Você está na página 1de 16

British Journal of Engineering and Technology

ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

CHOOSE THE BEST ELEMENT SIZE TO YIELD ACCURATE FEA RESULTS


WHILE REDUCE FE MODELS COMPLEXITY
Yucheng Liu, PhD, PE
Assistant Professor and Graduate Coordinator
Department of Mechanical Engineering
University of Louisiana at Lafayette
Lafayette, LA 70504, USA

ABSTRACT

n finite element models, element size is a critical


issue which closely relates to the accuracy of the
finite element models while directly determines
their complexity level. Thus, a primary problem in creating
finite element model is to choose appropriate element size
which can yield correct simulation results while reduces
the models complexity as much as possible. This paper
presents a systematic study on finding the effects of finite
element size on the accuracy of numerical analysis results,
based on which brief guidelines of choosing the best
element size in finite element modeling are provided. Static,
modal, and impact analysis are involved in this study to
discuss the effects of element size in numerical analysis.

Keywords: Element size, mesh density, static analysis,


impact analysis, modal analysis

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


1. Introduction
In finite element analysis (FEA), the accuracy of the FEA results and requested computing time are
determined by the finite element size (mesh density). According to FEA theory, the FE models with fine
mesh (small element size) yield to highly accurate results but may take longer computing time. On the
contrary, those FE models with coarse mesh (large element size) may lead to less accurate results but do
save more computing time. Also, small element size will increase the FE models complexity which is only
used when high accuracy is required. Large element size, however, will reduce the FE models size and is
extensively used in simplified models in order for providing a quick and rough estimation of designs. Due to
its importance, in generating FEA models, the foremost problem is to choose appropriate elements size so
that the created models will yield accurate FEA results while save as much computing time as possible. The
objective of this paper is to present guidelines for choosing optimal element size for different types of finite
element analyses. In order to achieve that goal, in this study, a series of static, modal, and impact analyses
were performed on thin-walled beam and plate models to reveal the effects of the element size on the
accuracy of the FEA results. An explicit solver, LS-DYNA, was used for modeling and analyses involved in
this work [1]. The paper is organized as follows: section 2 provides the background for this study, which
briefly reviewed previous related literature; section 3 explains the theoretical fundamentals of the applied
finite element; sections 4 to 6 discuss the effects of element size on the accuracy of static analysis, impact
analysis, and modal analysis, respectively; section 7 concludes the whole paper with guidelines of choosing
the best mesh density for different type of FEA presented.
2. Background
A number of investigators have studied the effects of elements size on the accuracy of numerical results of
different types of analysis and important conclusions have been drawn from previous research. Brocca and
Bazant [2] presented a finite element study of the size effect of compressive failure of geometrically similar
concrete columns of different sizes. It was observed from their analyses that the increasing elements size
caused reduction in nominal strength. However, a quantitative analysis showing the relationship between the
elements size and the nominal strength was still needed. Ashford and Sitar [3] evaluated the accuracy of the
computed stress distribution near the free surface of vertical slopes as a function of the element size. A
parametric study was carried out comparing stresses computed using FEA to those obtained from a physical
model composed of photoelastic material. It was found that for a slope height H, an element height of H/10
is adequate for the study of stresses deep within the slope. However, for cases where tensile stresses in the in
the vicinity of the slope face which are critical, element heights as small as H/32 are necessary. Saouma et
al. [4] discussed size effect in nonlinear finite element analysis with a metal-reinforced ceramics composite
material. In their study, a size effect investigation was numerically performed and the range of crack sizes
was presented, for which linear elastic fracture mechanics, nonlinear fracture mechanics or plasticity-based
models were applicable. Masakazu [5] conducted a numerical analysis of the size effect on the shear
strength of reinforced concrete (RC) beams through tow-dimensional nonlinear FEA, which was applied to
the simulation of RC members in flexure and shear failure. Three RC beam models with different sizes were
analyzed and the size effects on strain of reinforcement, strain of concrete, descriptive mode, and crack
situation were observed from the simulation and discussed. Perillo-Marcone et al. [6] assessed the effect of
mesh density on material property discretization and the resulting influence on the predicted stress
distribution through analyzing a three-dimensional, quantitative computed tomography based FE model of a
proximal implanted tibia. Significant variations were observed in the modulus distributions between the

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


coarsest and finest mesh densities. Poor convergence of the material property distribution occurred when the
element size was significantly larger than the pixel size of the source computerized tomography (CT) data.
From those results, they found an optimal element size of 1.4 mm on the contact surfaces which was enough
to properly describe the stiffness, stress and risk ratio distributions within the bone for that particular case.
Zmudzki et al. [7] discussed the influence of mesh density on the results of FE model analysis of mechanical
biocompatibility of dental implants. It was found that the increasing of mesh density leads to an
overestimation of loading stresses values and furthermore to an unjustified increase of pillars diameter. At
the other hand, too large elements might lead through an underestimation of loading stress level, to
overloading atrophy of bone tissue or to implant loss. From that work it can be found that a guideline of
choosing appropriate element size for certain finite element analysis is highly demanded. Roth and Oudry
[8] touched the influence of element size on the accuracy of dynamic analysis results and they mentioned
that for dynamic analysis, the minimum number of element required for correct simulation is according to
the loading case and material properties. Li et al. [9] investigated the sensitivity of the structural responses
and bone fractures of the ribs to mesh density in order to provide guidelines for the development of FE
thorax models used in impact biomechanics. It was demonstrated in their research that rib FE models
consisting of 2000-3000 trabecular hexahedral elements (weighted element length 2-3 mm) and associated
quadrilateral cortical shell elements with variable thickness more closely predicted the rib structural
responses and bone fracture force-failure displacement relationships observed in the experiments. Based on
the previous work and achievements, a systematic investigation is conducted here to fully discuss the size
effect on simulation accuracy of static, modal, and impact analysis for fundamental structural components
such as plates and beams.
3. Finite Element Formulation
All finite element models involved in this study were meshed with the full integration shell element: 4-node
Belytschko-Tsay shell element with five integration points through the thickness. Such element is based on a
combined co-rotational and velocity-strain formulation, among which the co-rotational portion of the
formulation avoids the complexities of nonlinear mechanics by embedding a coordinate system in the
element, and the choice of velocity-strain formulation facilitates the constitutive evaluation. In that element,
the velocity of any point in the shell is determined as:
(1)
v v m ze3
where vm is the velocity of the mid-surface, is the angular velocity vector, and z is the distance along the
fiber direction (thickness) of the shell element.
The velocity strain displacements need to be evaluated at the quadrature points within the element and
standard bilinear nodal interpolation is used to define vm, , and the elements coordinates ( x , y , z ). The
velocity strains are found to be
N
N
N N
N
N
N
N
v y z
x , 2 d xy
d x
v x z
y , dy
v x
v y z
y
x
x
x
y
y
y
x
x
y
N
N
v z Nx
(2)
2d xy
v z N y , 2d yz

x
y
In Eqn. (2), N is the shape function of the element. Constitutive evaluations are then obtained by using the
presented velocity-strain displacement relations.

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


4. Static Analysis
Static analyses were performed on a rectangular steel plate with the dimension 300 mm 200 mm and a
thickness of 3 mm. Material properties of the steel are listed in Table 1. During the analyses, one end of the
plate was fully constrained and a 1 Nm moment was applied at the other end for a duration of 1 second. 10
time steps were used to record the data so that 10 data points were collected during the analysis. A series of
FE models were generated for that plate whose long side was meshed from 2 (coarsest mesh) to 160 (finest
mesh) divisions (Fig. 1). Von mises stress and bending deformation yielded from each model were
calculated and compared to study the influence of element size on the static analysis results. Static analysis
results and comparisons are listed in Table 2. In that table, it is assumed that the FE model with the finest
mesh generated the most accurate results and percentage approximate errors were calculated by comparing
other results to the most accurate ones. It also needs to be mentioned that the using of full integration
Belytschko-Tsay shell element can effectively prevent the shear locking, which usually occurs in lower
order elements when those elements are subjected to bending.
Several observations were made by comparing those results. (1) The errors of bending deformation are far
lower than the errors of von Mises stress. According to FEA theory, stresses are not predicted as accurately
as the displacements because they are calculated from the displacements and it is assumed that the stresses
are constant over the element. (2) The difference of von Mises stress generated from the model with 10
elements along the long side of the plate and from the finest mesh model is less than 1%, which is
acceptable in engineering simulation. However the computing time for the coarse mesh model is only 3 sec,
which is less than 1/40 of the time cost by the finest mesh model. It can also be observed from Fig. 2 that
when the number of elements on the long side is higher than 60, the increase of mesh density does not
significantly improve the accuracy of von Mises stress any more. Such phenomenon was also observed in
comparing other static analysis results. (3) Fig. 3 shows the stress distribution and Fig. 4 compares the
bending deformation yielded from the coarsest mesh model (with 2 divisions) and the finest mesh model
(with 160 divisions). From Fig. 4 it can be found that even the coarsest mesh model generated a bending
deformation close to the finest mesh model (error = 0.22%), it failed to display a smooth and continuous
bending mode because its less number of elements. An FE model with finer mesh is still needed to correctly
simulate the bending behavior of the steel plate. Similar conclusion can be drawn from the Fig. 3. (4) It can
be concluded that for static analysis, the FE model whose longest side is meshed by 10 elements can give us
optimal combination of accuracy and efficiency.

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


Table 1. Steel material properties
Material Properties
Youngs modulus

207GPa

Density

7830kg/m3

Yield stress

200MPa

Ultimate stress

448MPa

Hardening modulus

630MPa

Poissons ratio

0.3

Table 2. Static analysis results and comparisons


# of division
2
5
10
20
30
40
50
60
70
80
90
100
120
140
160

von Mises stress


6.290 MPa
6.370 MPa
6.580 MPa
6.570 MPa
6.607 MPa
6.613 MPa
6.607 MPa
6.620 MPa
6.621 MPa
6.616 MPa
6.624 MPa
6.624 MPa
6.626 MPa
6.623 MPa
6.627 MPa

%
5.08
3.88
0.76
0.85
0.30
0.21
0.30
0.11
0.09
0.16
0.05
0.04
0.01
0.06

Bending deformation
0.6456 mm
0.6457 mm
0.6435 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm
0.6442 mm

%
0.22
0.23
0.11
0
0
0
0
0
0
0
0
0
0
0

Computing time
3 sec
3 sec
3 sec
3 sec
3 sec
4 sec
6 sec
7 sec
9 sec
13 sec
18 sec
26 sec
40 sec
66 sec
124 sec

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

(a)
(b)
Figure 1. FEA steel plate model (a) coarsest mesh (b) finest mesh
6.00
5.00

Error (%)

4.00
3.00
2.00
1.00
0.00
0

20

40

60

80

100

120

140

160

# of Mesh Elements along Plate Length

Figure 2. Element size vs accuracy for maximum von Mises stress

180

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

(a)
(b)
Figure 3. Stress distribution of (a) coarsest mesh model (b) finest mesh model

(a)
(b)
Figure 4. Bending of (a) coarsest mesh model (b) finest mesh model
5. Impact Analysis
After static analysis, impact analyses were carried out on a thin-walled steel beam with a square cross
section, whose dimension is 120 mm 120 mm and wall thickness is 3 mm. During the analyses, this beam
impacted a rigid wall at 15 m/s and buckled. A series of FE models were generated for that beam whose
axial direction was meshed from 2 (coarsest mesh) to 120 (finest mesh) divisions. The crash time was set as
0.01 seconds. Impact force, absorbed energy, and global displacement were computed for each FE model
and compared in Table 3, where the approximate error was calculated based on comparing each result to the
results yielded from the finest-meshed beam model. Fig. 5 displays the crushed model with coarsest mesh (2
divisions), medium mesh (60 divisions), and finest mesh (120 divisions). The effects of elements size on the
accuracy of important impact analysis results are plotted through Figs. 6 to 8. It needs to be mentioned that
hourglass modes (nonphysical, zero-energy modes of deformation that produce zero strain and no stress) did
not occur in these impact analyses. This is because that our models are meshed using Belytschko-Tsay shell

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


element with five integration points through the thickness, while the hourglass modes occur only in underintegrated (single integration point) shell elements. A default algorithm provided by LS-DYNA was also
selected to inhibit hourglass modes.
After comparing the listed results, following declaration can be made. (1) The errors generated during the
impact analyses were higher than those yielded from the static analyses. This is because that the impact
analysis dose involve fast, transient loading and it much more complicated than the static analysis. (2)
Similar to the static analysis, the finite element models predicted the displacement more accurate, which was
due to the same reason. (3) In general, the errors in predicting the absorbed energy were far lower than those
in estimating the impact force. A possible explanation is that the absorbed energy is related to the mass and
instantaneous velocity of the beam models, while the impact force is related to their mass and instantaneous
acceleration. There was no error in calculating the mass because different FE models faithfully represent the
volume and density of the steel thin-walled beam model. However, the velocity is calculated as the first time
derivative of the displacement and the acceleration is calculated as the first time derivative of the velocity,
therefore, the errors generated in calculating the velocity are transferred to the step of calculating the
acceleration. Due to this reason, it is understandable that the FE models predicted the absorbed energy more
accurate than the impact force. (4) In predicting the impact force, the beam model has to be meshed into 80
divisions longitudinally so that the approximate error would drop to below 10%, an accepted level in impact
simulation [10]. However, to reach the same accurate level, the beam model only has to be meshed into 20
divisions for predicting the absorbed energy and maximum displacement. (5) Similar to Fig. 4, it can be seen
from Fig. 5 that in impact analysis, the FE model with fewer number of elements could not correctly reflect
the real progressive buckling mode of the thin-walled steel beam. In other words, a certain number of finite
elements are required to correctly simulate the crash behavior and response of engineering structures during
the impact analysis. (6) In conclusion, in order to correctly simulate the crash process and predict important
impact results while saving as much computing time, the thin-walled beam model has to be meshed into 80
divisions along its axis. Table 3 also reveals that the optimal FE model with 80 divisions along its axis only
took less than 1/10 of the computing time requested by the finest mesh model.
Table 3. Impact analysis results and comparisons
# of division
2
5
10
20
30
40
50
60
70
80
90
100
120

Impact force
90.88 kN
61.39 kN
19.02 kN
10.09 kN
9.32 kN
9.46 kN
9.65 kN
9.82 kN
8.85 kN
5.26 kN
5.31 kN
5.35 kN
5.42 kN

%
1577
1033
251
86.2
72.1
74.7
78.1
81.4
63.3
2.86
2.02
1.21

Absorbed energy
41.48 kJ
39.99 kJ
22.88 kJ
20.19 kJ
19.85 kJ
19.85 kJ
17.33 kJ
19.97 kJ
19.84 kJ
19.86 kJ
19.45 kJ
19.32 kJ
18.90 kJ

%
119
106
21.0
6.83
5.00
5.02
8.30
5.62
4.95
5.08
2.87
2.19

Displacement
48.65 mm
50.03 mm
65.67 mm
85.5 mm
88.9 mm
91.36 mm
93.33 mm
97.37 mm
96.88 mm
96.33 mm
93.44 mm
94.43 mm
94.87 mm

%
48.7
47.3
30.8
9.88
6.32
3.70
1.62
2.63
2.11
1.53
1.51
0.47

Computing time
1 sec
1 sec
1 sec
2 sec
5 sec
9 sec
14 sec
21 sec
68 sec
111 sec
198 sec
357 sec
1198 sec

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

(a)

(b)

(c)

Figure 5. Deformed beam models with (a) coarsest mesh, (b) medium mesh, and (c) finest mesh

1800.00
1600.00
1400.00

Error (%)

1200.00
1000.00
800.00
600.00
400.00
200.00
0.00
0

20

40
60
80
100
120
# of Mesh Elements along Beam Model

140

Figure 6. Element size vs accuracy for maximum impact force

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


140.00
120.00

Error (%)

100.00
80.00
60.00
40.00
20.00
0.00
0

20

40
60
80
100
120
# of Mesh Elements along Beam Model

140

Figure 7. Element size vs accuracy for maximum absorbed energy

60.00
50.00

Error (%)

40.00
30.00
20.00
10.00
0.00
0

20

40

60

80

100

120

# of Mesh Elements along Beam Model

Figure 8. Element size vs accuracy for maximum displacement

140

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


6. Modal Analysis
Finally, modal analyses were conducted on above plate and beam models to determine their natural
frequencies and mode shapes during free vibration. It is common to use the FEA to perform this analysis and
the influences of the element size on the modal analysis results are discussed here. In this study, only the
lowest frequencies are listed and compared for each model because the lowest frequencies are related to the
most prominent modes at which the model will vibrate, dominating all the other higher frequency modes. In
performing the modal analysis, the steel plate was constrained on its short edge and the thin-walled steel
beam was constrained at its end. Table 4 only lists the lowest natural frequency calculated for each finite
element model. Fig. 9 plots the corresponding mode shapes of the coarsest meshed plate model and the
finest meshed plate model. Fig. 10 displays the mode shapes of the coarsest and finest meshed thin-walled
beam model.
From the displayed results it can be seen that developed FEA models correctly predicted the lowest natural
frequencies, with all the errors lower than 10%. The approximation error yielded from coarse models such as
the thin-walled plate model with 5 divisions and the thin-walled box beam model with 10 divisions are
below 1%. This is because the modal analysis results such as natural frequencies and mode shapes of a
model during free vibration are only depend on the models mass and stiffness matrix. An FEA model can
accurately predict its modal analysis results as long as it faithfully represents the models mass and stiffness.
Fig. 9 displays mode shapes of thin-walled plate models with different numbers of meshes. It is found that
the FEA models with different element size correctly plot the mode shape of the thin-walled plate and
predict the deflection of that mode shape. Table 5 lists deflections of the mode shapes respect to the lowest
natural frequency of several FEA models, which are very close to each other.
However, from Fig. 10 it can be seen that even though FEA thin-walled box models with coarse mesh can
correctly compute the natural frequency values, nevertheless, in order to closely simulate the mode shapes,
more finite elements are still required. It also deserves to be mentioned that the computing times for the FEA
models with different mesh density are not listed in Table 4. This is because the finite element modal
analysis is very fast. To finish one modal analysis, the coarsest mesh models took 1 second, the thin-walled
plate model with 140 divisions took 14 seconds and the thin-walled box beam model with 120 divisions took
12 seconds. Therefore, in deciding the optimal element size for the modal analysis, the computing time is
not an important issue.

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


Table 4. Modal analysis results and comparisons
Thin-walled steel plate
1st Natural
# of division
Difference %
frequency
2
27.0506 Hz
2.6351
5
27.6131 Hz
0.6105
10
27.7530 Hz
0.1069
20
27.7715 Hz
0.0403
30
27.7758 Hz
0.0248
40
27.7767 Hz
0.0216
50
27.7767 Hz
0.0216
60
27.7773 Hz
0.0194
70
27.7775 Hz
0.0187
80
27.7776 Hz
0.0184
90
27.7780 Hz
0.0169
100
27.7783 Hz
0.0158
120
27.7791 Hz
0.0130
140
27.7804 Hz
0.0083
160
27.7827 Hz

Thin-walled steel box beam


1st Natural
# of division
Difference %
frequency
2
19.4889 Hz
9.5836
5
17.5380 Hz
1.3860
10
17.9067 Hz
0.6871
20
17.7746 Hz
0.0557
30
17.7679 Hz
0.0933
40
17.7726 Hz
0.0669
50
17.7685 Hz
0.0900
60
17.7780 Hz
0.0365
70
17.7295 Hz
0.3093
80
17.7949 Hz
0.0585
90
17.8008 Hz
0.0917
100
17.7731 Hz
0.0641
120
17.7845 Hz
140
160

Table 5. Deflection of lowest frequency mode shapes of thin-walled plate models


# of division
2
5
10
100
160

Deflection
1654 mm
1683 mm
1689 mm
1691 mm
1691 mm

Difference %
2.19
0.47
0.12
0

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

(a) 2 elements

(c) 10 elements

(b) 5 elements

(d) 160 elements

Figure 9. Lowest frequency mode of plate with different element size

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013

(a) 2 elements

(c) 10 elements

(b) 5 elements

(d) 120 elements

Figure 10. Lowest frequency mode of beam models with different element size
7. Uniqueness of this Study
Several investigators have performed sound research in revealing the effects of mesh density on the FE
results of computer simulation of linear and nonlinear problems. DAmours et al. [11] presented an analysis
of crush response of hydroformed aluminum tubes. From the FE simulation, they found that in certain
instances, it may be better to use a fine mesh size for the hydroforming model and remap forming results to
coarser mesh sizes for crashworthiness models to save computational time. Aramayo [12] developed a FE
model of a Ford Explorer SUVwith arbitrary element size and size distribution. The general model was used
for frontal impact analysis with different scenarios, and the simulation results were verified by comparing
with experimental results of crash tests. In developing the FE model, the mesh density was parameterized in
different regions so that a fine mesh was employed in one half of the front of the object and a coarse mesh
elsewhere. Donadon and Iannucci [13] presented an objective algorithm for strain softening material models.
In order to evaluate the performance of the algorithm, a mesh sensitivity study was performed where a
simple coupon test simulation was performed on the virtual coupon models with six different mesh densities.
On comparing the structural response obtained using the different mesh types, it was found that the energy
dissipated in the formation of crack is mesh insensitive. Makino [14] examined the performance on the fine
mesh model and developed a 10-million shell elements car model in order to achieve the good accuracy in
crash analysis by LS-DYNA. Comparing with aforementioned research, the uniqueness of the present study
lies on following areas. (1) Despite the well known fact that fine mesh leads to higher accuracy while the
coarse mesh improves the computing efficiency, this study demonstrates how the element size affect the
accuracy and efficiency through a series of simulations and presents a primary guideline of choosing the
optimal element size for different type of analysis. (2) Instead of only focusing on impact analysis, this
research completely studies the effects of mesh density through static, impact, and modal analysis.

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


8. Conclusion
In this study, the effects of element size on accuracy of finite element models and simulation results were
thoroughly investigated through static analysis, impact analysis, and modal analysis. It was found that for
static analysis that assumes steady loading and response conditions, each side of a plate model should be
discretized into 10 divisions in order to obtain satisfied results (with approximation error < 1% in our
example) consuming less computer resources and computing time. For impact analysis which involves fast,
transient loading and considers dynamic structural response, a thin-walled beam model has to be meshed
into 80 divisions along its axis in order to simulate its crash response efficiently and correctly (with
approximation error < 10%). For modal analysis, the selection of the best element size is not very urgent
because (1) the FEA models with less number of elements can correctly predict the natural frequencies (with
approximation error < 1%) and (2) even the FEA models with finest mesh cost no more than 20 seconds to
complete a modal analysis. Nevertheless, it is suggested to apply 10 divisions on the box beam models to
correctly simulate their mode shapes in lowest natural frequencies.
The presented results reveal that different types of structural analysis require appropriate mesh generation
schemes. The optimal mesh density for static, impact and modal analysis can be used as guidelines in
creating other finite element models for structural analysis, which will lead to accurate and efficient
computer simulations.
One deficiency of this study is that the authors did not develop mathematical models to visually show the
relationships among element size, accuracy of results, and computing time for different analyses. This is
because that in any type of analysis, more than one type of result will be extracted and studied (this paper
only lists a few type of results as an example). As demonstrated in the paper, the element size has different
influence on different types of results. Therefore, it is neither possible nor necessary to derive mathematical
models for each type of result to show the influences of the element size. Another shortcoming of the FEA
models presented in this paper is that those models use automatic mesh only. Advanced mesh techniques
such as adaptive mesh are not considered. Also, this paper only discusses the structures with regular shape.
The mesh strategy recommended here can be applied to model more complicated structures with irregular
shapes and even engineering assembly for further validation. In the future, effects of adaptive mesh need to
be considered in studying the influences of element size on the accuracy of FEA results. The present study
can be applied to nonstructural problems such as heat transfer problem and fluid flow problem.

British Journal of Engineering and Technology


ISSN: 2326 425X

URL: http://www.bjet.baar.org.uk

Vol. 1, No. 7, pp 13-28, MAY 2013


Reference
1. LS-DYNA Theoretical Manual, Livermore Software Technology Corporation, May 2007.
2. M. Brocca and Z.P. Bazant, Size effect in concrete columns: finite-element analysis with microplane
model, Journal of Structural Engineering, 127(12), 2001, 1382-1390.
3. Ashford and Sitar, Effect of element size on the static finite element analysis of steep slopes,
International Journal for Numerical Analytical Methods in Geomechanics, 25(14), 2001, 1361-1376.
4. V.E. Saouma, D. Natekar and O. Sbaizero, Nonlinear finite element analysis and size effect study in a
metal-reinforced ceramics-composite, Materials Science and Engineering A, 323(1-2), 2002, 129-137.
5. T. Masakazu, The finite element analysis about size effect of concrete members, Memoirs of the
Faculty of Engineering, Fukuyama University, 25, 2001, 49-54.
6. A. Perillo-Marcone, A. Alonso-Vazquez and M. Taylor, Assessment of the effect of mesh density on
the material property discretisation within QCT based FE models: a practical example using the
implanted proximal tibia, Computer Methods in Biomechanics and Biomedical Engineering, 6(1), 2003,
17-26.
7. J. Zmudzki, W. Walke and W. Chladek, Influence of model discretization density in FEM numerical
analysis on the determined stress level in bone surrounding dental implants, Information Technologies
in Biomedicine, 47, 2008, 559-567.
8. S. Roth and J. Oudry, Influence of mesh density on a finite element model under dynamic loading,
Proceedings of 3 rd European Hyperworks Technology Conference, November 2nd 4th, 2009,
Ludwigsburg, Germany.
9. Z.-P. Li, M.W. Kindig, D. Subit and R.W. Kent, Influence of mesh density, cortical thickness and
material properties on human rib fracture prediction, Medical Engineering & Physics, 32(9), 2010, 9981008.
10. Y.-C. Liu and M.L. Day, Simplified modeling of thin-walled box section beam, International Journal
of Crashworthiness, 11(3), 2006, 263-272.
11. G. DAmours, A. Rahem, R. Mayer, B. Williams and M. Worswick, Effects of mesh size and
remapping on the predicted crush response of hydroformed tubes, Proceedings of 6th European LSDYNA Users Conference, May 29 30, 2007, Gothenburg, Sweden.
12. G.A. Aramayo and M.H. Koebbe, Parametric finite element model of a sports utility vehicle
development and validation, Proceedings of 7th International LS-DYNA Users Conference, May 19
21, 2002, Dearborn, Michigan.
13. M.V. Donado and L. Iannucci, An objective algorithm for strain softwening material models,
Proceedings of 9th International LS-DYNA Users Conference, June 2005, Dearborn, Michigan.
14. M. Makino, The performance of 10-million element car model by MPP version of LS-DYNA on
Fujitsu PRIMEPOWER, Proceedings of 10th International LS-DYNA Users Conference, June 2008,
Dearborn, Michigan.

Você também pode gostar