SimSolid “Real World” Validation Manual For

Stress Accuracy
Richard King 04/09/19

Abstract
This manual contains a series of SolidWorks models with static simulation studies that were run in Altair SimSolid™.
Some of them are complex, but an accurate solution was obtained in SolidWorks by the conventional procedure of
placing mesh controls in the regions of highest stress gradient and successively refining until the solution converged.
This provides a suite of accurate stress results to compare SimSolid’s results against. In all cases SimSolid’s stress results
compared to the SolidWorks comparison results within 10 percent.

Introduction
SimSolid is a very easy to use and fast product for structural analysis using novel alternative technology- a meshless
approach that based on similar theory to FEA. This approach can be shown theoretically to converge to correct
solutions, as discussed in the whitepaper on the technology [1], where it is shown that the mathematical foundation for
the method used in SimSolid is on as solid of a foundation as FEA. But in stress analysis the quantity being calculated is
the derivative of the directly calculated variables (displacements) so is more difficult to achieve accurate results. This is
true of FEA codes as well as SimSolid. In FEA codes, the solution is to have a fine enough mesh in regions of high stress
gradients. This can sometimes be done adaptively using automatic mesh refinement, but often requires one or more
user interventions, for example to create mesh controls, as described in appendix 1. On complex models this can be
tedious and time consuming.

SimSolid has an automated approach to address this issue. In a first step, the model is analyzed with automatically
chosen displacement functions to accurately calculate displacements, assure continuity at all interfaces and satisfaction
of boundary conditions. This captures overall load paths and reaction forces accurately [2]. In the second step, it adapts
to local features where there are high stress gradients by introducing additional functions it those areas. From a user’s
standpoint, this “two stage workflow”, as it is called in the training material, is very easy:

1. Assign materials, loads, and constraints directly to the CAD model. Hit the “Solve” button to run the model with
defaults settings. This run is typically very fast. When complete, examine the results to make sure the
deformation looks reasonable.
2. Turn on solution settings to adapt to features by clicking a single checkbox shown in the figure. If the model
contains thin shell or sheetmetal type parts, an additional button, “adapt to thin solids” can be checked. This is
all done for the entire model for a single part model. For an assembly, the user picks specific parts in which high
stress has been observed to use for adapting to features or thin solids.

I have used various finite element codes with adaptivity, including SolidWorks Simulate, Autodesk Fusion 360, and
Mechanica (of which I was co-inventor and one of the authors). In my opinion SolidWorks Simulate is a good quality
commercial code, with a robust automesher, so it was a good choice for comparing the accuracy of SimSolid. But all FEA
codes have the drawback of difficulties associated with meshing, so none of these is as easy to use in practice as the two
stage workflow for SimSolid described in the Altair SimSolid Fast Start Training Guide [2].

Solutions for Validation

Most FEA codes have validation manuals, and SimSolid does as well, but they typically contain test cases that tend to be
simple, so that theoretical solutions are available. I consider them “necessary but not sufficient”. To further investigate
accuracy, this manual uses a variety of more complicated models for which in most cases there is not theoretical
solution for comparison. Instead, an accurate solution for all these models was obtained in SolidWorks, by using
automatic mesh refinement and by the conventional procedure of placing mesh controls in the regions of highest stress
gradient and successively refining until the solution converged. This provides a suite of accurate stress results to
compare SimSolid’s results against. The procedure for obtaining accurate FEA results in SolidWorks simulation is
described in appendix 1. The specific procedure used to obtain accurate stresses in SimSolid is described in appendix 2.

Note on Singular Stresses

Many situation such as reentrant corners lead to stresses that are singular (theoretically infinite). Some examples are
shown in appendix 3. If the maximum stress in your model occurs at one of these it is best to fix the situation, for
example by creating a fillet. Alternatively, if you are interested in the stress elsewhere in the model, it can be examined
using a probe. This was necessary, for example, for the hanger-beam tutorial.

Model Location
All the models described in this document are in the folder “valmanModels”. There is a SimSolid workspace (.ssp) file for
each, as well as SolidWorks part and/or assembly files.

Summary

Model Description/name error in maximum stress

Single Part models
1. Plate with hole (pwh) 1.2%
2. Bar with fillet in bending (petebendfillet) 0.8%
3. plate with narrow elliptical hole (Ingliss) 1.5%
4. model with thin slot (model with thin slot) 1%
5. bar with edge slits (edgeslit) 1.6%
6. stepped model with fillet and slot (stepfillet) 3.2%
7. Crank (crank) 5.3%
8. Tube joint with thin solids (newtj) 2.0%
9. Spherical pressure vessel with hatch (spherehatch) 4.4%
10. Canoe paddle (canoepaddle) 0.6%
11. rectangular tank (recttank) 5.9%
Assembly Models
12. multiple cylindrical columns with fillets (railcylassy) 0.7%
13. multiple bars with edge slits (edgeslitassy) 2.5%
14. testing machine and specimen (clevisassy) 4.1%
15. multiple plates with steps and fillets (filletassy) 0.2%
16. structure with thin plates and girders (floorsystem) 8.6%
17. pullup bar (SimSolid tutorial) 3.5%
18. hanger-beam (SimSolid tutorial) 2.6%

Solution Timing: Solution time was similar for all the smaller models. The last 3 models are relatively large and the timing
for SolidWorks vs SimSolid is described below. For these models the total solution time (including meshing time in
SolidWorks) is significantly less in SimSolid. For all models, considerably less user intervention was require in SimSolid to
achieve an accurate result.
Single Part Models

Model: pwh
Description: Plate with hole

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
7.5 Mpa 7.6 Mpa 1.2%
(3 solution passes)
Notes:
• Theoretical Solution for this model is 7.6 Mpa [3]
• The SimSolid results did not change significantly when “adapt to features” was turned on with 3 solution
passes, so that was taken as the converged result.
• SolidWorks adaptive mesh refinement result was accurate for this model
Model: petebendfillet
Description: Bar with fillet in bending

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
826 Mpa 819 Mpa 0.8%
Notes:
• The handbook solution for this model is 823 Mpa [4]
• SimSolid Stage II Solution did not change appreciable between runs with 4 and 5 solution passes so the result
from 4 passes was taken as the converged result
• SolidWorks adaptive mesh refinement result for this model was 771 Mpa which is 6.6% low. The SolidWorks
Refined mesh result was obtained by applying a mesh control at the fillet and successively refining it
manually.
Model: Ingliss
Description: plate with narrow elliptical hole

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(6 solution passes)
6.21E4 Pa 6.12E4 Pa 1.5%
Note: SolidWorks adaptive mesh refinement result for this model was 5.66E4 Pa which is 8.9% low. The SolidWorks
Refined mesh result was obtained by applying a mesh control at the hole and successively refining it manually
Model: thinslot
Description: model with thin slot

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
4367 Pa 4412 Pa 1%
Note: SolidWorks adaptive mesh refinement was accurate for this model and agreed with the solution obtained by
successive manual mesh refinement
Model: edgeslit
Description: bar with edge slit

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(5 solution passes)
298 ksi 293 ksi 1.6%
Model: Crank
Description: Crank

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
12.21 Mpa 12.86 Mpa 5.3%
Note: SolidWorks adaptive refinement gets an accurate result for this model.
Model: newtj
Description: Tube joint with thin solids

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(6 solution passes)
167 Mpa 170 Mpa 2%
Notes:
• SolidWorks adaptive refinement gets an accurate result for this model.
• In SimSolid I used “adapt to features” and “adapt to thin solids”
Model: spherehatch
Description: Spherical pressure vessel with hatch (spherehatch)

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(5 Solution passes)
1.31 Mpa 1.24 Mpa 5.5%
Note: SolidWorks adaptive refinement gets an accurate result for this model.
Model: canoepaddle
Description: Canoe paddle

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
66.4 Mpa 66.8 Mpa 0.6%
(4 solution passes)
Note: SolidWorks adaptive refinement gets an accurate result for this model.
Model: recttank
Description: rectangular tank

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
3.68 Mpa 3.90 Mpa 5.9%
(4 Solution passes)
Note: SolidWorks adaptive refinement is off by 12% on this model
Model: railcylassy
Description: multiple cylindrical columns with fillets

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
1.057E4 Pa 1.05E4 Pa 0.7%
Notes:
• SolidWorks adaptive refinement gets an accurate result for this model.
• For this Model the Max stress occurs at a singularity where the column has a reentrant corner with the base.
Solidworks did not detect this singularity because the mesh was not refined there. The maximum meaningful
stress is in a fillet on the column, and was determined in SimSolid using a probe.
Model: edgeslitassy
Description: multiple bars with edge slits

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
0.128 Mpa 0.123 Mpa 3.9%
Note: SolidWorks adaptive refinement is off by 13% on this model
Model: clevisassy
Description: testing machine and specimen; tensile load 100000 N

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
165.7 Mpa 172.5 Mpa 4.1%
(4 Solution passes)
Note: SolidWorks adaptive refinement is off by 22% on this model
Model: filletassy
Description: multiple plates with steps and fillets

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
55.4 Mpa 55.3 Mpa 0.2%
(4 Solution passes)
Note: SolidWorks adaptive refinement gets an accurate result for this model.
Model: floorsystem
Description: structure with thin plates and girders

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
2.56 Pa 2.75 Pa 7.4%
Notes:
• This model has shell elements so adaptive refinement is not available in SolidWorks
 • This model took 54 seconds to solve with 4 solution passes in SimSolid. The SolidWorks solution time for 3
mesh refinement runs was 290 secs total. There was also considerable user intervention required.

Model: hanger-beam
Description: hanger-beam model from SimSolid Tutorial.

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Solution Error
(4 solution passes)
41.9 Mpa 40.6 Mpa 2.6%
Notes:
• This is design study “baseline” from the tutorial
• SolidWorks picks up a singular stress result for the maximum, elsewhere in the model. For the comparison I
probed the SolidWorks in the vicinity of the fillet where SimSolid detected the maximum
• SolidWorks adaptive mesh refinement gets a value of 30.8 for the maximum near the fillet which is 26% low.
• Timing: SimSolid took 14 secs to solve this model with 4 solution passes. The SolidWorks solution time for 3
mesh refinement runs was 140 secs total. There was also considerable user intervention required.
Model: pullup bar
Description: pullup bar model from SimSolid Tutorial.

SolidWorks Refined mesh: SimSolid stage II solution:

Maximum Stress Result

SolidWorks Refined mesh: SimSolid Stage II Error
Solution (4
solution passes)
143.4 Mpa 138 Mpa 3.5%
Notes:
• This is design study baseline, analysis “structural 1” from the tutorial. I had trouble getting SolidWorks to
process to bolts in the same manner as SimSolid, which affected the results. To work around this I placed an
additional immovable support on the face highlighted in yellow above
• The maximum stress in the model occurs at a reentrant corner between the pullup bar and a vertical face
welded to it. The stress is singular there and theoretically infinite, so cannot be used for accuracy
comparisons. I modified the model to fix this by placing a fillet.
• SolidWorks automatic mesh refinement failed for this model, with the message “Mesh Adaptation Failed”.
The comparison results were obtained by placing a mesh control on the fillet and successively refining it until
the stress converged.
• Timing: The SimSolid solution, with adapt to features and 4 solution passes, took 18 secs for this model. The
SolidWorks solution time for the 3 mesh refinement runs was 83 secs total. There was also considerable user
intervention required.
Appendix 1-Obtaining Accurate Stresses With FEA
This is a topic to which an entire book can be devoted (in fact I am working on one). I’ll describe here in detail the
procedure I used in all the test cases in this book. If you have a FEA stress result that does not agree with SimSolid, it is
recommended this procedure be followed to make sure the FEA result is accurate.

The default mesh from commercial FEA codes does a pretty good job of following the geometry of the model and will
often produce accurate displacement results. Stress results are more localized and may not be accurate unless manual
or adaptive mesh refinement is done in local regions of high stress gradient.

The following settings for an automesher are crucial as a prerequisite for achieving accurate stresses. They are turned on
by default in Solidworks Simulate, but not always in other codes (for example they are off by default in SolidEdge
Simulate), because they can make the automesher fail. But if they are not on there is little chance of accurate stresses.

• Project midside nodes on surfaces. If this is not on you may be modeling the wrong physical model, such as a
plate with a hexagonal hole instead of a circular hole.
• Curvature-based mesh refinement: more elements should be placed along highly curved edges of the model.

Here is an example of a decent default mesh near a curved cutout (this is actually the 14th validation model, the testing
machine assembly with a notched specimen:

This is a reasonable mesh. The geometry is followed well, there are more elements where the curvature is higher, and
the elements near the feature (the curved end of the notch) are not large compared the same size of the feature. They
are pretty large for a region of high stress gradient, however, which raises a red flag from the standpoint of accurate
stresses: elements used in FEA codes represent the displacements quadratically, so stresses, which are related to the
derivatives of displacement, are approximately linearly. But the stresses near a curved feature will vary sharply,
something like shown in the red line below, with distance from the edge of the notch:

You can see this in the fringes of the FEA result:

The stress goes from high to low in the distance of a single element. Mesh refinement is clearly needed in the vicinity,
either using manual or adaptive mesh refinement.

Here is the mesh obtained using the manual mesh refinement procedure described below:

Now the high stress variation now occurs across several elements.

Here is the adaptive automesh for this same model:

It looks reasonable also. But it turns out this result is only in error by 22%. The default mesh is in error by 41% for this
model.

Suggested Refinement Procedure:

1. Run the model with the default mesh
2. If your FEA code has adaptive mesh refinement, run the model with adaptive refinement. for example, in
SolidWorks, right-click the analysis name, choose properties, then select “h-adaptive” from the properties form:

3. If the adaptive stress results agree well with the default, this is probably an accurate result. Of the 19 models in
this manual, that was true for only one, the plate with a hole. Otherwise further manual refinement is needed.
In the testing machine assembly, for example, the default solution and adaptive solution disagree by 12%, so I’d
recommend checking further with manual refinement.
4. Examine the mesh near the vicinity of maximum stress. Place a mesh control there, for example, the notch
surface:
 Set the mesh density to high or “fine” in the control and remesh. Make sure the new mesh looks reasonable in
that vicinity:

If not, you can manually tighten the mesh density more. Once the mesh looks reasonable, run the model again.
In this case the first mesh from the mesh control still doesn’t look good with fairly large elements still adjacent
to the notch, so I reduce the element size on the control and try again:

This looks better, so I solved with this mesh.

5. You should repeat this process at least once more, a good rule of thumb is to increase the density of your
control by a factor of two (= lower the element size by 50%), and run again. If the results of your last two runs
agree within a few percent the result is probably accurate. Otherwise, you need to repeat the refinement and
re-run step until the result stops changing significantly.
This process can be time consuming for large models but is necessary. For the 19 models run in this study, SolidWorks
adaptive refinement either did not work or got an inaccurate answer in 6, or almost a third of the models. So while it is a
very good first step, I recommend double-checking it with manual refinement.

The 5 step procedure I described was used for all the FEA results in this manual.

Appendix 2- Obtaining Accurate Stresses With SimSolid

There is a good overview of the two stage procedure for obtaining accurate stress results in SimSolid in the Fast Start
Training Guide [2]. With the default settings in SimSolid, the displacements and load paths through the model are
usually accurate for static analyses. However, stress is more difficult to calculate accurately because it often is
concentrated in regions of high gradients. The training guide suggests an excellent two stage approach for handling this:
Analyze the problem with default settings (stage I), then “zoom in” on regions of interest where the might be high
stresses, and use the automatic “adapt to features” setting (stage II). You can also at this point request more adaptivity
passes in the solution. Then the analysis is run again with these modified settings, to obtain accurate stresses. This
approach is much more efficient than simply setting “adapt to features”, and increasing the number of adaptivity passes,
for the entire model. That may work fine for relatively small models but be time consuming for large assemblies. The
two stage approach lets you quickly compare multiple design alternatives for large models with stage I, then verify the
accuracy of stress results in a final stage II run. The user intervention required for stage II is trivial compared to having to
manually do mesh refinement in FEA codes.

Now let’s look at this procedure in detail using the same example used in appendix 1, the testing machine assembly.
After solving the default analysis (“stage I”) we see the max stress in the model is in the notch of the specimen:

Now the procedure is to turn on adapt to features, but only for the part or parts where the stress is highest:
 Note there is a pulldown for refinement level: standard, increased, or fine. The higher this level, the more accurate the
stresses. I set it to fine for all the models in this manual. Since we’re only doing this for a small subset of the model, it
does not add to the solution time too much. The other option on this form is “Max number of adaptive solutions” which
defaults to 3. I recommend leaving it at 3 and re-running:

If this agreed with the default solution within a few percent, we’d be confident in an accurate solution, but it is off
considerably. So now I recommend bumping up the Max number of adaptive solutions to 4 and trying again:
This did not change the result much so we can accept this result. Otherwise it would be necessary to adjust the number
of solutions again and repeat the process.

This procedure was done for all the models in this manual. Out of 19 models, one needed only 3 for the number of
solutions, thirteen needed 4, two needed 5, and two needed 6.

Appendix 3- Avoiding Singularities

Results at singularities are not physically meaningful because the stress is theoretically infinite. They cause mesh
refinement (or p-adaptivity) in FEA codes, or adaptive analysis in SimSolid, to diverge because the stress just keeps
increasing towards infinity.

They can be caused by geometry, such as reentrant corners, or by loads, constraints, and special elements. Here are
some examples:

Some singular stress situations

Singularities caused by loads can be avoided in SimSolid by applying the load to a “spot” or by using remote loads.
Reentrant corners (or similar situations where there is a kink where two surfaces meet) can be resolved with fillets.

If the maximum stress in your model occurs at a singularity it is best to rethink the true physical situation you are trying
to model and resolve it. For example, tube joints do not meet at an infinitely sharp corner, in reality there should be a
fillet there. If you are interested in the stress state elsewhere in your model, you may need to use probes to evaluate the
stress at that other location.
About The Author
Richard B King, Ph.D.

I received my Ph.D. in Applied Mechanics from Stanford University in 1980. My training included theoretical stress
analysis, numerical methods including the boundary element and the finite element methods, and variational methods
such as the Ritz and Galerkin methods (described in the SimSolid Technology Overview [1[, these are the predecessors of
both the finite element method and the external method used in SimSolid). I have decades of experience in stress
analysis.

In the early part of my career I used all of these methods in investigating the mechanical reliability of pipelines and naval
vessels at the National Bureau of Standards, and of disk drives at IBM research. At IBM it became clear that the state of
the art of finite element analysis was found to be clearly inadequate for design purposes, which led to the idea of using
the adaptive p-method of finite element analysis for this purpose. This led to my co-invented Mechanica and cofounding
Rasna Corporation.

Rasna was acquired by PTC in 1995, and I continued to work on enhancing the adaptive analysis capabilities in
Mechanica (later renamed Creo Simulate). This included advanced adaptive nonlinear features including snapthrough,
nonlinear thermal, large strain plasticity, large displacement and frictional contact.

After leaving PTC I did freelance consulting and teaching, and developed stressRefine, an accessory for p-adaptive
solution of existing FEA meshes, which is now available open-source. I also investigated the capabilities of commercial
FEA adaptive mesh refinement solutions include SolidWorks Simulation, Ansys, and Autodesk Fusion 360 Simulation.
More recently I have looked into SimSolid.

References
1. Altair SimSolid Technology Overview, Altair Corporation,
https://altairhyperworks.com/ResourceLibrary.aspx?category=Technical%20Papers&altair_products=SimSolid
2. Altair SimSolid Fast Start Training Guide,
https://altairhyperworks.com/ResourceLibrary.aspx?category=Technical%20Papers&altair_products=SimSolid
3. Timoshenko, S, and Goodier, J, Theory of Elasticity, McGraw Hill, 1970
4. Pilkey, W. D., Pilkey, D. F., & Peterson, R. E. (2008). Petersons stress concentration factors. Hoboken, NJ: John
Wiley.