Está en la página 1de 17

FINITE ELEMENT MODELING OF A BAR

WITH FILLETED SHOULDERS

Caleb Kreeger
City College of New York
ME 371 FEM 1 Project
Overview:

This is a project in which I will report my first experience using the Finite Element Method (FEM) with
SolidWorks Simulation. FEM is used to calculate the displacement, strain and stress of a Solid Model under a
load or thermal stress by using approximations based on Finite Elements. The object I will use for this study is
a 1020 steel bar with two filleted shoulders. (Fig 1) The plate will be subjected to a tensile force of 100 KN
applied to the shorter 100 mm flat side.

Fig 1: Bar with filleted shoulders.

To advance with the FEM study, I created a mesh. This mesh divided the object in tetrahedrons
(finite elements) to create an approximation of the object’s displacement, strain and stress as seen in Figure
2.
The precision of my results was dependent on two factors:

(1) The size and number of tetrahedrons (finite elements), a decrement of size and increment of
total tetrahedrons improved accuracy
(2) The number of nodal points on each tetrahedron. Nodal points are points which trace the
position of the tetrahedrons after the load has been applied and the tetrahedrons are deformed.
The more nodal points, the more accurate results I will get.

For this study I will use 1st order elements (Fig 3) and 2nd order elements (Fig 4).

Fig 3: 4-node tetrahedron element- 1st Order Solid Element- Interpolation among two nodes on each
side (linear).

Fig 4: 10-node tetrahedron element (4 vertices and 6 midside) – 2nd Order Solid Element – Interpolation
among three nodes on each side (quadratic).

Meshes are divided into two types, local mesh and global meshes. Local meshes are used for mesh
refinement meaning they are aimed to increase the accuracy of a certain region that is undergoing high stress or
strain. Global meshes focus on the total object to increase the overall accuracy of the calculations. I also
experimented with curvature based meshes and automatic transition to minimize run time and increase accuracy.
Procedure:

Predictions:

I will predict the stress distribution by using a water flow analogy. Flow analogy will help me
understand how the notches of the plate will affect the stress distribution. If the steel plate were hollow,
and I were pumping water from one side the other, the velocity would increase and be maximum at an
area of high stress concentration when an abrupt change occurs in the circled region below. (Fig4)

Fig 5: Stress Distribution- Flow analogy, showing where the velocity is maximum and minimum.

In this specific case, velocity serves as an analogy for stress. My prediction for where stress is
maximum will be in the cross section where velocity is maximum. My prediction for where stress is
minimum will be in the cross section where velocity is minimum.

Displacement will be very simple to predict, since the left side is fixed it will be the place with
minimum displacement, and the right end of the plate will have maximum displacement from its original
place because every part of the plate is elongating towards that direction.

Creating the Solid Model & Applying Boundary Conditions:

I began by creating the solid model of the plate following the dimension in Figure 1, then selected the
material of the object as 1020 steel. I applied a restrain by fixing the left side of the plate, then applied a force
of 100 KN directed towards the right side on the right side of the bar, as seen in Figure 6.
Fig 6: My filleted bar after applying 100 kN load and applying fixed restraint
Creating a Mesh using 1st Order Elements:

For this step, I created a mesh using 1st order elements, in SolidWorks. This is called draft quality
mesh. For mesh parameters, I selected Standard Mesh and for mesh density, I selected Coarse. This is the
largest element size recommended by SolidWorks. Creating a mesh with these characteristics gave me a
rough approximation since 1st order elements aren’t reliable, and the element size is relatively large.

Creating a Mesh using 2nd Order Elements:

My next step was to change the mesh recently created from 1st order elements to 2nd order elements

by turning off draft quality mesh. Second order elements use more nodes than 1st order elements as seen in
Figure 3 & Figure 4. Second order elements allowed for quadratic interpolation instead of linear interpolation.
This change of the element’s order increased the accuracy of the study.

Decreasing the Global Element Size:

For this step, I gradually decreased the element size from coarse to fine and created a mesh for
each of them as seen in Figure 7. From left to right, as seen in Figure 7, the element size measured from
one edge of the tetrahedrons to another of each mesh was: 14.58 mm, 11.30 mm, 8.20 mm, 5.56 mm,
3.64 mm, and 1.8 mm. This change in element size is called global mesh refinement and was done to
increase the accuracy of the simulation.
Applying Local Mesh Refinement:

For this step, I will apply local mesh refinement for the areas of my interest, which are the ones
undergoing a high stress. After running a stress study on last step’s fine mesh (figure 8), I concluded I
needed to do a local refinement around the filleted shoulders.

Fig 6: max stress occurred in red where I then applied local mesh controls to increase accuracy

I approached the local mesh refinement in two different ways:

1) Applying the local mesh/mesh control around the 4 edges (2 top and 2 bottom) of the
filleted shoulders as seen in figure 9 (in blue):
Figure 9: Mesh control on the edges of the fillets (only one shown)

2) Applying the local mesh/mesh control on the two surfaces (1 top and 1 bottom) of the
notches create on the plate as seen in figure 10:

Figure 10: Mesh control on the surface of the fillets (only one shown)

Then, I created 10 local meshes (2 for each case) with element sizes of: 7.29 mm, 5.74 mm,
3.92 mm, 2.87 mm and 1.82 mm using a global size element of 14.58 mm.
After using local mesh refinement I experimented with automatic transition. Shown in Figure 11 is a standard mesh
with a global element size of 8.20. Below that in Figure 12 is the same standard mesh with same size, but with
automatic transition checked. Automatic transition automatically applies mesh controls to small features, and is
useful with meshing large models with small features and helps avoid using an unnecessary number of elements.
Results:

Table 1: P-refinement (from 1st order to 2nd order)


Global
Number of Number of Number Run Maximum Displacement
Order Size
Nodes Elements of DOF Time (s) Stress (MPa) (mm)
(mm)
1st order 14.58 388 1074 1074 2 126.46 .0813
nd
2 order 14.58 2207 1074 6360 2 179.12 .0826

Table 2: H-refinement (element size) with 2nd order elements


Global
Number of Number of Number Run Maximum Displacement
Size
Nodes Elements of DOF Time (s) Stress (MPa) (μm) M
(mm)

14.58 2737 1300 8076 2 179.12 .0826


11.30 4400 2401 12687 3 182.38 .0826
8.20 9007 5192 26544 4 185.68 .0826
5.56 21413 12586 63144 7 191.47 .0826
3.64 65365 41218 193788 22 192.65 .0826
24
1.8 65365 41218 193788 192.95 .0826
h-refinement- nodes vs. stress (Table 2)
195

Max Von Mises Stress (MPa)


190

185

180

175

170
2207 4400 9007 21413 65365 453601
# of Nodes

Figure 11: convergence graph for h- refinement. (Table-2)

Table 3: Local refinement along the edges of the notch with global element size of 14.58 mm

Local Number Number Run Maximum


Number Displacement
Size of of Time Stress
of DOF (μm)
(mm) Nodes Elements (s) (MPa)

7.29 2368 1163 6843 2 192.22 219


5.74 2608 1279 7563 2 192.59 219
3.92 3353 1696 9798 2 191.21 219
3
2.87 4248 2237 12483 193.2 219
1.82 8539 4672 25482 3 194.11 219
.8 193.39
.4 4248 2237 12483 3 193.13 219
Local ref edges- nodes vs. stress (Table 3)
194

Max Von Mises Stress (MPa)


193.5

193

192.5

192

191.5

191
0 10000 20000 30000 40000 50000
# of nodes

Figure 11: convergence graph for local refinement along the edges
Table 4: Local refinement along the surfaces of the fillet with global element size of 14.58 mm

Local
Number of Number of Number Run Maximum Displacement
Size
Nodes Elements of DOF Time (s) Stress (MPa) (μm)
(mm)
7.29 2368 1144 6744 2 193.15 0.000219
5.74 2668 1280 7506 2 184.22 0.000219
3.92 3353 1604 9222 2 195.66 0.000219
2.87 4248 2163 12015 4 192.71 0.000219
1.82 6390 3033 16617 3 194.02 0.000219
.8 22037 193.36
.4 76305 3033 16617 3 193.44 0.000219

Local ref surface nodes vs. stress (Table 4)


198
Max Von Mises Stress (MPa)

196
194
192
190
188
186
184
182
0 10000 20000 30000 40000 50000 60000 70000 80000 90000
# of nodes

Figure 11: convergence graph for local refinement along the surface of fillet
Table 5: Curvature based mesh

Using the curvature based mesh allowed for me to specify a minimum and maximum element size while at the same time determining the
element size as the minimum number of elements in a hypothetical circle. Curvature based meshes take greater advantage of multi core
processors which often makes a quicker run time than most standard meshes. This is due to its ability to adapt to the geometry and use
smaller elements where the curvature is the highest. I ran a curvature based mesh with a maximum element size of 7.29 mm and a
minimum element size of 2.0 mm. I then compared the run time to a simulation done earlier with a similar amount of elements. Shown in
Figure 12 are the results, showing a run time of 4 seconds with 45,396 nodes. I compared this to

Analytical Solution:
𝐹 = 100𝑘𝑁
Radius of Notch,𝑅 = .002 𝑚
t=thickness=.01 m
d=.1 m
D=.2 m
𝐷
=2
𝑑
𝑟 𝑎
𝑘𝑇 = 𝐵 ( )
𝑑

a=-.321
B=1.1
. 002 −.321
𝑘 𝑇 = 1.1 ( ) = 184.40 𝑀𝑃𝑎
.1

From this experiment, I learned how to use Finite Element Modeling with
SolidWorks. This will be of great advantage to me whenever I will analyze a certain
object and attempt to reduce its stress concentration. From the data acquired, I learned
the displacement doesn’t noticeably change with H – refinement (element size). I
learned that P-refinement from 1st order to 2nd order elements improves enormously the
accuracy of the study. I learned local and global H-refinement also improves the accuracy
of the study. I also learned local refinement is important to apply on areas with high stress
concentration in order to get more accurate results with FEM. I saw as well how the
number of nodes and degrees of freedom are related linearly.
From the data collected, I observed how the three graphs were converging towards
a value close to the analytical solution. The highest % difference between the convergence
values of the 3 graphs and the analytical solution was only 3-4 %. This seems a bit high,
but results with FEA are not always assumed to be exact or correct. In addition, Ican
recognize it is better to use the local refinement on surfaces instead of edges since it
gives a slightly smoother curve on the graph and has a smoother convergence. I also
separately ran a single study with a very small element size of 0.5 mm, which took 15
minutes to process and got a maximum stress of 193.07 MPa, which is in between all the
three convergence values for the graphs. The analytical solution for maximum
deflection was lower than any deflection from the studies I ran, the reason for this after
looking at thki mo any other place, and it is increasing the total displacement.
My prediction for maximum stress was right as I concluded it would be in the
filleted shoulders. The prediction for minimum stress was somewhat accurate but didn’t
take place in the exact part I had originally predicted. I had predicted the minimum tress
would be in the middle of the top surface of the bar, but instead it took place at the edges
of the bar, in between the fixed end and the filleted shoulder. My prediction for minimum
and maximum displacements were right though since I just had to use the boundary
conditions to predict which part of the bar deforms the most.

With a curvature-based mesh I attempted to reduce the run time of my


simulations and increase the efficiency of my studies. I failed to increase the
efficiency to a drastic extent and take advantage of the curvature-based mesh’s
ability to mesh complex curvature geometries. This was simply due to the lack of
curvature in my model. With models that have a more complex curvature I assume
I would see a greater decrease in run time and increase in efficiency when using a
curvature-based mesh instead of a standard mesh.

I then experimented with automatic transition meshes and looked at the


difference in mesh complexity in areas of my model where automatic mesh control
was applied. For the automatic transition mesh, an increase in the complexity of the
mesh in my filleted shoulder proved to me the capability of an automatic transition
mesh to automatically apply mesh control to the filleted features of my model. In
my comparison to a standard mesh, I concluded that there was an increase in mesh
complexity for the automatic transition mesh, affirming my prediction that
automatic mesh control was applied.

One of my suggestion to improve this assignment, would be to additionally analyze


strain. Before doing the tutorials recommended by Prof. Benenson, I thought that the
place undergoing the highest stress, was also undergoing the highest strain. But this isn’t
true, I think it is a very important characteristic of the FEM study we are missing.
As a last note, I wanted to point out that most of us aren’t familiar to how much
100 KN of force really is, so after doing a short research, I calculated it would be similar to
hanging two African Elephants from this steel plate. Which is hard to believe, since I would
have thought this 1020 steel plate would have broken while doing the FEM.

También podría gustarte