm in
size, and triangular elements must be avoided. See Figure 26.
3. Unaveraged node-at-element stresses are recovered for the nodes at the weld toes, based
on the elements adjacent to the weld (not the weld elements themselves) and used to make
the fatigue calculations. The worst result for each node is reported to the results file.
4. The method determines the contribution of bending to the total stress and from this
determines whether the weld is essentially “stiff” or “flexible”. Each of these cases requires a
different S-N curve.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 54

Figure 27 Seam weld modelling of a simple T-joint

The method is now described in more detail.

Meshing and FE analysis rules

In order to get good results, some simple meshing rules need to be observed.
• The sheets are modelled by thin shell elements located at the mid-planes of the sheets.
The element thickness should be equal to the plate thickness
• A row of nodes should be positioned (at the mid planes) adjacent to the each weld toe.
(See figure 27)
• The weld bead should be modelled using 4 or 3 noded shell elements (CQUAD4 and
CTRIA3 if using NASTRAN). The thickness of these elements should be equal to that of
the effective weld throat.
• The elements representing the part of the plate immediately adjacent to the weld toe
MUST be 4 noded shell elements (CQUAD4). These are the elements from which the
stress results are recovered for use in the fatigue calculation, so it is particularly important
that these are the correct type and of good shape. The element length should be
approximately 5 mm.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 55

Figure 28 Basic meshing practice for seam welds.

There are a number of additional requirements for weld corners, and weld starts that should be
observed. These are detailed in the following figures for different joint geometries.

Figure 29 a) Overview of continuously welded T-joint. b) Detail of corner weld.

Note the well-shaped elements at the weld toe and the way the corner is radiused.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 56

Figure 30 a) Overview of intermittently welded T-joint. b) Detail of corner showing shape of weld toe
elements. c) Detail of corner – note the angles used to radius the weld end.

Figure 31 Modelling of an edge to edge thin walled section. a) Overview of joint. The weld is modelled with
4-node and 3-node shell elements. Note that the modelling of the weld still is according to the guidelines.
b) Side view of weld corner (see direction marked A). Note that the weld elements in the edge to edge
connection are modelled in the same plane as the elements representing the section sides.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 57

Figure 32 Intermittently welded overlap joint. a) Overview of joint. b) and c) Detail of weld end. Note that
the weld end is modelled in the same was as for the intermittently welded T-joint except that the triangular
element at the end of the weld bead should have its free edge normal to the sheets.

FE Stresses
When running the analysis in NASTRAN, the parameter SNORM should be set e.g.
PARAM,SNORM,55.0 and stresses recovered to the output2 file using the STRESS(CUBIC)
option. A simple NASTRAN file header might be as follows:
SOL 101
CEND
$
TITLE = static analysis
SPC=1
SUBCASE 1
SUBTITLE = Simple loadcase 100 N
LOAD = 2
STRESS(CUBIC,PLOT,PRINT)=all
DISPLACEMENT(PLOT)=all
BEGIN BULK
PARAM,POST,-1
PARAM,AUTOSPC,YES
PARAM,EPZERO,1.E-5
PARAM,PRGPST,NO
PARAM,K6ROT,1.
PARAM,SNORM,55.
$
$ GRID Data
$
GRID 1 0.0 0.0 0.0
The SNORM parameter defines a common surface normal at each node, and the CUBIC stress
recovery option uses a strain gauge approach with cubic bending correction to recover the nodal
and element centroid stresses.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 58

The fatigue analysis method as formulated in this version of FE-Fatigue has been developed and
validated to predict fatigue failure of welds from the weld toe, and this prediction should be based
on stresses recovered from the weld toe elements at the nodes along the weld toe.
The stresses recovered from the FE results are the unaveraged, absolute maximum principal
stresses for each static loadcase, mode or timestep, at the location of the weld toe. The results
are node-at-element results from the elements in the sheets adjacent to the weld toe only – not
from the weld elements themselves. At each node along the weld toe therefore, more than one set
of stress results will generally be recovered. Results are recovered from top and bottom surfaces.
FE-Fatigue will make separate fatigue analyses based on these unaveraged stresses and report
the worst (shortest life or greatest damage) result. Note that, unlike a normal FE-Fatigue analysis,
the stress results stored in the .FES file are the top and bottom surface absolute maximum
principal stress values; the full stress tensor is not recovered.
The user should also be aware that in this version, the results translator FE2FES is unable to
distinguish between the elements in front of the weld toe, which we should be using to calculate
the life, and those “underneath” the weld toe, which we should not be using and may lead to more
conservative predictions. The user can avoid this by controlling the set of elements for which
stresses are recovered to exclude the unwanted elements.

Figure 33 Results recovery

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 59

S-N Curves and Fatigue analysis method

The structural stress at the weld toe may be used to correlate stress-life data from a variety of
welded specimen configurations and loading directions. The results illustrated below are taken
from Reference 6. Loadings are constant amplitude, and all data is corrected to R = -1 using the
Haig diagram (constant life diagram) in figure 34.

Figure 34 S-N data for a variety of different weld geometries.

The most distinctive feature of this plot is that the data correlates quite well giving rise to two
distinct S-N curves. Closer inspection of the results revealed that the the stresses at the weld toe
for the test results on the upper curve were characterised by a predominance of bending, whereas
those falling on the lower curve had mainly tension/compression stress. Another way of looking at
this is to say that the results correlated to the upper curve were for joints that were “flexible” in
character and those correlated to the lower curve were “stiff”.
Now the stresses used in the fatigue calculation are the absolute maximum principal stresses
from the top and bottom surface of the weld toe elements, at the weld toe nodes. Let us call these
σtop and σbtm.
Now the stress at the weld toe can be divided into a bending component:

σ b = 0.5(σ top − σ btm )

and a direct component:

σ n = 0.5(σ top + σ btm )

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 60

We can now define a flexibility ratio r

σb
r=
σb + σn

This can be used to distinguish between the “flexible” and “stiff” cases, by setting a threshold
value of the flex ratio r, which may be in the range 0 ≤ r ≤ 1. The default value for this threshold in
the software is 0.5.
In order to account for mean stresses, a Haig diagram is used, as illustrated below.

Figure 35 Haig (constant life) diagram for welded joints.

In this diagram, two mean stress sensitivity factors are defined, M1 for R ≤ 0 and M2 for R ≤ 0. For
steels, typical values for M1 and M2 are 0.25 and 0.1 respectively. The value of M1 may be defined
in the material database, and the user must define the ratio M1/M2 to be used during the analysis.
This ratio will therefore normally be expected to have a value of 2.5 for steels.
Clearly, S-N curves for use with this method should be determined by testing welded steel
specimens, preferably at load ratio R = -1. Ideally, a number of different geometries would be
tested. In order to obtain the structural stress values required to plot the results on a S-N diagram,
it is necessary also to make FE models of the specimens used, following the guidelines described
above. Failure is deemed to have occurred when there is a visible crack.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 61

The analysis process as implemented in FE-Fatigue is summarised in the following diagram:

STATIC FE MODAL DIRECT

ANALYSIS TRANSIENT TRANSIENT
FE ANALYSIS FE ANALYSIS

Top and bottom Top and bottom surface

surface stresses for stresses for each
each static load case contributing mode
(largest principal) (largest principal)
Modal
Load-time participation
histories Linear Linear
factors
superposition superposition

Time histories of
top and bottom
surface stresses
σtop (t) and σbtm(t)

Determine minimum value of Rainflow count top

flexibility ratio r. If rmin < and bottom surface
threshold, use “stiff” curve, stresses and
otherwise use “flexible” curve calculate lives

Report worst
(shortest life) result
at each node

Figure 36 Analysis process summary for Volvo seam weld approach.

Figure 35 illustrates the fatigue analysis process using the “Volvo” method as implemented in
FE-Fatigue. Note that the analysis may be based on a static superposition, modal superposition or
transient approach. Note that, in this version, when the static or modal superposition approach is
used, the stresses that are superposed are the absolute maximum principal stresses for each of
the static modes or loadcases, and not the full stress tensors.
There are some indications that it might be wise to use a Miner’s sum of less than 1 to calculate
the life when using this method with variable amplitude loadings. However, more research is
needed to clarify this point.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 62

3.7.5 References
1. British Standard BS7608:1993 “Code of practice for fatigue design and assessment of steel
structures”. British Standards Institution.
2. BS5400 Pt 10:1980 “Code of practice for fatigue design of steel concrete and composite
bridges”. British Standards Institution
3. BS8118:1991 “Structural use of aluminium. Part 1:Code of practice for design”. British
Standards Institution
4. T. Kasahara, Y. Maruo, T. Nakamaru, S. Magara and K. Koibuchi. “Method of Fatigue Life Esti-
mation for Arc-Welded Structures”. SAE Technical Paper 2001-01-0781.
5. M. Andréasson and B. Frodin. “Fatigue Life Prediction of MAG-Welded Thin Sheet Structures
– Theory and Experiments”. MSc Thesis. Chalmers University of Technology and Volvo Car
Corporation. Nov 1997.
6. M. Fermér, M. Andréasson and B. Frodin. “Fatigue Life Prediction of MAG-Welded Thin Sheet
Structures”. SAE Technical Paper 982311. Proc. IBEC ’98, 1998.
7. M. Fermér and H. Svensson. “Industrial Experiences of FE-Based Fatigue Life Predictions of
Welded Automotive Structures”. Fatigue and Fracture of Engineering Materials and Structures,
Vol. 24, No. 7, 2001.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 63

3.8 Multiple mean stress curve analysis

3.8.1 Interpolating life from a set of multiple mean stress life curves
FE-Fatigue supports the input of and calculation from materials data in the form of multiple S-N
curves, where each curve represents the stress amplitude and life for a given mean stress or
R-ratio. An example is shown below:

Figure 37 Multiple S-N curves

In order to use these curves instead of a mathematical mean stress correction such as Goodman,
interpolation must be used.
The cycle consists of an amplitude and a mean value, which may or may not lie on the curve. If
the amplitude/mean pair lies inside the set of curves, for example in the case above, an amplitude
of 400 MPa and a mean of 50 MPa, then interpolation can be used using the four curve values
surrounding the cycle value ( in this case, 2 values on the curve mean=25 and 2 values on the
curve mean=75. The program first calculates the value on the adjacent curves at the required
amplitude using a log-log interpolation. It than uses the mean value to interpolate between the
curves.
In the example shown, the values on the adjacent curves are as follows:
Mean curve at 25 MPa, stress = 435 MPa, life = 239768 cycles.
Mean curve at 25 MPa, stress = 365 MPa, life = 617931 cycles.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 64

Mean curve at 75 MPa, stress = 435 MPa, life = 126302 cycles.

Mean curve at 75 MPa, stress = 365 MPa, life = 361833 cycles.
The program first calculates the life at 400 MPa (L) on each of the curves.
At 25 MPa:
log10(L) = log10(239768) - ((log10(435)-log10(400)) x (log10(239768)-log10(617931) / (log10(435)-log10(365)))
L25 = 377016 cycles
At 75 MPa:
log10(L) = log10(126392) - ((log10(435)-log10(400)) x (log10(126302)-log10(361833) / (log10(435)-log10(365)))
L75 = 208905 cycles
The life for a mean of 50 MPa is then interpolated as follows:
log10(L50) = log10(377016) + ((log10(208905) - log10(377016)) x (50-25) / (75-25))
L50 = 280643 cycles
Note that the mean stress is not logged, as the value can be zero or negative.
In the plot below an interpolated curve at 50 MPa has been added:

Figure 38 Includes interpolated curve at 50 MPa

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 65

If the data lies outside of the nest of curves, extrapolation is used only when the mean is higher
than the highest mean curve. For example, if the cycle has an amplitude of 400 and a mean of
zero, the program gives the value on the lowest curve as a conservative result, i.e. 377016 cycles.
If the mean is greater than 225 MPa, ( e.g. 250 MPa) the same equations are used with the two
nearest curves and this extrapolates the value, i.e.
At 175 MPa:
log10(L) = log10(48127) - ((log10(435)-log10(400)) x (log10(48127)-log10(109419) / (log10(435)-log10(365)))
L175 = 71274 cycles
At 225 MPa:
log10(L) = log10(30497) - ((log10(435)-log10(400)) x (log10(30497)-log10(66306) / (log10(435)-log10(365)))
L225 = 44210 cycles
The life for a mean of 250 MPa is then extrapolated as follows:
log10(L250) = log10(71274) + ((log10(44210) - log10(71274)) x (250-175) / (225-175))
L250 = 34819 cycles
NOTE: this extrapolation may also be "turned off" by using the environment keyword MEANCEIL
(set to "ON"), which will then use the non-conservative value of the highest mean stress curve in
the set.
Extrapolation will also be performed if the amplitude value is less than the lowest amplitudes
described in the set of curves, for example in this case if the amplitude was 100 MPa and the
mean 50 MPa, the program would extrapolate values for 100 MPa on the curves at 25 and 75 and
then interpolate between as normal. There is an upper limit of 1E30 on the value of life that can be
extrapolated in this way.
For R-ratio curves, the data is converted to the equivalent mean values so that the same
interpolation rules can apply.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 66

3.9 Vibration fatigue

Fatigue damage is traditionally determined from time signals of loading, usually in the form of
stress or strain. This approach is satisfactory for periodic loading but requires very large time
records to accurately describe random loading processes. These may prove prohibitive for many
finite element analyses especially when modelling dynamic resonance. Alternatively, a compact
frequency domain fatigue calculation can be utilised where the random loading and response are
categorised using Power spectral density (PSD) functions and the dynamic structure is modelled
as a linear transfer function.
This chapter will review the available methods for performing fatigue analysis from PSDs and
shows that the Dirlik method gives the best comparable results with the traditional time domain
approaches. It also describes how these techniques are implemented in the Finite Element
environment to rapidly identify critical areas in the structure.

3.9.1 Introduction
It is often easier to obtain a PSD of stress rather than a time history. Take, for instance, the
dynamic analysis of complicated finite element models. Here it is often beneficial to carry out a
rapid frequency response (transfer function) analysis instead of a computationally intensive
transient dynamic analysis in the time domain. The offshore oil industry faced this problem in the
early 1980's. An offshore oil platform is a hugely complex structure that is subjected to random
wind and wave loading. A typical design analysis may have to consider over 70 load combinations
on the structure. The analysis is further complicated because the imposed loads are random and
dynamically excite the structure. A transient dynamic analysis in the time domain proves
impossible to carry out in this case.
A Finite Element analysis based in the frequency domain can simplify the problem considerably.
The designer can now carry out a frequency response analysis on the FE model to determine the
transfer function between wave height and stress in the structure. Using this, he simply multiplies
the PSD of wave height by the transfer function to arrive at the PSD of stress. A review of this
method is presented.
To take advantage of the fast frequency domain techniques for fatigue analysis, we require a
method of deducing damage from the PSD of stress.
In this chapter we will briefly review the technology behind time domain S-N analysis and then
draw a parallel approach in the frequency domain. A comparison study between the different
fatigue analysis techniques is presented. The chapter concludes by describing the Finite Element
approach for computing fatigue damage on vibrating components.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 67

3.9.2 Review of S-N analysis in the time domain

The starting point for any fatigue analysis is the response of the structure or component. In the
time domain this is usually expressed as a stress or strain time history. Fatigue occurs as a result
of stress or strain reversals in the time history. These are known as cycles. The significant aspects
of these are the ranges of stress in the cycle and also there mean stress. Today the range and
mean information is usually extracted from the time history using a procedure known as 'Rainflow
Cycle Counting'. Matsuishi and Endo first introduced the concept of rainflow ranges to the
scientific community over twenty years ago. An example of the way rainflow ranges are extracted
from a time signal is given by Downing [1].
The output from a rainflow cycle counting exercise is usually expressed as a range mean
histogram such as that shown below. The stress range of each cycle is given along the x axis, its
mean stress is shown on the y axis and the z axis gives the number of cycles contained in the time
history for each particular range and mean. This data was taken from the Howden HWP330 wind
turbine; Bishop [2] discusses the analysis.

400
stress MPa

300

200

100

-100

-200
0 50 100 150 200 250
Time Secs.

20

15
Cycles

10

0
0
2
2
6. 4
8 2.
17 .7 a
58 .9 P
2 4 1 M
34 1.
43
3 an
R 7. e
an 51 .6 -2 M
3 8 10
ge 60 9. .3
M 68
P
a

Figure 39 Range-mean histogram derived from time history by rainflow cycle counting
Each cycle will induce a certain amount of fatigue damage on the component. The total damage
caused by the time history can therefore be obtained by summing the damage caused by each
cycle shown in the stress range histogram. This approach is known as the Palmgren-Miner
accumulated damage rule after the two independent people who proposed it.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 68

The damage caused by each cycle is calculated by reference to the material life curve, in this case
the SN curve. The SN curve shows the number of cycles to failure, Nf, for a given stress range, S.
The total damage caused by N number of cycles is therefore obtained as the ratio of cycles to the
number of cycles to failure. The Palmgren-Minor rule can therefore be expressed as Equation (1).

Ni
Accumulated Damage = åNi
(1)
f

Where, Ni is the number of cycles with a particular stress range and mean; i is a ranging variable
covering all the possible range and mean combinations; and Nf is the number of cycles to failure
for a particular stress range and mean.
The accumulated damage is expressed as a proportion of the damage required to fail the material.
Therefore the fatigue life for the component can be determined from Equation (2).

Length of time history (2)

Fatigue Life =
Accumulated Damage

The fatigue life for the data shown in Figure 39 was found to be 0.2 years based on an aluminium
alloy 2024_HV_T4.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 69

3.9.3 Simple fatigue analysis using time history recreation

The most obvious method for calculating fatigue from a PSD is to regenerate a characteristic time
history and proceed as described above. This section describes the theory of time history
regeneration from PSDs and highlights some of the fundamental assumptions involved.
The frequency domain is simply another domain in which to view a time signal; the x-axis now
represents frequency instead of time. To convert a time signal into the frequency domain we
effectively split it up into a number of discrete sinusoidal waves of varying amplitude, frequency
and phase. When these are added together they form the original time signal. The algorithm used
to split the time history into its constituent sinusoidal components is the 'Fourier Transform'. This
returns a complex vector of values where each value represents the amplitude and phase of the
particular sinusoidal wave at a particular frequency. The amplitude can be obtained from the
modulus of the complex number while the phase is determined from the argument.
In practice we usually represent the frequency domain as a 'Power Spectral Density (PSD)' plot.
This is a normalised density plot describing the mean square amplitude of each sinusoidal wave
with respect to its frequency. A typical PSD plot is shown below.

3E4

2E4

1E4

0
0 10 20 30
Frequency Hz.

Figure 40 Typical PSD of a random time history

The mean square amplitude of a constituent sinusoidal wave can be determined by measuring the
area under the PSD over the desired frequency range. The amplitude can then be estimated using
Equation 3. The PSD does not contain any phase data.

(3)
Amplitude = 2 ⋅ Mean Square

To regenerate a time signal from the frequency domain we would usually perform an 'Inverse
Fourier Transformation' on the complex vector of frequency domain results. This would yield
exactly the same time history as we started with. When starting with a PSD this method is
inappropriate because the PSD does not contain any of the original phase information. For certain
time histories, however, we are able to make assumptions about the original phase content and
therefore regenerate a statistically equivalent time history. If the time history is taken from an
'ergodic stationary Gaussian random process', it is found that the phase is purely random between
-π and +π radians.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 70

A process is said to be stationary if its statistics are not affected by a shift in the time origin. (For
example, the statistics of a time history X(t) are the same as a time history X(t + τ) for all values of
τ.) To test for stationarity we take a number of recordings of the random process at different times.
The process is stationary if the probability distributions of the ensemble are the same for all points
in time. If the ensemble probability density function is Gaussian then the process is known as a
Gaussian random process. A stationary process is called an ergodic process if statistics taken
from one sample are the same as those obtained for the ensemble. With an ergodic stationary
random process, therefore, we can effectively take a single sampled time history from the process
and safely assume that this contains all the required statistical properties of the parent process.
For nonstationary processes the statistics obtained from a sampled time history would not be
representative of those of the whole random process as these would be continuously changing. In
practice we see that many time signals obey this criterion. Naturally occurring phenomena like
wind speed, wave height, etc. usually obey this along with vibrations from generators and
engines. Strictly speaking, the later examples are not random processes, however, the vibrations
occur as a result of many deterministic sources and the central limit theorem tends to make them
behave in the desired fashion.
To regenerate a time signal from a PSD we therefore assume that the original process was
'ergodic stationary Gaussian and random'. We can then generate random phase angles and add
these to the amplitude data given in the PSD. At this point we use the Inverse Fourier
Transformation to determine a statistically equivalent time history.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 71

3.9.4 Fast fatigue analysis methods in the frequency domain

This section describes a variety of approaches for computing fatigue life, or damage, directly from
the PSD of stress as opposed to a time history. The techniques fall into two broad categories,
those that estimate fatigue life directly and those that compute range mean histograms as an
intermediate stage. For more background information the reader is referred to Bishop [3] and
Halfpenny [4]. As before, these methods are only applicable for PSDs describing an ergodic
stationary Gaussian and random process.
In 1964 Bendat [5] proposed the first significant step towards a method of determining fatigue life
from PSDs. Bendat showed that the probability density function (pdf) of peaks for a narrow band
signal tended towards a Rayleigh distribution as the bandwidth reduced. Furthermore, for a
narrow banded time history Bendat assumed that all positive peaks in the time history would be
followed by corresponding troughs of similar magnitude regardless of whether they actually
formed stress cycles. Using this assumption the pdf of stress range would also tend to a Rayleigh
distribution. To complete his solution method, Bendat used a series of equations derived by Rice
[7] to estimate the expected number of peaks using moments of area under the PSD. Bendat's
narrow band solution for the range mean histogram is therefore expressed in Equation (4).

ìï S S ü
2

8⋅m0 ï
−
N (S ) = E [P ] ⋅ T ⋅ í ⋅e ý (4)
ïî 4 ⋅ m 0 ïþ

Where N is the number of cycles of stress range S occurring in T seconds. m0 is the zeroth
moment of area under the PSD (i.e. the area beneath the curve), and E[P] is the expected number
of peaks obtained by Equation (5).

m4
E[P] = (5)
m2

m4 and m2 are the 4th and 2nd moments of area of the PSD respectively where the nth moment of
area is obtained as:

mn = ò f n ⋅ G ( f )df

and G(f) is the value of the single sided PSD at frequency f Hz.
The term in brackets in Equation (4) is the Raleigh distribution. The figure below shows the range-
mean histogram obtained from the PSD of the time history given in Figure 39. The range mean
histogram given by Bendat contains no cycle mean data.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 72

RMS Power MPa^2. Hz^-1

3E4

2E4

1E4

0
0 10 20 30
Frequency Hz.

Counts Cycles

80

60

40

20

0
0
.6
05 .2
1 1
21 .7 a
16 3
3 2. P
42 .9 M
27 5 an
R 5 3. 9 e
an 63 73
-4
28
M
ge .6
M 44 .9
8
P
a

Figure 41 Range-mean histogram derived from PSD using Bendat's method

The problem with Bendat's narrow band solution is that it is extremely conservative when wider
band time histories are used. (We now observe a fatigue life of only 280.1 hours.) The reason for
this lies in the assumption that peaks are matched with corresponding troughs of similar
magnitude. This effect is illustrated in Figure 42. A narrow band time history is characterised by
each peak having a corresponding valley of similar magnitude. In comparison, a wide band time
history is characterised by smaller waves riding on a low frequency carrier. As Bendat assumes
that all positive peaks are matched with corresponding valleys of similar magnitude, the damage
is grossly exaggerated for wide band histories as shown.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 73

Narrow band

Stress
time

Wide band

Stress

time

Cycle counting using Bendat’s method

Stress

time

Figure 42 Why Bendat's method is conservative

During the 1980's the need for a rapid fatigue analysis method based in the frequency domain
became apparent to the offshore oil industry. Large jacket platforms were being designed and
fatigue failures had to be avoided. The transient dynamic analysis proved too intensive for the
time domain software because of the large structural models and high number of possible load
combinations. The dynamic wave and wind load data was already provided in the frequency
domain and it therefore seemed sensible to make use of the speed advantages inherent in a
frequency domain analysis. The problem was how to calculate a reasonably accurate fatigue life
using the resultant PSDs from the frequency domain analyses.
Sea state spectra are relatively wide banded and this effectively rules out the use of Bendat's
narrow band fatigue analysis because the results prove too conservative. Several methods were
developed to address this problem, the notable ones being Wirsching, Kam & Dover and
Hancock. These are semi-empirical approaches based on the narrow band solution. The latter two
methods are both in the form of an equivalent stress parameter and neither tends to work
particularly well when used outside the offshore platform industry. Wirsching's approach was
developed for the offshore industry but has been found to be applicable to a wider class of
industrial problems.
In other industries advances were also being made. Steinberg and Tunna both worked on the
problem for the electronics and rail industries respectively. Again neither of these methods tend to
work well outside their respective industries. For more information on these methods see
Bishop [3].
In 1985 Dirlik proposed an empirical closed form solution to the problem following extensive
computer simulations using the Monte Carlo technique. Although apparently more complicated
than some alternative methods it is still only a function of four moments of area of the PSD, these
being m0, m1, m2 and m4. This method has been found to be widely applicable and constantly
outperforms all of the other available methods. The Dirlik formulation is given in Equation (6).

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 74

(6)
N (S ) = E [P ]⋅ T ⋅ p (S )

Where: N(S) is the number of stress cycles of range S N/mm2 expected in time T sec.
E[P] is the expected number of peaks obtained by Equation (5).

−Z −Z 2 −Z 2
D1 Q D2 ⋅ Z 2⋅R 2
⋅e + ⋅e + D3 ⋅ Z ⋅ e 2
Q R2
p (S ) =
2 ⋅ m0

D1 =
(
2 ⋅ xm − γ 2 )
1+ γ 2

1 − γ − D1 + D12
D2 =
1− R

D3 = 1 − D1 − D2

S
Z=
2 ⋅ m0

1.25 ⋅ (γ − D3 − D2 ⋅ R )
Q=
D1

γ − x m − D12
R=
1 − γ − D1 + D12

m2
γ =
m0 ⋅ m4

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 75

m1 m2
xm = ⋅
m0 m4

Figure 43 shows the range mean histogram obtained using Dirlik's method for the wind turbine
data used in Figure 39. Again the method ignores the cycle mean stresses but now gives an
improved range mean histogram that is comparable with that obtained using a time domain
approach. The fatigue life is correctly calculated as 0.2 years.
RMS Power MPa^2. Hz^-1

3E4

2E4

1E4

0
0 10 20 30
Frequency Hz.
Counts Cycles

120
100
80
60
40
20
0
0
2
.0 0
95 19 1
5. a
28 .1 P
0 1 M
38 5. 1 n
R 47 0. 1 ea
an 57 5. 1
-3 M
ge 66 0.
86
M 76
P
a

Figure 43 Range-mean histogram derived from PSD by Dirlik's method

In terms of accuracy, Dirlik's empirical formula for rainflow ranges has been shown to be far
superior to the previously available correction factors. However, the need for certification of the
technique before its use meant that theoretical verification was required. Bishop [6] achieved this
when a theoretical solution for predicting rainflow ranges from the moments of area of the PSD
was produced. The theoretical approach devised by Bishop is computationally intensive and
shows little improvement on accuracy over Dirlik's empirical approach. For this reason, Bishop's
method gives credence to the Dirlik method but is seldom used for analysis purposes.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 76

3.9.5 Comparison between fatigue analysis techniques

Many comparative studies have been carried out to ascertain the accuracy of all the various
methods. The studies made by Bishop [2] are of notable significance. These studies took data
from the Howden HWP330 wind turbine, computed fatigue lives in the frequency domain and then
compared these with the life obtained from the time domain approach. A summary of the results is
given in the table below.

Load Case Narrow Dirlik Wirsching Bishop Chaudhury Hancock

Band

y12a 5.14 1.03 3.91 1.52 2.13 2.75

y19a 5.15 1.00 3.92 1.54 2.14 2.77
y27a 14.34 1.59 10.91 1.74 5.12 5.83
y35a 81.87 2.34 62.23 1.95 30.08 25.08
y12b 1.91 0.77 1.46 1.13 0.98 1.25
y19b 1.98 0.81 1.50 1.22 1.04 1.31
y27b 3.67 1.07 2.79 1.29 1.47 1.92
y35b 18.34 1.48 13.95 1.84 5.68 6.10
y12c 1.98 0.76 1.51 0.86 0.95 1.25
y19c 1.87 0.73 1.43 0.86 0.92 1.20
y27c 2.03 0.74 1.54 0.72 0.87 1.14
y35c 3.22 0.76 2.45 0.66 1.15 1.42
y12d 2.09 0.84 1.59 1.15 1.03 1.33
y19d 2.03 0.83 1.54 1.17 1.02 1.31
y27d 2.92 1.01 2.22 1.15 1.23 1.62
y35d 7.50 1.12 5.70 1.23 2.75 3.29
y12e 2.80 0.99 2.13 1.27 1.50 1.95
y19e 3.06 1.01 2.33 1.44 1.64 2.12
y27e 3.50 1.03 2.67 1.53 1.65 2.16
y35e 8.81 1.11 6.71 1.99 3.31 4.15
y12f 3.86 0.98 2.93 1.43 1.66 2.18
y19f 3.97 1.00 3.02 1.61 1.78 2.33
y27f 3.96 1.01 3.01 1.57 1.76 2.31
y35f 5.59 0.98 4.25 1.65 2.17 2.80

avrg 7.98 1.04 6.08 1.36 3.08 3.32

Figure 44 Comparison between different frequency domain fatigue analyses

The table clearly shows that the Dirlik approach is remarkably robust. It shows an average
discrepancy of only 4% from the fatigue life calculated in the time domain.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 77

3.9.6 FE based vibration analysis in the frequency domain

The previous section discussed methods of determining fatigue life from PSDs of stress. This
section considers how these PSDs are obtained from finite element analysis.
The dynamic behaviour of components can be determined both in the time and frequency
domains. In the time domain this involves a complicated and often lengthy transient analysis. In
the frequency domain it uses a very fast 'transfer function' technique. Essentially the frequency
domain breaks down a signal into its constituent sinusoidal waves following Fourier's theory. The
transfer function relates the amplitude of the input force or moment to the amplitude of the output
stress for each frequency of sinusoidal wave. An example transfer function plot is shown below.
This relates input force (N) to output stress (MPa) for a certain node on a FE model. The peaks
represent resonant modes in the component.
Transfer function [MPa/N]

2000

1000

0 50 100
Frequency [Hz]

Figure 45 Transfer function relating input force to output stress at a node

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 78

3.9.6.1 Calculating the linear transfer function

This section describes the calculation performed by the Finite Element solver when calculating the
linear transfer function.
Consider the simple bracket component shown below subjected to a single input load. PSDs of
stress are required at each node on the model in order to determine the fatigue life of the
component.

Figure 46 Simple FE model of a mounting bracket

The equation of motion for the component is given in Equation (7).

[M ] ⋅ x(t ) + [C ] ⋅ x(t ) + [K ] ⋅ x(t ) = f (t ) (7)

Where:
[M] is the global mass matrix
[C] is the global damping matrix
[K] is the global stiffness matrix
x(t) is the vector of nodal displacements
f(t) is the vector of nodal forces
In the case of a single input system the vector f(t) has only one non-zero component.
If the input load is sinusoidally varying with respect to time then the displacement vector v(t) will
also be sinusoidally varying. Expressing this in exponential form yields the expressions for load
and displacement shown in Equation (8).

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 79

x(t ) = X ⋅ e i ⋅ω ⋅t
(8)

f (t ) = F ⋅ e i ⋅ω ⋅t

Where:
F is the amplitude of the sinusoidally varying force of frequency ω rad/sec and X is the amplitude
of the induced displacement.
Differentiating displacement with respect to time yields expressions for the velocity and
acceleration of the component. These are given in Equation (9).

x (t ) = i ⋅ ω ⋅ X ⋅ e i⋅ω ⋅t
(9)
x(t ) = −ω 2 ⋅ X ⋅ e i⋅ω ⋅t

Substituting these into Equation (7) and simplifying yields the following formula for the amplitude
of displacement with respect to frequency:

X (ω ) = H (ω ) ⋅ F (ω ) (10)

Where: H(ω ) is the linear transfer function given by:

[
H (ω ) = − [M ] ⋅ ω 2 + [C ] ⋅ i ⋅ ω + [K ] ]−1

X(ω) is the vector of displacement amplitude

F(ω) is the vector of force amplitude
It is usual to compute the transfer function between input force and output stress instead of
displacement shown in Equation (10). This is carried out in the FE program following the back
substitution stage.
Classical FE analysis results in a tensor of 6 component stresses at each node. These consist of
3 axial and 3 shear stress components and are expressed mathematically in Equation (11).

Axial H xx (ω ), H yy (ω ), H zz (ω )
(11)

Shear H xy (ω ), H yz (ω ), H xz (ω )

N.B. for the two dimensional plate elements shown in Figure 7, the values of Hzz(ω), Hyz(ω) and
Hxz(ω) are all zero.
Frequency can also be expressed in terms of Hertz after using the appropriate conversion.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 80

3.9.6.2 Single force input

Having found the relationship between component stress and applied load, the component stress
PSD 'G(f)' for a single input load PSD 'W(f)' can be found very rapidly by multiplying the input PSD
by the transfer function as in Equation (12).

G( f ) = H ( f ) ⋅ W ( f )
2 (12)

3.9.6.3 Multiple force input

This section introduces the analysis of multiple random inputs. Figure 63 shows a simple example
of a flagpole with two flags flying at different heights. A typical application would be to determine
the bending stress at the base of the flagpole as a result of the two random wind speeds seen by
the flags.
The time history of wind speed at the location of the flags can be recorded using anemometers.
PSDs of wind speed can be calculated from these; however, the PSDs alone do not provide
information on the phase relationships between the two measured time histories.

Wind
Speed
x(t)

Wind
Speed
y(t)

Bending stress
required at the base
of the flagpole

Figure 47 Example of two random process inputs

With multiple random processes we also require the sequential relationship between the two time
histories. If the two flags are far enough apart then the wind speed witnessed by one will be
completely independent of that on the other. As they are moved closer together then a correlation
between the two time histories will be seen. The two time histories are correlated because the
random wind turbulence incident on one flag has a sufficiently large range of influence to also
affect the response at the other.
To calculate the stress at the base of the flagpole it is insufficient to simply sum the reactions from
the two input PSDs, instead we must sum the reactions from the input and cross-power spectra.
The cross-power spectra contain information on the joint statistics of the two processes. If the two
processes are correlated then the sequencing effects may act to increase or decrease the base
bending stress depending on whether the forces are in or out-of-phase. For a mathematical
explanation of this see Newland [8].

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 81

The single-sided PSD function of component stress at the base of the flagpole, Gzz(f), is therefore
determined by Equation (13).

2 2
G ( f ) = åå H a ( f ) ⋅ H b ( f ) ⋅ Wab ( f ) (13)
*

a =1 b =1

H1(f) and H2(f) are the transfer functions relating stress at the base of the flagpole to wind load
incident on flags x and y, respectively. The asterisk indicates the complex conjugate. W11(f) and
W22(f) are the PSD functions of wind speed at flags x and y, respectively, and W12(f) and W21(f)
are the cross-power spectral density functions. For a general loading with n simultaneous forcing
functions the PSD of stress can be obtained from Equation (14).

n n
G ( f ) = åå H a ( f ) ⋅ H b ( f ) ⋅ Wab ( f ) (14)
*

a =1 b =1

3.9.6.4 Finding the response PSD for the maximum principal stress
To this point we have considered calculating the response PSD for a component of stress. In
reality each node or element has 6 component stresses pertaining to 3 axial and 3 shear
components. This was discussed in Equation (11). We must now consider how to obtain the
response PSD for the maximum principal stress.
In general, we cannot calculate the principal stresses using component stress PSDs as they do
not have information on the relative phase between each component. However, if we assume the
stresses are uniaxial then all component responses will be in-phase. The response PSDs for the
three principal stresses can therefore be obtained using an Eigenvalue routine as shown in
equation (15).

éGσ 1 ( f )ù éG xx ( f ) G xy ( f ) Gxz ( f )ù
ê ú ê ú
êGσ 2 ( f )ú = eigen values êGxy ( f ) G yy ( f ) G yz ( f )ú
(15)
êGσ ( f )ú êG xz ( f ) G yz ( f ) G zz ( f )ú
ë 3 û ë û

This is the method that is currently implemented in FE-Fatigue. It requires that all the loading (both
auto- and cross-spectral densities) has been included in a random vibration finite element
analysis. The FE results will be component stress PSDs (with units 'stress^2/Hz') and these are
translated by the FE2FES module into a FES file of principal stress PSDs. This is valid where
there is a single loadcase or a uniaxial stress condition.
An alternative method (not currently implemented in FE-Fatigue) for evaluating principal stresses
is also described by Halfpenny [9]. This method does not require a uniaxial stress state but
requires a rather more complicated analysis procedure.
The absolute maximum principal stress is used for fatigue calculations.
The figure below shows the fatigue log damage plot arising on the bracket due to a vertical
vibration at the bolt hole. The critical fatigue life is determined as 206 hours.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 82

Critical region

Figure 48 Fatigue contor plot of Log Damage

3.9.7 Conclusions
The concept of frequency domain fatigue analysis has been presented, where the random loading
and response are categorised using Power spectral density (PSD) functions. All the current
methods are briefly reviewed and conclusions are drawn showing that the Dirlik method is
recommended for general use. The Dirlik formulation is presented along with an explanation of the
FE analysis involved.

3.9.8 References
1. Downing SD. and Socie DF. (1982). "Simple rainflow counting algorithms." Int. J Fatigue,
January 1982, 31-40.
2. Bishop NWM, Hu Z, Wang R, Quarton D. (1993). "Methods for rapid evaluation of fatigue
damage on the Howden HWP330 wind turbine." British Wind Energy Conference, York.
3. Bishop NWM. and Sherratt F (1989). "Fatigue life prediction from power spectral density data."
Environmental Engineering, 2.
4. Halfpenny A. and Bishop NWM. (1997). "Vibration Fatigue." nCode International Ltd. 230
Woodbourn Road, Sheffield, S9 3LQ. UK.
5. Bendat JS. (1964). "Probability functions for random responses." NASA report on contract
NAS-5-4590.
6. Bishop NWM (1988). "The use of frequency domain parameters to predict structural fatigue."
Ph.D. thesis, University of Warwick, UK.
7. Rice SO. (1954). "Mathematical analysis of random noise." Selected papers on noise and
stochastic processes, Dover, New York.
8. Newland DE. (1984). "An introduction to random vibrations and spectral analysis” (2nd edition),
Longman Inc., New York.
9. Halfpenny A. (1999). "A frequency domain approach for fatigue life estimation from Finite
Element Analysis.” International Conference on Damage Assessment of Structures (DAMAS 99)
Dublin.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 83

4. Time Histories
4.1 Why time histories are required
Components under static loading at normal ambient temperatures will typically not fail unless that
loading exceeds the static fracture limit of the material. To induce fatigue failure at loads below this
limit, the load must be cyclic.
There are two ways in which FE data can be subjected to cyclic loading;
• by running a transient or time-step analysis in the FE itself
• by applying time histories of load or modal participation to static or modal FE stress
analysis results
Sections 4.2 to 4.5 are concerned with the use of external time histories, section 4.6 with the use
of transient or time step analyses.

4.2 Constant amplitude versus variable amplitude

If only a single static load case has been applied to the FE model, the loading can be defined as
either a constant amplitude or variable amplitude. Constant amplitude loading assumes that the
same cyclic loading sequence will occur throughout the lifetime of the component. Variable
amplitude loading allows a sequence of load values to be defined which will typically result in a
number of differently sized cycles being experienced by the component.
A constant amplitude loading time history typically looks like this:

Peak to peak
Peak to peak
1.5

0.5
Load scale factor

0
1 2 3 4

-0.5

-1

-1.5

Figure 49
This example shows a “peak to peak” or “fully reversed” history which will subject the FE model to
a peak positive loading equivalent to the load applied in the FE model followed by a peak negative
loading equivalent to -1 times the load applied in the FE model. This defines 1 cycle and the
damage for this cycle is then used to calculate the expected lifetime of the component in cycles.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 84

Another example of a constant amplitude history is the zero-peak example, as shown below:

Zero to peak
Zero to peak
1.2

0.8
Load scale factor

0.6

0.4

0.2

0
1 2 3

Figure 50
This example shows only tensile loading, from zero to the maximum load applied in the FE model
and back to zero. Finally FE-Fatigue recognises a third type of constant amplitude definition,
where the user can specify the maximum and minimum values to be applied. This is particularly
necessary when a unit loading has been applied in the FE model and therefore the time history is
describing the maximum and minimum load, as follows :

Sc aled data
Sc aled data
2000

1500
Load scale factor

1000

500

0
1 2 3 4

-500

Figure 51
In this case, the stress data in the FE model, which was the result of a unit load being applied, is
multiplied by 1776 to give the stress as a result of the peak positive loading and then by -300 to
give the result of the negative loading. This defines a stress cycle that can be used to calculate

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 85

damage.

Note: There are no specified X-axis units. Although time histories are typically measured with
time as the X-axis, this time is almost never used as a measure of life. The normal descrip-
tion is in cycles to failure, or number of repeats of the history (in the constant amplitude
case, these are the same thing.) The type of loading and the units of loading must corre-
spond to the loading of the FE model.
Warning: The software makes no checks with respect to the suitability of the units.

The following graph shows a typical variable amplitude load sequence from the SAE validation
cases (Appendix C). The file contains 17081 points and has been normalised to unit maximum.

DISPLAY OF SIGNAL: SAETRN.DAC

17081 points.
1
9 pts/Secs

Displayed:

17081 points.

from pt 1
Load (No units)

Full file data:

Max = 1

at 0 Secs

Min = -0.4955

at 1743 Secs

Mean = 0.3857

S.D. = 0.2352

-0.6 RMS = 0.4517

Tim e (Secs)
0 1897.778

Figure 52
During the fatigue life calculation process, variable amplitude time histories are converted into a
series of loading cycles using peak-picking and rainflow cycle counting. For details of these
algorithms, refer to the Fatigue theory manual.
Section 4.5 describes how to pre-process the time histories using these techniques to speed up
the calculation process.

4.3 Obtaining and creating variable amplitude time histories

FE-Fatigue can process nCode .DAC format, and MTS RPC format time histories.
The format description for DAC files are detailed in the document called format.pdf and the MTS
file format can be obtained from MTS via their website on www.mts.com. Typically time history data
comes in many other forms and so it is necessary to use the nSoft tools provided with FE-Fatigue
to convert data from other formats.
The following table shows which nSoft functions can be used to import, create and view time

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 86

history data. Consult the nSoft online manuals for details of the use of these functions.

Program Function Performed

atd Converts data from a variety of ASCII formats
remdac Converts data from MTS RPC II and RPC III formats
coe Used to type in values from the keyboard
waves Creates sine, swept sine, triangular and square waves.
regen Creates time histories from rainflow and Markov matrices
qld Views single channel time history data
mfd Views multiple channel time history data
muledt Views and edits multiple channel DAC files and RPC files

Many software packages for data acquisition and analysis create time histories directly into .DAC
or RPC III format. Consult your software supplier for further information on third party .DAC
support.

4.4 Multiple load cases

When two or more independent load case results have been obtained from the FE analysis there
are special conditions applied to the time histories that can be used in the analyser. The analysis
procedure requires that the time histories have the same number of points, and that each point in
each time history occurs at the same time as the same point number in the other histories. The
histories can then be said to be synchronised, or in-phase, with respect to each other, although
this does not necessarily mean that their frequencies are in-phase.
Also, they must have the same number of points, start time and sampling rate. If this is not the
case, nSoft utilities can be used to synchronise the time histories.
Imposing these conditions on the time histories allows the principle of linear superposition to be
applied to the loads and stresses at each point in the time history, thereby creating a local stress
or strain time history that combines the load cases together.
See “6. Principles of Linear Superposition” on page 96. for details of how linear superposition
works.
Pre-calculated rainflow matrices cannot be used as input for multiple load case analysis as the all
important phase information has been lost. See section 4.5 for details on using pre-calculated
rainflow data for single load case calculations.

4.5 Using nSoft time history processing to increase calculation speed

For single load case calculations, the analyser automatically runs a rainflow cycle count on the
input time history and then calibrates the rainflow matrix for each node or element to be
processed, thereby preventing the time consuming cycle counting at each calculation point. It is
also possible to pre-calculate the rainflow matrix using the nSoft module cyc, and to use this as an
input to a single load case calculation in FE-Fatigue. The results will be the same if the same limits

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 87

are used for the rainflow matrix. Alternatively if rainflow data is available from another source
nSoft can be used to convert it into a rainflow file compatible with FE-Fatigue.
For multiple load case calculations, rainflow matrices cannot be used as they lose the
synchronisation information between the channels.
Typically, multiple load case analysis is the most time consuming and it is therefore useful if some
form of data reduction can be performed to speed up the calculations. There are two ways to do
this, each of which can be applied independently or they can be applied together.
• The first is to reduce the number of nodes or elements to be calculated.
• The second is to remove data from the input time histories that is not contributing to the
damage.
Both methods involve some risk in that it is possible to create situations in which they will lose
information, and ultimately it is strongly advised that a complete superposition analysis be
performed on the whole model if the correct results are to be obtained. However, for the purposes
of quick interactive analysis, both methods can be used with caution and engineering judgement.
This section focuses on the second method, where a peak-slicing technique is used to extract the
turning points in each time history. Rainflow cycles are extracted from a time history using the
turning points, and in the case of proportionally loaded components this means we can reasonably
use the peaks and valleys thus extracted and discard the points in-between. This presents a
problem as shown in the data below:
Time series 1 Time series 2
0 0
1 -3
2 -5
3 3
5 7
10 9
7 6
6 1
9 -4
3 -1
-5 3
-10 7
-5 -3
0 -7
3 -5
0 0

Figure 53

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 88

15

10

Series1
0
Series2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

-5

-10

-15

Figure 54
The first turning point in Series 2, which exists at point 3, is not matched by a turning point in
Series 1. If only the peaks and valleys are extracted, the following reduced histories are produced.

15

10

Series1
0
Series2
1 2 3 4 5 6 7

-5

-10

-15

Figure 55
Although there are the same number of peaks and valleys (and this is not generally the case), the
synchronisation has been lost and points which now correspond on the X-axis did not originally
correspond before the peak-valley extraction. To compensate for this, peak-slicing keeps not only

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 89

all the peaks and valleys but also the corresponding points in all other channels. This leads to the
following reduced histories in our small example:

15

10

Series 1
0
Series 2
1 2 3 4 5 6 7 8 9
-5

-10

-15

Figure 56
By keeping 2 extra points in total the synchronisation is retained.
The nSoft program pvxmul performs this task on .DAC format files, creating reduced .DAC files
with extension .PVX that can be used as input to the loading form of FE-Fatigue. In addition a
peak picking or cycles gate can be applied to reduce the length of the data further. However
extreme caution must be employed when using this option as data which contributes to damage
may be lost.
When the loading is not proportional the risk in using this technique is increased significantly. In
full multiaxial loading situations the damage does not necessarily occur at the time when a turning
point exists in any of the input loading channels.

4.6 Transient or time-step analysis

In a transient or time step analysis, the time history data is embedded in the FE results and hence
in the FES file. No external time history file is required and hence the FE-Fatigue loading form is
not required.
If the transient problem is one that can reasonably be solved using a modal formulation, it may be
handled more efficiently by using a modal superposition approach. This method is discussed in
more detail in the worked example Transient Analysis and Modal Superposition using NASTRAN.
.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 90

5. Materials Data
Good quality materials data is crucial to good fatigue life prediction. FE-Fatigue includes over 150
data sets for steel and aluminium alloys in its standard materials database. It also has the ability to
generate cyclic fatigue data from static properties. If there is no appropriate data in the database
nCode has a large library of data from tests conducted at its own laboratories which it can make
available, or tests for a specific material can be made to determine the necessary properties.

5.1 Materials database

nCode’s materials database is managed by an nSoft program called mdm (materials data manager),
which is available from the nSoft menu system or from the group information form in FE-Fatigue
(see ”5.5 Auto-generated data on page 92’). The mdm program allows the entry, display and
editing of materials data in either a shared or local database. Access to the shared database is
secured by a password known to your systems administrator. Data integrity and traceability issues
dictate that the ability to change shared materials data should be restricted and it is strongly
advised that in an organisation where a number of users have access to a shared database a
materials administrator be appointed to care for the entry and editing of data in a shared database.
Proprietary data can be merged into nCode’s standard database, or a new database can be
created which contains only company data or a mixture of company data and some of the data
from the standard database.
Consult the mdm on-line manual for details on setting up and using nCode materials databases.
See ”5. Materials Data on page 90’ re. using the databases provided with the system.

5.2 S-N data

S-N data can be characterised and stored in many forms. At present the only data format available
for use in FE-Fatigue is a parametric form which describes the S-N relationship as two straight
lines on a log-log scale.

Stress

SRI1
Slope=b1
UTS
Slope=b2

NC1 Life

Figure 57 The S-N relationship

The required parameters are as follows :

• SRI1 Intercept of first slope at Life=1 cycle in range
• b1 Slope of first line
• NC1 Intercept point of the two slopes
• b2 Slope of second line

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 91

If b2=0, this means that at the value of stress indicated by NC1 there is in effect a fatigue limit,
where any cycle with a stress below this value will cause zero damage.
If SRI1 is greater than the UTS of the material then the software adds an extra line between the
UTS (multiplied by 2 to convert from amplitude to range) and the value on curve one at 1,000
cycles. Below 1000 cycles the S-N method is likely to be invalid, and any stress range values
greater than twice the UTS will cause a static fracture failure and to calculate a finite life for such
values of stress would be incorrect.
In addition to these parameters an optional parameter SE can be entered which describes the
standard error in log life. This is obtained as part of the regression analysis performed to derive
the linear parameters from the life test results.
If S-N data is available for the model to be analysed which is not in the form described above,
contact nCode support for advice on how to convert data to this form.
Spot weld data is stored in this format. There are typically two separate entries in the database,
one for the data measured for the sheet, and the other containing data measured for the nugget.
To use the data, the spot weld flag must be set to indicate that the data is for spot welds.

5.3 E-N data

The strain life analysis method allows fatigue calculation on low cycle problems, where significant
plasticity may occur. In addition to the damage curve the relationship between stress and strain is
also required. This is described by the Ramberg-Osgood equation:

e = (s/E) + (s/K’)1/n’
where e is strain, s is stress, E is Young’s Modulus and n’ and K’ are parameters derived from
regression analysis of test data.
The life curve is described by the Coffin-Manson equation:
′
Sf b ′ c
e = ------- ⋅ ( 2 Nf ) + ef ( 2 Nf )
E
where e is strain and Nf is the number of cycles to failure. The two parts of the equation come from
regression analysis on the elastic and plastic life lines from strain controlled life tests. The
obtained parameters are the coefficients sf’ and ef’ and the exponents b and c. The standard
errors for the three regression analyses may also be entered. This allows a confidence factor to
be entered in the calculations, which takes into account the variability in the physical tests.
In addition to these parameters, a material cut-off may be entered which describes the life beyond
which the damage is assumed to be zero.

5.4 Multiaxial data

In the materials data manager properties can be entered for either multiaxial strain life or
multiaxiial safety factor or both. For multiaxial strain-life, all the regular strain life properties must
be entered along with UTS, Youngs Modulus and Yield Stress. Yield Stress is used by the
Fatemi-Socie model. All the multiaxial damage models use common uniaxial fatigue parameters.
However, two of the algorithms, Fatemi-Socie and Wang-Brown, require one additional parameter
each, n and S respectively. Both parameters can and should be obtained from results of pure
uniaxial and pure torsion fatigue tests. Default values of 0.6 and 1 can be used, but it has been

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 92

observed that the Wang-Brown method can be quite sensitive to the value of S whereas Fatemi-
Socie is less sensitive to the value of n.
For multiaxial safety factor, three values are required. McDiarmid requires the type A and type B
fatigue endurance limits, Dang Van requires the tension fatigue limit (entered as Type A fatigue
limit) and the hydrostatic stress sensitivity. See the section 3.5.4. Dang Van Criterion for more
information. When using McDiarmid only, set HSS to 0.05. When using Dang Van only, set the two
fatigue limits to the same value.

5.5 Auto-generated data

Data for E-N analysis may be generated automatically in FE-Fatigue using the following rules:

5.5.1. Multiaxial E-N data

The Fatemi-Socie parameter is set to 0.6 and the Wang-Brown parameter is set to 1 when data is
generated.

5.5.2. Multiaxial safety factor

Generated properties are not available for multiaxial safety factor methods.

5.5.3. E-N data

The first method of approximating the strain life relationship from monotonic properties was
proposed by Manson and later modified by Muralidharan[1]. The procedure is usually referred to
as the method of universal slopes and can be applied to any metal. However it requires
knowledge of the parameter RA (reduction in area) which is not always available. If this property is
known, properties can be calculated using this method and entered into the mdm database.
However this is not implemented directly in the FE-Fatigue analyser.

Parameter Universal Slopes Modified Universal Slopes

(Manson) (Muralidharan)
sf' 1.9 Rm 0.623 Rm 0.823 E0.168

b -0.12 -0.09
ef' 0.76 ef0.6 0.0196 ef 0.155 (Rm / E) -0.53

c -0.6 -0.56
K' sf ' /( ef ')0.2 sf' /(ef') 0.2
n' 0.2 0.2

where :
Rm = the ultimate tensile strength
ef’= the true fracture strain calculated from ln (1 / (1 - RA)) and
RA = the reduction in area.
Baumel Jr. and Seeger [2] have compiled an alternative approach based on the results of more

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 93

than 1500 fatigue tests. Currently the approach is limited to plain carbon and low to medium alloy
steels, aluminium and titanium alloys.

Parameter Steels Aluminium/Titanium alloys

sf' 1.5 Rm 1.67 Rm

b -0.087 -0.095
ef’ 0.59 α 0.35
c -0.58 -0.69
K' 1.65 Rm 1.61 Rm

n' 0.15 0.11

The ductility factor α is calculated from:

α = 1.0 for values of R m / E < 3 x 10 -3

α = (1.375 - 125 R m / E) for R m / E > 3 x 10 -3

where Rm is the UTS and E is Young’s modulus. Since only Rm and E are required this is a
simpler option and has been included directly in the analyser.

5.5.4. S-N data

Approximate S-N damage curves can be generated purely on the basis of ultimate tensile
strength, UTS. The curves are constructed by fixing the stress axis intercept, 1 cycle, at the value
of the UTS, fixing the stresses at 1000 cycles and the endurance limit according to the fractions of
UTS detailed below:

Cycles Stress
Ferrous Alloys. 1 1.000 x UTS
1,000 0.900 x UTS
1,000,000 0.357 x UTS
Titanium Alloys. 1 1.000 x UTS
1,000 0.800 x UTS
1,000,000 0.307 x UTS
Aluminium Alloys. 1 1.000 x UTS
1,000 0.700 x UTS
500,000,000 0.258 x UTS
Other Alloys. 1 1.000 x UTS
1,000 0.800 x UTS
100,000,000 0.274 x UTS

S-N data can be derived directly from the elastic portion of the strain-life curve by converting strain

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 94

to stress and reversals to cycles. This data is only valid above the transition life (1000 to 10000
cycles) as it does not take any plasticity into account.

5.5.5. Validity of generated data

The use of generated properties is only encouraged when no good quality measured data is
available. It is useful for approximate calculations and for assessing the likely increase or
reduction in life as a result of changing to a stronger or a weaker material.
Generated data is not suitable for spot weld analysis.
See the section on multiaxiality safety factor theory for suggestions on the estimation of the
required properties for those algorithms.

5.6. Surface finish correction

Note: This option is not available for multiaxial safety factor or spot weld.

A very high proportion of all fatigue failures nucleate at the surface of components and so surface
conditions become an extremely important factor influencing fatigue strength. The usual standard
by which various surface conditions are judged is against the polished laboratory specimen.
Normally, scratches, pits, machining marks influence fatigue strength by providing additional
stress raisers that aid the process of crack nucleation. Broadly speaking, high strength steels are
more adversely affected by a rough surface finish than softer steels, for this reason the surface
correction factor, Csur , is strongly related to tensile strength. The surface finish correction factor
is often presented on diagrams that categorize finish by means of qualitative terms such as
polished, machined or forged.
Similarly, fatigue lives, particularly for high cycle environments, can be extended significantly by
treating the surface using methods such as nitriding and shot-peening. The nCode Fatigue Theory
Book gives a full description of these methods and their effects.
In FE-Fatigue the surface finish and treatment are simply selected from a list. The program then
uses a series of look-up tables to calculate the surface effect factor which is then applied as a
fatigue strength reduction factor (see 5.7) The actual calculated values for the surface factor can
be viewed within the materials manager, mdm.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 95

5.7. Fatigue strength reduction factor

Note: This option is not available for multiaxial E-N or multiaxial safety factor.

Each group of nodes and elements within FE-Fatigue can have a fatigue strength reduction factor
associated with it. The default is 1. This value affects the fatigue limit as shown in the following
diagram :

S tres s

S1

S2

L ife

Figure 58

If S1 is the original fatigue limit and k is the fatigue strength reduction factor then S2=S1*k and the
slope is recalculated accordingly. In the case of surface effect the value k is less than 1 for rougher
surfaces ( as shown in the diagram) and greater then 1 for smoother surfaces, or treated surfaces.
The fatigue strength reduction factor can be used to address other factors such as size, loading or
notch effects.
1] MURALIDHARAN, MANSON (1988) A Modified Universal Slopes Equation for Estimation of Fatigue
Characteristics of Metals, Journal of Engineering Materials Technology, Vol. 110, pp. 55-58.
2] BAUMEL JR., A and SEEGER, T. Materials Data for Cyclic Loading, Supplement 1, Materials Science
Monographs, 61 published by ELSEVIER ISBN: 0444 88603 6.

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 96

6. Principles of Linear Superposition

In FE-Fatigue, multiple input out-of-phase loading is dealt with by assuming that the stresses from
the independent load input calculations can be added together to produce a net total stress. There
are some important factors which allow this assumption to be made:

1 – The stresses must be in the same direction. Component stresses in the same direction are
typically used. It is not safe to add together principal stresses as the direction of the principals is
often different.

Note: This also affects nodal averaging. Stresses must only be averaged at a node if they have
the same co-ordinate system.

2 – The stresses are linear elastic. Elastic-plastic stresses should not be superimposed in this
way.

3 – That each point in the input load-time histories occur at the same time as each other and are
independent of the points before and after.

So at each point in the time history we can calculate the total stress by dividing all stresses by the
value of the input load to produce a unit load case stress, multiplying by the value of the point in
the time history and adding the values from each load case together.

Therefore, if we have two load cases with the following stress values:

Table 1:
X Y Z XY YZ XZ
LC 1 20 -50 10 -25 -28 15
LC 2 -30 50 0 -10 0 0

Where load case LC1 had 2 units of load applied and LC2 had 5 units of load applied, and the time
histories corresponding to the load cases were as follows:

Table 2:
Load case 1 Load case 2
0 10
10 -5
20 10
-5 -5
25 10
-10 -5
-15 10
-5 -5
0 10

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 97

Then the combined stress time history for the X stress at point 1 is calculated as follows :

Total X= (X1*(LC1 value)/(LC1 applied load)) + (X2*(LC2 value)/(LC2 applied load))

= (20.0*0.0/2.0) + (-30.0*10.0/5.0)
= -60.0

Total Y= (Y1*(LC1 value)/(LC1 applied load)) + (Y2*(LC2 value)/(LC2 applied load))

= (-50.0*0.0/2.0) + (50.0*10.0/5.0)
= 100.0

and so on, giving the following results:

Table 3:
X Y Z XY YZ XZ
-60.0 100 0 -20 0 0
130 -300 50 -115 -140 75
140 -400 100 -270 -280 150
-20 75 -25 72.5 70 -37.5
190 -525 125 -332.5 -350 187.5
-70 200 -50 135 140 -75
-210 475 -75 167 210 -112.5
-20 75 -25 72.5 70 -37.5
-60 100 0 -20 0 0

The 6 stresses can then be used directly, or more commonly be combined together to produce a
maximum principal stress or Von Mises stress value.

The principle of linear superposition works equally well for quasi-static cases (where the
instantaneous structural response is simulated by the linear superposition of scaled static loading
cases) and transient-dynamic cases, where, so long as the problem can reasonably be assumed
linear, the instantaneous structural response may be simulated by the linear superposition of
scaled mode shapes. This method is discussed in more detail in the worked example Transient
Analysis and Modal Superposition using NASTRAN.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 98

7. Calculation of fatigue life from rainflow matrices

This appendix shows how fatigue life is calculated from a rainflow matrix.

7.1. Strain-life Fatigue Analysis from a Rainflow Matrix

Technical Overview

The theory of strain-life analysis from a rainflow matrix is based on the theory used for a time
history. The only difference is in the interpretation of the loading environment.

Types of rainflow matrix

The result of a rainflow cycle count of a strain history can be described as a list of closed cycles
and unclosed reversals (See the technical background of rainflow cycle counting for details). In
this document we assume that all the reversals are converted to closed cycles by adding an extra
data point to the sequence which corresponds to the absolute largest value of the sequence.
Therefore the result of the rainflow count will consist solely of closed cycles.
Each cycle can be described as follows:

• Maximum strain value

• Minimum strain value
• Direction of loading

These values can then be classified into a variety of histograms, the form of which is dependent
on user preference. Typically these forms are:

• Maximum-Minimum-Counts (X-Y-Z)
• Range-Mean-Counts (X-Y-Z)
• Range-Counts (X-Y)
• From-To-Counts (X-Y-Z)

where Range = (max-min)

Mean = (max+min)/2

and From-To uses the maximum and minimum values but positions the count in a different part of
the histogram depending on whether the maximum or minimum occurs first.

For Example:
Consider the following time series:

1000,200,800,500,1000,500,800,0,800,500,1000

It is displayed below:

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 99

1200

1000

800

600 Series1

400

200

0
1 2 3 4 5 6 7 8 9 10 11

Figure 59

The rainflow procedure yields the following results :

Table 4:
Cycle maximum Minimum direction
1 800 500 down
2 1000 200 down
3 800 500 up
4 800 500 down
5 1000 0 down

Figure 60
After a time history fatigue analysis, a typical hysteresis plot would look like this:
Cross Plot of Data : CYC01
Material : Generated UTS : 100
Factor : 1 Kf : 1
Life : 1.83E5 Repeats

60

40
Stress(MPa)

20

-20

-40
0 200 400 600 800 1000
Strain(uE)

Hysteresis loops for file CYC01

Figure 61

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 100

The three cycles which have the same strain maximum and minimum each have a different value
of mean stress. It is possible to ascertain whether the cycles are “hanging” from the top limb of a
loop or “standing” on the bottom limb by the direction of loading ( “down” or “up” ) but once the
data is in a histogram the sequence is lost and it is not possible to identify from which loop the
cycle is hanging from or standing on. This is only important if a mean stress correction is used and
the method of estimating the mean stress in this case is described in the calculation procedure
below.

1 - Range-Mean

To classify the data into a range mean matrix, calculate the range and mean and insert the counts
into a 2D histogram. The example above can be classified into an 8x8 matrix as follows :

Table 5:
Range
0:150 150: 300: 450: 600: 750: 900: 1050:
300 450 600 750 900 1050 1200
0:150 0 0 0 0 0 0 0 0
150:300 0 0 0 0 0 0 0 0
300:450 0 0 0 0 0 0 0 0
Mean 450:600 0 0 0 0 0 0 1 0
600:750 0 0 3 0 0 1 0 0
750:900 0 0 0 0 0 0 0 0
900:1050 0 0 0 0 0 0 0 0
1050:1200 0 0 0 0 0 0 0 0

Figure 62

Note that the choice of bin size and limits and the policy of where to place a count when the value
is between bins may have a significant effect on the life calculation. Consider the cycle whose
maximum is 1000 and minimum is 200. The range is 800 which falls neatly into the bin -750:900
but the mean is 600 which lies on the border of two bins. In this case we have put it into the bin
with the larger limits - this is a conservative approach as larger mean values give more damage.
Similarly the 3 cycles which have a range of 300 are put into the bin with the higher range. In
either case the approach of using a matrix leads to an approximation of the answer derived from
using the time history directly. It is also important to ensure that the outer limits of the histogram
are sufficient to contain all potential cycles as information on cycles which fall outside the matrix is
lost.

2 - Range only
This option ignores the mean entirely. In our simple case the histogram becomes single
dimensional as follows :

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 101

Table 6:
Range
0:150 150:300 300:450 450:600 600:750 750:900 900:1050 1050:1200
0 0 3 0 0 1 1 0

If the mean is to be ignored this approach makes sense as the data storage saved can be used to
refine the resolution of the bins in the range axis. However it is normally recommended to save the
mean information as this may be required later and once lost cannot be retrieved.

3 - Max-Min
This option bins the maximum and minimum values directly, e.g.

Table 7:
Max
0:150 150:300 300:450 450:600 600:750 750:900 900:1050 1050:1200
0:150 0 0 0 0 0 0 0 1
150:300 0 0 0 0 0 0 0 1
300:450 0 0 0 0 0 0 0 0
Min 450:600 0 0 0 0 0 3 0 0
600:750 0 0 0 0 0 0 0 0
750:900 0 0 0 0 0 0 0 0
900:1050 0 0 0 0 0 0 0 0
1050:1200 0 0 0 0 0 0 0 0

The values in this matrix must by definition populate only a triangular region above the diagonal
where max>=min. The lower diagonal must therefore contain zeroes. This area can be utilised in
the from-to matrix.

4 - From-To
This matrix utilises the lower area to include directional information. In the example above, the
matrix becomes :

Table 8:
To
0:150 150:300 300:450 450:600 600:750 750:900 900:1050 1050:1200
0:150 0 0 0 0 0 0 0 0
150:300 0 0 0 0 0 0 0 0
300:450 0 0 0 0 0 0 0 0
From 450:600 0 0 0 0 0 1 0 0
600:750 0 0 0 0 0 0 0 0
750:900 0 0 0 2 0 0 0 0
900:1050 0 0 0 0 0 0 0 0
1050:120 1 1 0 0 0 0 0 0
0

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 102

Cycles with a “down” direction (“hanging”) populate the lower diagonal and the “up” (“standing”)
cycles populate the upper diagonal. The outside loop, which goes from the maximum in the series
to the minimum is placed according to which has the absolute largest value. In this case the
maximum has the absolute largest and so the entry is in the lower diagonal.

Calculation Procedure
When calculating strain-life fatigue lives from a pre-counted rainflow matrix there are two
problems :

In a range-mean or single-sided max-min rainflow matrix the sequence of cycles is completely lost
and it is therefore not possible to correctly define the hysteresis loops. This is not important unless
mean stress is a consideration.
The actual value of the maximum and minimum is lost and so must be approximated.

The second problem can be resolved in one of two ways. Two values can be calculated which
correspond to the worst and best cases and these, together with their average, can be reported to
give a results envelope. The worst case occurs when the range and mean ( or max/min ) are at
their largest corner, the best case when the values are selected from the smallest corner, e.g. for
the outside loop in the range-mean matrix above the worst case would have a range of 1050 and
a mean of 600, the best case a range of 900 and a mean of 450. In a low resolution matrix like this
the difference is large but in a 128x128 matrix this difference would be much smaller and the
impact on the life much less. The second way would be simply to take the middle value from the
bin, e.g. in the case above the range would be 975 and the mean 525. Both methods are
approximations and neither is “right”.

The first problem is resolved according to whether the matrix contains the directional information.
The range-mean and max-min matrices do not contain any directional information and so we do
not know whether the loops hang or stand. Therefore we adopt a best-worst approach, where the
worst case involves hanging all cycles from the top of the outside loop ( maximising the mean
stress) and the best case involves standing all cycles on the bottom of the outside loop. This
corresponds to the time histories identified below :

1200

1000

800
Actual
600 Worst
Best
400

200

0
1 2 3 4 5 6 7 8 9 10 11

nCode n 1st Page Back Main Menu

FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 103

These time histories have exactly the same values but in a different order. The sequence effect
leads to different mean stress values for the smallest cycles. The loops for the best and worst
cases look like this:
Cross Plot of Data : BEST
: Material : Generated UTS : 100
Factor : 1 Kf : 1
Life : 8.08E4 Repeats

60

40
Stress(MPa)

20

-20

-40
0 200 400 600 800 1000
Strain(uE)

Hysteresis loops for file BEST

Figure 63 The best case

Cross Plot of Data : W ORST

Material : Generated UTS : 100
Factor : 1 Kf : 1
Life : 7.72E4 Repeats

60

40
Stress(MPa)

20

-20

-40
0 200 400 600 800 1000
Strain(uE)

Hysteresis loops for file WORST

Figure 64 The worst case

The average damage, and hence the average life, can then be calculated. When we have the
From-To matrix we have slightly more information, in that we know there are two hanging and one
standing cycles rather than three whose status is unknown.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 104

Cross Plot of Data : FROMTO

Material : Generated UTS : 100
Factor : 1 Kf : 1
Life : 7.84E4 Repeats

60

40
Stress(MPa)

20

-20

-40
0 200 400 600 800 1000
Strain(uE)

Hysteresis loops for file FROMTO

Figure 65 Average damage

However, this is still not totally accurate as one of the small cycles in reality hangs from an inner
loop.

In conclusion, the use of histograms to store rainflow data is not ideal, but by selecting a high
resolution matrix and storing directional information a reasonable answer can be obtained,
particularly if the cycles are well distributed and a good sample size is captured.

7.2. Stress-life Fatigue Analysis from a Rainflow Matrix

Types of rainflow matrix

The basic types of matrix are the same as for the strain-life approach. However the From-To
option is not typically required as non-linear behaviour is not modelled in the S-N approach.

Calculation Procedure
Because the sequence effect is not considered, the only approximation in this case is related to
using the bin corners and the same methods can be used to estimate the cycle range and mean,
i.e. take the best and worst cases from the bin corners, or use the mid-point of the bin.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 105

8. Auto elimination
A key influence on the length of time FE-Fatigue takes to process a model is the number of
nodes or elements. FE-Fatigue includes an auto elimination option to enable the user to
preselect nodes or elements on which to focus the analysis.
This method exploits the fact that for most structures the high stresses are concentrated in local
areas and large proportions of the model experience little or no stress. Auto-elimination uses the
magnitude of the stress due to each loadcase and the magnitude of the load to calculate an
approximate worst case stress for each node or element. This value of stress is then used to filter
out the lower stressed nodes/elements.

Auto elimination methods

The following auto elimination methods are available:
% of maximum stress – Eliminates the nodes or elements with a calculated stress less than n
percent of the model's maximum.
% of model to retain – Selects only the highest n percent of nodes or elements for analysis.
Nodes that are eliminated will have no results. So care must be taken if contour plotting these
results.
% to retain by group – To avoid the problem of only retaining elements from a single “hot-spot”
on the model, auto-elimination on a group basis retains the top n percent nodes/elements from
each material group defined in the FES file.

Calculation method
All auto-elimination methods use the combination of the Von Mises stress for each load case and
the range of each time history to produce a worst case superposition value for each element.
Mathematically, this total value can be expressed as:

σtotal = i=1 to nΣ[(αi/βi) . σvm,i . (Lmax,i - Lmin,i)]

where,

i is the loadcase number

n is the total number of loadcases

σvm,i is the Von Mises stress for loadcase i

(Lmax,i - Lmin,i) is the range or largest cycle of the loading time history associated with
loadcase i (maximum value subtract the minimum value)

αi is the scale factor for loadcase i (default 1.0)

βi is the 'divide by' factor for loadcase i (default 1.0)

This total value is used to form an approximate ranking of nodes or elements likely to be
damaged.
For the stress-based option, the program also requires a percentage value to be entered which
then removes all elements whose stress combination is less than that percentage of the
maximum.
For example, if a value of 10% is entered and the worst stress combination is 500 MPa, then all

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 106

elements whose combination was less than 50 MPa would be omitted from the calculation.
For the point-based options, the program also requires a percentage to be entered which is the %
of nodes or elements to be analyzed. For example for a 10,000 element model, if 10% is entered
then the 1,000 elements with the largest “total Von Mises” stress will be used in the calculation.
The auto-elimination method is very useful but should be used with caution, especially where a
large proportion of the model is being discarded, and where other factors such as mean stresses
may increase damage.

Spot weld auto elimination

To perform auto elimination, a ranking stress parameter is required for each spot weld. This
ranking stress can be expressed as:

∆σtotal = Σ ∆σ max,i
i=1 to n

where:
i is the loadcase number
n is the total number of loadcases
∆σmax,i is the maximum range of stress associated with each loadcase and loading history

∆σmax,i is calculated for each loadcase using the forces and moments from the .fes file, together
with the maximum and minimum values of the associated time history and any scaling factors.
The spot weld structural stress equations (see section 3.6.3) can be differentiated to find the
angles of maximum and minimum stress for each load case.
These angles, θ max,min (180 degrees apart) are calculated from:

θ = arctan ((σmax (Fy1) + σmax(Mx1) )/( σmax(Fx1) - σmax(My1)))

These angles are then substituted back into the structural stress equations and the ‘effective’
stress is calculated at θ max,min for the maximum and minimum values of each associated time
history. The bigger of the two ranges calculated at θ max,min is ∆σmax,i.
If the FATFE preferences are used to select ranking without consideration of the loading history,
the extreme values of stress used to calculate ∆σmax,i will be based on the forces and moments in
the .fes file scaled by +1 and –1.
The ranking parameter for each spotweld is the larger of the values of ∆σtotal for sheet 1 and sheet
2 (the nugget is neglected for ranking purposes). When 3 sheet spot welds are considered, the
ranking value of the middle sheet is neglected. If one element of a 3 sheet weld is selected for
analysis, the other will also be included.

nCode n 1st Page Back Main Menu

 FE-Fatigue Rel. 5
1st Page Back Main Menu
Theory Page 107

9. Virtual strain gauges

This feature allows the user to place a virtual strain gauge on the part being analysed. Element or
node strains that are calculated in the FE solver are exported such that the strain output will
emulate a strain gauge rosette at the same location on the actual part. This provides an improved
facility for road to CAE correlation and calibration to occur and improves the confidence of FE
models by a more direct comparison of strain results.
The following types of gauge are supported:
• Single
• Rectangular (45 degrees)
• Delta (60 degrees)
• Tee (90 degrees)
It is recommended that virtual strain gauge data is exported from shell elements rather than
nodes. Shell element data will typically be in local coordinate system which will help the user to
compare orientation with actual strain gauge positioning.

3D to 2D conversion
If a file contains 3D stresses or strains, these must be converted to surface resolved strains. At the
moment the existing eigenvector method must be used, although this is not 100% reliable. If an
error is generated in the course of calculating the eigenvectors, no output can be produced. Note
that the X-axis in the plane of the surface is effectively arbitrary after conversion from 3D data.
The eigenvector method transforms stresses, so stress to strain conversion must be used before
and after as necessary.

Angle calculation
To calculate the strain at an angle T to the X axis, the following equation should be used:

E(T) = (Ex * cos2(T)) + (Ey * sin2(T)) + (Exy * sin(T) * cos(T))

Where Ex, Ey and Exy are the X, Y and shear strains.

Elastic-plastic conversion
This is a two-pass operation, where the elastic outputs are created first and then converted to
elastic-plastic using the Mroz-Garud cyclic plasticity model described in section 3.3.8.

nCode n 1st Page Back Main Menu