Professional Documents
Culture Documents
98 Heckman
98 Heckman
ABSTRACT
Pressure vessels are a commonly used device in marine engineering. Until recently the primary analysis method had been hand
calculations and empirical curves. New computer advances have made finite element analysis (FEA) a practical tool in the study
of pressure vessels, especially in determining stresses in local areas such as penetrations, O-ring grooves and other areas difficult
to analyze by hand. This project set out to explore applicable methods using finite element analysis in pressure vessel analysis.
Having tested three dimensional, symmetric and axisymmetric models, the preliminary conclusion is that finite element analysis is
an extremely powerful tool when employed correctly. Depending on the desired solutions, there are different methods that offer
faster run times and less error. The two recommended methods included symmetric models using shell elements and axisymmetric
models using solid elements.
Contact elements were tested to determine their usefulness in modeling the interaction between pressure vessel cylinder walls
and end caps. When modeled correctly, contact elements proved to be useful, but the operator also needs to be able to interpret
the results properly. Problems such as local stress risers, unrealistic displacements and understanding how to use such data
become extremely important in this kind of analysis. This highlights the key to proper use of finite element analysis. The
analyst should be able to approximate the solution using classical methodology (hand calculations) in order to verify the
solution.
INTRODUCTION
This project set out to verify finite element analysis, or FEA, when applied to pressure vessel design. While finite element analysis
offers another way to analyze structures, it requires an understanding of the program and subject being modeled. If the operator
does not use the correct model, time is wasted and more importantly the data is useless.
Finite element analysis is a powerful tool in the field of engineering. Initially, finite element analysis was used in aerospace
structural engineering. The technique has since been applied to nearly every engineering discipline from fluid dynamics to
electromagnetics.
The difficulty in analysis of stress and strain in structural engineering depends on the structure involved. As the structure grows in
complexity, so does the analysis. Many of the more commonly used structures in engineering have simplified calculations to
approximate stress and strain. However, these calculations often provide solutions only for the maximum stress and strain at
certain points in the structure. Furthermore, these calculations are usually only
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (1 of 7)
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (2 of 7)
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (3 of 7)
Run Time
(s)
s1 Error
(%)
107
Output
File Size
(MB)
4.52
6.5
557
5.67
4.8
6.2
747
30
16.66
2.47
3.3
3.9
6.3
122
2.97
4.8
8.21
3.3
Element Type
180
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (4 of 7)
s2 Error
(%)
DZ Error
(%)
DR Error
(%)
7.0
6.7
0.5
6.2
3.7
0.1
6.9
200
2
5.73
0.48
40.9
3.2
0.85
6.4
14
2.24
3.2
6.3
13.3
0.7
3.1
4.5
4.7
5.2
1.4
1.1
3.0
1.9
6.2
DISCUSSION
The axisymmetric model had by far the shortest run time with comparatively small computational error. Even the highly refined
axisymmetric model ran in half the time of the fastest shell model. The quadratic shell had the longest of the initial mesh size times.
However, when the number of linear shell elements was doubled, the run time increased above that of the quadratic models. The
solid model took roughly twice as long as the shell model to run.
Not surprisingly, run time is proportional to the number of elements for a given element type. In axisymmetric models, when the
mesh doubled, run time doubled and when mesh size quadrupled, the run time nearly quadrupled.
Symmetric models offered an impressive improvement in run time. For the shell elements run time was reduced by roughly
four times in each case.
File space was dependent on model size. The axisymmetric models were the smallest. The standard mesh axisymmetric model was
five times smaller than the next closest model. Doubling the mesh size resulted in slightly less than doubling the file size. A
quadratic model was larger than a linear model of the same mesh size, but significantly smaller than a twice as refined linear
model. The solid model was the largest of the initial mesh models.
The calculated error for all of the models was roughly the same. Only in one case did the error exceed seven percent. Even with the
refined meshes, error did not necessarily improve. Comparing the axisymmetric models showed that the error fluctuated around
three to six percent even with the higher refinement. This error is most likely the result of constraints applied to the model.
Constraints do not completely simulate reality and slight errors are a result.
It is also worth comparing the stress plots for each model. The shell (Fig 5) plots look different from the solid model (Fig 6)
and axisymmetric (Fig 7) plots due to the fact that the solid model shows the bending from the thick endcaps,
while the shell elements do not give the appearance of as much bending.
Since error between each element is fairly similar the choice of which to use will be dependent on the analysis being run. In cases
where the required wall thickness is unknown, shell elements are an excellent choice since the wall thickness can be iterated by
simply changing a constant without having to change the model geometry.
When the wall thickness is known solid and axisymmetric models can provide greater data resolution. Shells only provide data at
the wall edges. Solids allow for meshing inside the wall so the stress distribution can be studied through the wall thickness as well as
on the outer surface. However, solid models have extremely high run times and consume large amounts of disk space.
Axisymmetric models run considerably faster with no greater error. In fact, an axisymmetric model can be used with much greater
resolution than a full solid model, yet will still run in less time and use less disk space.
Problems Noted
Shell elements were found to have some unique problems that should be noted. When modeled using symmetric boundary
conditions, a small number of elements can result in incorrect stresses along the boundary. In a low element number model (Fig 8)
stresses along the boundary was slightly higher than for the rest of the cylinder. Another model
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (5 of 7)
were tested with a slightly finer mesh (Fig 9). The stress along the boundary decreased in the center of the cylinder, but was
still apparent near the corners.
When interacting with solids, shell element thickness does not completely apply. A model was tested where shells were used for
the cylinder wall. A cap was placed on the end of the cylinder for the end cap. When the cylinder walls were composed of solid
elements the cap showed a fairly even stress distribution (Fig 10). When a pressure was
applied the solid and shell interaction was not that of a flat surface against a flat surface. The shell acts as a point load on the solid
(Fig 11). In such models solids should be used instead of shell elements.
Contact Elements
For contact between the cylinder and end cap three-dimensional point to point (CONTAC52) elements were used. The CONTAC52
works by preventing the two nodes from passing through each other. An element is generated between both nodes and given a
stiffness. If the stiffness is too low the outer node will pass through the inner node. However, if the stiffness is set too high the
problem will not converge and no solution will be found. Thus, some experience in using these elements is useful
The first test model succeeded in preventing the outer nodes from penetrating the inner nodes. The model consisted of two cylinders
composed of linear shell elements. The outer cylinder was longer than the inner cylinder to illustrate the effects of the inner and outer
shell interactions. When the outer shell compresses and impacts the inner shell some of the load is transferred to the inner shell so the
displacement of both shells will be lower than the end of the outer cylinder that is unsupported.
For the initial tests all inner and outer nodes lined up radially so that as the cylinders compressed the nodes would directly impact
one another (Fig 12). Under these conditions the outer cylinder stopped as soon as it impacted the inner cylinder (Fig 13). The
inner cylinder showed the greatest stress at the end where the outer cylinder started bending as it
was unsupported (Fig 14).
Axisymmetric models were tested using CONTAC52 elements. Initial tests were successful in using the contact elements to
prevent penetration between the endcap and cylinder (Fig 15). Stress risers are located at the point where they impacted (Fig
16).
A larger model was next run using contact elements. This pressure vessel was far more complex, including a center hole and Oring grooves (Fig 17). The solution was within five percent of hand calculations for the cylinder wall (Fig 18) and the center of
the end cap (Fig 19). However, extremely high stresses were found at the contact point, roughly
945 MPa (137 ksi) (Fig 20). The only other area in the pressure vessel that came even close to this stress was at the other end
where stress reached 303 MPa (44 ksi).
The extremely high stress was primarily the result of the sharp corner of the cylinder model. What actually would happen in a
ductile material is the edge of the cylinder would deform to reduce the stress. This deformation would increase the contact area
over which the force between the cylinder and end cap acts, which in turn reduces stress.
CONCLUSIONS
Finite element analysis is an extremely powerful tool for pressure vessel analysis when used correctly. Tested models were run with
errors ranging from seven to nearly zero percent error and could be run in a relatively short time. However, even with such results
the operator still is required to be knowledgeable of not just how to run the finite element analysis, but also how to read the
results. Data must be verified with hand calculations to confirm that solutions are relatively accurate. Where results are
questionable, such as in the final contact element model, one must understand just what the finite element model is modeling and
how well this approximates the actual subject. For this pressure vessel, the model had a sharp corner, where in the actual pressure
vessel there is a small radius which reduces the stress.
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (6 of 7)
For pressure vessels finite element analysis provides an additional tool for use in analysis. However, it must be compared to other
available data, not taken as being correct just because it looks right. Used with this understanding, finite element analysis offers
great insight into the complex interactions found in pressure vessel design.
ACKNOWLEDGEMENTS
Thanks to David Beals for help in the editing of this paper and assistance in modeling.
REFERENCES
ANSYS Basic Analysis Procedure Guide, Release 5.3. (1996). Ansys, Houston, Pennsylvania
ANSYS Expanded Workbook, Release 5.4. (1997). Ansys, Houston, Pennsylvania ANSYS
Modeling and Meshing Guide, Release 5.3. (1996). Ansys, Houston, Pennsylvania ANSYS
Structural Analysis, Release 5.3. (1996). Ansys, Houston, Pennsylvania
Craig, R. (1996). Mechanics of Materials. John Wiley and Sons, New York, New York. 639 Pages
Young, W. (1989). Roarks Formulas for Stress & Strain. McGraw-Hill, New York, New York. 763 Pages
http://www.mbari.org/education/internship/98interns/98internpapers/98heckman.html (7 of 7)