15th International LS-DYNA® Users Conference NVH

Advances in Fatigue Analysis with LS-DYNA

Yun Huang, Zhe Cui
Livermore Software Technology Corporation

Abstract
Fatigue analysis is critical to the design and optimization of metal structures and components. This paper reviews the recent
development in fatigue analysis with LS-DYNA. Both frequency domain and time domain fatigue solvers have been implemented to LS-
DYNA. They can be used towards different simulation situations. Some examples are provided in this paper, to illustrate how to use
these fatigue analysis methods. The plan on future development of the fatigue solvers in LS-DYNA, is also discussed.

Introduction

Fatigue analysis is important in many industry areas, including automotive industry, aircraft industry and pressure
vessel design. This is because for many structures and components, they are subjected to continuous and cyclic
loading during their service life and fatigue failure is the dominant failure mode in these cases.

Since version 971 R6, a set of fatigue solvers was provided in LS-DYNA in frequency domain, including random
vibration fatigue (*FREQUENCY_DOMAIN_RANDOM_VIBRATION_FATIGUE) and SSD (Steady State
Dynamics) fatigue (*FREQUENCY_DOMAIN_SSD_FATIGUE).

Sometimes users need to run fatigue analysis in time domain. For example, they may need to study the fatigue of
gun pipes due to cyclic thermal loading. Another example is the fatigue of the aircraft turbine blades, which are
made of alloy and subject to combined loads of high temperature and high air pressure. In these cases, the loading
level is high and the plastic deformation is observed in the materials. So it is more appropriate to use low cycle
fatigue criterion and solve the problems in time domain. To help these uses, we developed time domain fatigue
solvers in LS-DYNA. A new keyword *FATIGUE is introduced to run the time domain fatigue. This keyword
provides user with the opportunity to perform fatigue and durability analysis following the transient analysis in
LS-DYNA. The transient analysis, which provides stress or strain cycles for fatigue analysis, can be linear or
nonlinear, can use a wide variety of material models (elastic, plastic, ...), and can work with a variety of loading
conditions (thermal, mechanical, ...).

The results of fatigue analysis, in both frequency domain and time domain, are provided in the binary plot database
D3FTG, which can be easily accessed by LS-PrePost®.

Some examples are provided in the paper to demonstrate the fatigue analysis capabilities of LS-DYNA, and show
how to use these frequency domain or time domain fatigue solvers towards different cases. Some of the examples
mentioned in this paper were taken from previous publications or presentations [1].

June 10-12, 2018 1

15th International LS-DYNA® Users Conference NVH

Database and post-processing for fatigue analysis

The fatigue analysis results are saved in a binary plot database D3FTG. A keyword
*DATABASE_FREQUENCY_BINARY_D3FTG is used to activate the output of D3FTG database. Currently
this keyword is also used even if it is a time domain fatigue analysis. D3FTG can be easily accessed by
LS-PrePost. A typical GUI for D3FTG is shown in Figure 1.

For random vibration fatigue analysis, the following results are provided in D3FTG:

1. Cumulative damage ratio

2. Expected fatigue life
3. Zero-crossing frequency with positive slope
4. Peak-crossing frequency
5. Irregularity factor
6. Expected fatigue cycles

Figure 1. GUI in LS-PrePost for fatigue analysis results

For fatigue analysis based on *FREQUENCY_DOMAIN_SSD, and fatigue analysis in time domain, only the
cumulative damage ratio is provided in D3FTG. The other fringe components are zero.

More details about the post-processing of the fatigue analysis results can be found in [1].

June 10-12, 2018 2

15th International LS-DYNA® Users Conference NVH

Frequency domain fatigue analysis with LS-DYNA

A series of frequency domain analysis features have been implemented to LS-DYNA, since version 971 R5. A
comprehensive review of the frequency domain features, and their applications in auto NVH and acoustic
problems can be found in [2, 3].

Solutions for both deterministic and non-deterministic vibration are provided in frequency domain. For non-
deterministic vibration, one can define the loads in terms of PSD (Power Spectral Density) and run random
vibration analysis (*FREQUENCY_DOMAIN_RANDOM_VIBRATION); for deterministic vibration, one can
run steady state dynamics (*FREQUENCY_DOMAIN_SSD) and get the response for each excitation frequency.

With the knowledge of stress cycles from the vibration analysis, one can proceed to fatigue analysis, upon
receiving materail’s fatigue properties (S-N fatigue curve). A Miner’s rule of cumulative damage ratio, is used:

E (D ) = ∑
ni
(1)
i Ni

Where, E(D) is the expected cumulative damage ratio, ni is the number of cycles at stress level Si, and Ni is the
number of cycles for failure at stress level Si, given by a material’s S-N curve.

Fatigue analysis in random vibration environment

In this section, we describe the fatigue analysis method in random vibration environment.

With random vibration analysis, PSD of stress components are provided (and saved in D3PSD). Various methods
(Three-band method, Dirlik method, etc.) have been developed to calculate the PDF (Probability Density
Function) of stresses. The number of stress cycles can be calcuated for a given time of exposure [4].

The material’s S-N curve can be defined by *MAT_ADD_FATIGUE. More details about defining the materials‘
S-N curve can be found in [4].

The example in this section considers a metal bracket shown in Figure 2. It is constrained to a shaker table through
the two small holes (marked as red in Figure 2) for a random vibration test.

Figure 2. A bracket model constrained to shaker table

June 10-12, 2018 3

15th International LS-DYNA® Users Conference NVH
The S-N curve is represented by the equation

N ⋅ S 2.5 = 1.56 × 10 8 (2)

where, N is the number of cycles for failure (or endurance) under the stress level S.

The model is subject to base acceleration. The acceleration PSD curve is shown in Figure 3.

1.E-02
Acl. PSD (g^2/Hz)

1.E-03

1.E-04

1.E-05

1.E-06
10 100 1000 10000

Frequency (Hz)
Figure 3. The base acceleration PSD curve

The duration of excitation is 1 hour. For the first loading case, we consider the acceleration in z-direction only.

Figure 4. Cumulative damage ratio for the case Figure 5. Safe / Failed zone for the case of z-
of z-acceleration loading acceleration loading
Dirlik method is employed to perform fatigue analysis in this example. Von-mises stress is used as the stress
index.

Figure 4 shows the cumulative damage ratio plot under the z-acceleration excitation. The peak value of the
cumulative damage ratio is around 2.48. This indicates that the model will fail due to fatigue in the test. The peak
value appears at the edge of one constrained hole. This suggests that the initial crack will take place at that
location.

Using the keyword *INITIAL_FATIGUE_DAMAGE_RATIO, one can accumulate the damage on a structure
from multiple loading cases. For example, the same structure shown in Figure 2 can be subject to base acceleration
in x-, y- and z-direction sequentially. When we run fatigue analysis for the last loading case (e.g. acceleration in
z-direction), we can define the keyword *INITIAL_FATIGUE_DAMAGE_RATIO and set the path and name
for the binary databases for the previous fatigue results (due to acceleration in x- and y-directions). The combined
cumulative damage ratio plot, due to the three loading cases, can be found in Figure 6.

June 10-12, 2018 4

15th International LS-DYNA® Users Conference NVH

Figure 6. Cumulative damage ratio for x-, y- Figure 7. Safe / Failed zone for x-, y- and z-
and z-acceleration loading cases acceleration loading cases

Figure 7 shows the safe / failed zone as the final result of the three loading cases. As we can see, the hole edge
area has a higher chance for failure than other areas in this test.

Sometimes it is important to account for the effect of pre-stress or nonzero mean stress on fatigue analysis. For
random vibration analysis, one can use a predefined initial stress as the mean stress. The initial stress is introduced
by keywords *INITIAL_STRESS_SHELL, *INITIAL_STRESS_SOLID, etc. To define the mean stress
correction method, use the keyword *FATIGUE_MEAN_STRESS_CORRECTION. The following mean stress
correction methods are available in LS-DYNA.

σa
Goodman method S= (3)
1−σm /σu
σa
Soderberg method S= (4)
1−σm /σ y
σa
Gerber method S=
1 − (σ m / σ u )
2 (5)

Goodman tension-only method, and Gerber tension-only method are also available. For the two tension-only
methods, LS-DYNA performs mean stress correction for tensile mean stress only, using the equations (3) and (5).

In the equations (3)-(5), S is the corrected stress magnitude or alternating stress; σ a is the original stress
magnitude or alternating stress; σ m is the mean stress; σ u is the ultimate tensile strength and σ y is the yield
strengh of the material.

June 10-12, 2018 5

15th International LS-DYNA® Users Conference NVH

For the model shown in Figure 2, an initial stress (mean stress) state is assumed (see Figure 8)

Figure 8. Initial Von mises stress (mean stress)

for the loading case with z-acceleration, the updated cumulative damage ratio results, with mean stress correction
by Goodman’s method (equation 3), and by Soderberg’s method (equation 4), are shown in Figures 9 and 10.

Figure 9. Cumulative damage ratio for z- Figure 10. Cumulative damage ratio for z-
acceleration PSD loading only (use Goodman’s acceleration PSD loading only (use
equation for mean stress correction) Soderberg’s equation for mean stress
correction)

Table 1 shows the maximum cumulative damage ratios with and without mean stress correction.

Mean stress correction Maximum damage ratio

method
No mean stress correction 2.480
Goodman’s equation 2.767
Soderberg’s equation 4.726
Gerber’s equation 2.510

Table 1: Maximum damage ratio with and without mean stress correction

One can see, with the mean stress correction, the damage ratio of the structure is increased, which means that
the fatigue life of the structure becomes shorter due to the existence of the mean stress.

June 10-12, 2018 6

15th International LS-DYNA® Users Conference NVH

Fatigue analysis in steady state vibration environment

Sometimes we have steady state vibration environment, for example, the vibration in a sine sweep test. With the
keyword *FREQUENCY_DOMAIN_SSD_FATIGUE one can run fatigue analysis based on steady state
vibration. This approach is also based on Miner’s rule of cumulative damage and we sum up the fatigue damage
from different stress level.

The example shown here is about the fatigue analysis of a solid structure (a simplified auto front bumper model),
in a steady state vibration test.

Figure 11. A front bumper model in steady state vibration

The bumper is subjected to continuous vibration from ground excitation during driving. It is assumed that the
bumper is constrained to auto frame by the edge of the two holes (see the red nodes shown in Figure 11).

The material’s S-N curve is defined *MAT_ADD_FATIGUE using a curve shown in Figure 12.

1.E+02
S: stress range (MPa)

1.E+00

1.E-02

1.E-04

1.E-06
1.E+03 1.E+05 1.E+07 1.E+09
N: number of cycles

Figure 12. S-N curve for the bumper model

June 10-12, 2018 7

15th International LS-DYNA® Users Conference NVH

The base acceleration spectrum is given in Table 2 (see below).

Frequency (Hz) Amplitude (g) Duration (sec.)

200 1.0 600
300 1.0 600
400 1.0 600
500 1.0 600
600 1.0 600
700 1.0 600
800 1.0 600
900 1.0 600
1000 1.0 600
1100 1.0 600
1200 1.0 600
1300 1.0 600
1400 1.0 600
1500 1.0 600
Table 2. Base acceleration spectrum

The cumulative damage ratio fringe plot can be found in Figure 13.

Figure 13. Cumulative damage ratio plot for the bumper

To locate the failed zone quickly, one can use the “Safe / Failed zone” check box and get the failed zone indicated
by red elements as below (Figure 14).

Figure 14. Safe / Failed zone under the steady state vibration

One can see that the failed elements are almost the same elements which are constrained to shaker table directly.
For those elements we expect higher stress concentration due to the constraints and that is the reason for the higher
chance for fatigue failure at the same location.
June 10-12, 2018 8
15th International LS-DYNA® Users Conference NVH

Time domain fatigue analysis with LS-DYNA

For many applications, the loading conditions are defined in time domain and the response of the structures or
parts are also computed in time domain. This is true for door slam, and for thermal loading on structures, and for
many other cases where the nonlinear effect is significant and the problem cannot be linearized. To provide fatigue
solution for those cases, time domain fatigue solvers are being implemented in LS-DYNA. The solvers can be
put into 2 categories: one is based on stress index and the other is based on strain index.

A new keyword for general time domain fatigue analysis has been implemented

*FATIGUE
$# pid ptype
1 20
$# dt
0.01
$# strsn index restrt
0

To do mean stress correction, the keyword *FATIGUE_MEAN_STRESS_CORRECTION is needed.

A simple way to run fatigue analysis in time domain is to use the stress/strain data saved in ELOUT (see
*DATABASE_ELOUT and *DATABASE_HISTORY_SHELL, *DATABASE_HISTORY_SOLID, etc.).

To use ELOUT, one just need to add option _ELOUT to *FATIGUE and skips the first two cards in this keyword.
To save strain results to ELOUT, one can set STRFLG = 1 in *DATABASE_EXTENT_BINARY.

The Material fatigue properties can be defined by keyword *MAT_ADD_FATIGUE (for SN curve) or
*MAT_ADD_FATIGUE_EN (for EN curve). More details of these keywords can be found in LS-DYNA Users’
Keyword manual.

Stress-based fatigue analysis

For stress-based time domain fatigue solver, a rain flow counting method is adopted to get the number of stress
cycles for each stress level. One can also make mean stress correction based on the Goodman method, the
Soderberg method, the Gerber method and tension-only version of the Goodman method, and the Gerber method
(see keyword *FATIGUE_MEAN_STRESS_CORRECTION).

The example concerning the stress based time domain fatigue analysis is a metal pipe model (Figure 15), subject
to temperature oscillation (see Figure 16). For real life application, this pipe can be cannon or gun pipe and the
temperature oscillation can be due to continuous gun fires. The thermal stress cycle due to the temperature
oscillation induces fatigue damage to the metal. After sufficient number of gun fires, the cumulative damage ratio
can reach 1 and indicates the end of the service life of the gun pipe.
The piple is modelled by *MAT_ELASTIC_PLASTIC_THERMAL. And the thermal loading is defined by
*LOAD_THERMAL_LOAD_CURVE.

June 10-12, 2018 9

15th International LS-DYNA® Users Conference NVH

Figure 15. The metal pipe model Figure 16. The temperature load curve

The material’s fatigue property is defined by *MAT_ADD_FATIGUE. The SN curve looks like

Figure 17. SN fatigue curve of the material

The cumulative damage ratio plot is given in Figure 18. One can see that the peak value of the cumulative damage
ration appears in the middle part of the pipe, which agrees with the location for peak value of the von mises stress
(Figure 19).

Figure 18. Cumulative damage ratio plot Figure 19. Distribution of von mises stress at
termination time

June 10-12, 2018 10

15th International LS-DYNA® Users Conference NVH

Figure 20 shows the xy- shear stress cycles at an element in the middle of the pipe. One can see a lot of stress
cycles have taken place in the 1000 seconds period.

Figure 20. The xy-stress history at one element

Strain-based fatigue analysis

For strain-based fatigue analysis, we use the local strain life equation
∆ε σ ′f
= (2 N f )b + ε ′f (2 N f )c
2 E
(6)

With Smith-Watson-Topper mean stress correction, equation (6) is changed to

∆ε
σ max =
(σ ′f )2
(2 N f )2b + σ ′f ε ′f (2 N f )b+c (7)
2 E

With Morrow’s mean stress correction, equation (6) is changed to

∆ε σ ′f − σ m
= (2 N f )b + ε ′f (2 N f )c (8)
2 E

Here is the definition of the variables used in the equations

∆ε strain amplitude
N endurance, or number of cycles for failure
b the fatigue strength exponent (Basquin’s exponent)
σ ′f the fatigue strength coefficient
σ max maximum stress
σm mean stress
c the fatigue ductility exponent (the Coffin-Mson exponent)
ε ′f the fatigue ductility coefficient, which is the plastic strain amplitude at 2 N f = 1
E the elastic modulus

June 10-12, 2018 11

15th International LS-DYNA® Users Conference NVH

If pure elastic materials are used in the stress-strain cycle computation (e.g. using an elastic finite element
analysis), Neuber’s rule is used to convert the elastic stress/strain to local true stress/strain

σε = σ eε e (9)

Where, σ and ε are the local stress and strain, σ e and ε e are the elastic stress and strain.

For illustration purpose, we consider a bracket model shown in Figure 21. The bracket is modeled by
*MAT_PIECEWISE_LINEAR_PLASTICITY.

Figure 21. A bracket model with sinusoidal nodal force excitation

The model is constrained at the two rings. The sinusoidal nodal force is defined as

F (t ) = 5 sin (1000 t ) N (10)

In the example, the material’s fatigue constants are defined as

E = 209,000 MPa ; K ′ = 1230 MPa ; n′ = 0.161 ; σ ′f = 1006 MPa
b = −0.0809 ; ε ′f = 1.104 ; c = −0.6207

Figure 22. Effective strain at lower layer Figure 23. Effective strain at upper layer
integration points integration points

The cumulative damage ratio due to the sinusoidal nodal force for 1 second is computed and plotted in Figure 24.
The location for the peak value of the cumulative damage ratio matches well with the location for highest strain
values. Figure 24 shows the strain history at the element which has the highest cumulative damage ratio.

June 10-12, 2018 12

15th International LS-DYNA® Users Conference NVH

Figure 24. Cumulative damage ratio Figure 25. Effective strain at upper layer
integration points

Conclusion and future work

Some basic frequency domain and time domain fatigue analysis methods have been implemented to LS-DYNA.
For frequency domain fatigue analysis, the index variable is stress; for time domain analysis, both stress and strain
can be used as the index variable. These methods provide user the opportunity to run fatigue and durability
analysis on the structures, following the vibration analysis in frequency domain or time domain. The results not
only provide user an estimate of the fatigue life of the structures, but also suggest the location of “hot spot” for
future revision or optimization.

Regarding future work, firstly, more testing and validations are needed, to check the accuracy, reliability of the
current fatigue solvers. Secondly, an approach to consider multi-axial stress /strain state is needed, for more
realistic simulation of the fatigue failure process. The current methods are based on the stress or strain invariants
(e.g. the Von mises stress / strain, or the maximum shear stress / strain). We are interested in exploring the new
methods based on critical plane approach.

References
[1] Yun Huang, Zhe Cui, Philip Ho, Chengju Zhang, Application of LS-DYNA in Structural Fatigue Analysis, Proceedings of the 14th
Internatinal LS-DYNA Users Meeting & Conference 2016, June 12-14, 2016, Detroit.
[2] Yun Huang, Zhe Cui, Recent updates in LS-DYNA frequency domain solvers, proceedings of the 10th European LS-DYNA
Conference 2015, Wurzburg, Germany.
[3] Prasanna S. Kondapalli, Tyler Jankowiak, Yun Huang, Application of LS-DYNA to NVH Solutions in the Automotive Industry.
Proceedings of the 14th International LS-DYNA Users Conference 2016, June 12-14, 2016, Detroit.
[4] Arnaud Ringeval, Yun Huang, Random Vibration Fatigue Analysis with LS-DYNA. Proceedings of the 12th International
LS-DYNA Users Conference.

June 10-12, 2018 13