Designer Oriented

Software -
Is it Accurate? Part 2
Ian Symington, NAFEMS

his series takes a look at the tools currently available for designers aiming to

T develop their designs with an analysis capability. When selecting a particular

software code, there are a range of criteria that need to be considered, such as
cost, usability, speed and accuracy. In this series, we focus on performing an
assessment of the accuracy of different designer oriented codes using eight
benchmarks in the area of linear elastic, small displacement structural mechanics.
Part 1 features SimSolid by Altair and is available to download for free from the
NAFEMS resource centre [1]. Here in part 2, the focus shifts to midas MeshFree [2] a
software code developed by Midas IT.
Before taking a look at the tool, I wanted to recap what I'm looking for in a designer
oriented tool. It is worth stressing that I'm not looking for a replacement for a
general-purpose FEA package. I'm expecting that anyone needing reliable and
robust results will already have access to one, or be able to outsource the analysis to
an organisation with this capability. With the focus being on designer oriented
software, I'm looking for a tool that can help inform design decisions. In particular
I'm looking at tools that are:
1. Easy to use – require little specialist knowledge
2. Quick to run – this means the solution needs to run in the time it takes to grab a
cup of coffee AND there should be no need for hours of preprocessing cleaning
up the geometry so that it can be used in by the tool.
3. The accuracy must be sufficient to guide the direction that a design should take
but doesn't need to be able to hit the level of precision that is achievable using a
general-purpose analysis package.

2
3
 Figure 1: Comparison of discretisation using a traditional FEA approach (Left) and Non-conforming grid approach (Right).

Figure 2: Computational Domain with an illustration of the adaptive integration scheme

Unboxing midas MeshFree analysed. Cells which intersect with the boundary of the
Let's start by looking at how the midas MeshFree part are subject to a proprietary adaptive numerical
marketing team describe their product. integration scheme which optimises the number of
quadrature points for an arbitrarily shaped domain of
interest.
"MeshFree performs finite element analysis on
the original CAD model without need for meshing or
defeaturing. MeshFree frees you from all the
difficulties of today's design and analysis software." The Simulation Process & Solution
Accuracy
midas MeshFree provides solution capabilities for While the underlying numerical method may be new to
analysing static and dynamic structural mechanics most readers the basic workflow will be familiar and is
problems, steady-state and transient heat transfer shown in Figure 3 (note the absence of a mesh
problems and the ability to perform sequentially coupled generation phase).
thermal stress analysis.
The accuracy of the solution is largely driven by the
The tool is marketed as "Mesh Free" and the user can density of the non-conforming cells used in the initial
perform the entire simulation process without defining or discretisation phase. midas MeshFree contains controls
reviewing the discretised domain. MeshFree utilises allowing the user to manually specify the cell density for
concepts similar in principle to the Implicit Boundary every part used in the analysis. This control allows the
Method using a non-conforming mesh [3]. user to specify the number of non-conforming cells in
three orthogonal directions however, for the purpose of
This approach embeds the geometry of the part into a this evaluation the automatic grid density settings have
larger, regularly shaped computational domain which is been utilised. The users are still able to influence the
discretised using a non-conforming grid (Figure 1). accuracy of the solution but this is done by setting the
Putting it simply, MeshFree puts the model geometry "reference memory" that the simulation process is able
inside a regularly shaped domain and then breaks this to utilise (Figure 4). Increasing the memory available to
larger domain down into a 3D array of regular cells that the simulation process increases the number of cells in
do not align with the boundaries of the part that is being the non-conforming grid.

Figure 3: midas MeshFree workflow

4
 Designer Oriented Software – Is it Accurate? Part 2

Figure 4: Analysis control dialogue box

The memory settings are available in the "Analysis Control" During this evaluation, results were recalculated
dialogue box in the form of a simple slider. Leaving the slider using 0.5, 2, 4, 6 and 8GB reference memory
at the left-hand end of the scale will result in a quick settings. All runs were performed in-house by
solution but it should be less accurate than an analysis NAFEMS.
performed with the slider set at the right-hand position.
Additional settings can be used to increase the
Users can see the effect that the Reference Memory setting solution accuracy under certain scenarios. It is not
has on the model size by opening up the log file that is anticipated that a designer would modify these
generated as the solution progresses. The log file includes during a design exploration study. Unless otherwise
details of the density of the non-conforming grid and the stated, after discussion with MeshFree the
resulting number of degrees of freedom. While I would not modifiable solution parameters were set to:
expect a designer to spend time parsing this file for
information, the number of degrees of freedom have been • Automatic – Relative Grid Density = Medium
provided in the table that accompanies each benchmark.
• Geometry Representation Details = 0.1
This allows the rate at which the solution converges against
the number of degrees of freedom to be observed. The • Adapt Grid Orientation to Shape = Yes
density of the non-conforming grid can also be viewed
graphically as shown in Figure 5. • Enable Increased Integration Accuracy = Yes

Figure 5: Benchmark 1 showing non-conforming background grid

5
 Table 1: midas MeshFree - Benchmark 1 Details

The Benchmark Problems Benchmark 1 – Pressure Component

The benchmark problems are described more Benchmark 1 is a linear elastic stress analysis problem considering
comprehensively (including links to download an internally pressurised pipe. The solution quantity of interest is
the benchmark geometry) in a white paper the peak Von Mises stress on the inside surface of the main pipe.
accompanying this initiative [5]. This paper is The details of the analysis settings, number of degrees of freedom
available to NAFEMS members and collates in the model, predicted result and target solution are reported in
the results of the different codes that have Table 1.
been part of this evaluation.
Table 1 demonstrates that there is a linear relationship between the
Seven structural, static, linear elastic, small reference memory assigned to the problem and the number of
displacement benchmarks are considered degrees of freedom that are generated. Assuming that the degrees
where the target solutions are either a peak of freedom are being created in appropriate regions, it confirms
stress value (e.g. Maximum Von Mises Stress), that the reference memory can be considered as a simple control,
peak displacement value, or a spring rate. allowing the user to transition from a quick and coarse solution to a
Contour plots are supplied in order to allow the more time consuming and accurate analysis. This relationship
reader to see how the output in question is between reference memory and model size holds true for the rest
distributed throughout the component. One of the problems considered in this study.
linear dynamic, modal benchmark is also
considered where the target solution is the The coarsest solution setting produces a maximum Von Mises
first 5 fundamental modes of vibration. All Stress distribution that is within 1% of the target solution. The
target solutions have been produced using a contour plot of Von Mises shows a stepped variation that runs in the
traditional finite element approach, with axial direction of the main pipe. This effect reduces as the reference
confidence in the target solution generated memory assigned to the solution increases. The highly stressed
using mesh refinement studies. regions of the inner surface of the pipe are easily identifiable.

Figure 6: midas MeshFree – VonMises Stress -Contours scoped to 420-

6 540MPa – Iteration 1 Left Hand Side, Iteration 5 Right Hand Side

 Designer Oriented Software – Is it Accurate? Part 2

Table 2: midas MeshFree - Benchmark 2 Details

Benchmark 2 – Coil Spring The coarsest solution setting resulted in a spring rate
Benchmark 2 tests the ability of the designer- that deviated by 12.5% from the target solution. As the
oriented package to predict the compliance of the reference memory assigned to the solution was increased
coil spring shown in Figure 7. The details of the the deviation decreased. Interestingly iteration 4 with 6GB
models and predicted results are shown in Table 2. of reference memory provides a result that was almost
identical to the target solution while the more
The spring rate is obtained by loading the top computationally expensive 5th iteration was accurate to
surface of the spring with a 1mm displacement within 5.3%. This is a challenging benchmark and,
applied in the axial direction and extracting the understandably, automatically discretising the complex
resulting reaction force. The reaction forces are geometry is difficult. If the background grid density is
calculated in midas MeshFree by graphically manually set for each of the three primary directions then
selecting the desired surface and hitting the the solution quickly converges to the target value as
calculate option (see Figure 8). shown in the results of iteration 6-8.

Figure 7: Geometry of the coil Figure 8: midas MeshFree – Iteration 4 - Contours showing deformation –
spring used in Benchmark 2 Reaction force associated with extending the spring shown in dialogue box

7
 Benchmark 3 – Thin Skew Plate in Bending
The third benchmark considers a thin skewed plate (see MeshFree analysis runs are shown in Table 3, while the
Figure 9), restrained in a simply supported manner and distribution of maximum principal stress in MeshFree
loaded with a uniform pressure. The thin plate was analyses can be viewed in Figure 10.
selected to expose any deficiency in the numerical
method's ability to appropriately capture bending The peak stress in the midas MeshFree model was
behaviour of thin sections. extracted by generating a linear path between two
corners of the lower surface of the plate. Under the
With the traditional FE approach, the geometry would be coarsest solution setting the maximum principal stress
midsurfaced and then discretised using a shell element. in the plate centre was predicted with an error of less
This additional work is not required with midas than 1%.
MeshFree. The details and results of the midas

Figure 9: Benchmark 3 – Geometry

Table 3: midas MeshFree - Benchmark 3 Details

Figure 10: Benchmark 3 – midas MeshFree - Maximum principal stress on plate lower surface

8
 Designer Oriented Software – Is it Accurate? Part 2

Benchmark 4 - Stress Concentration - Plate

with Hole
The fourth benchmark tests the ability of the "designer
oriented" code to capture a stress concentration in a plate
containing a small hole. The benchmark has been
designed so that the extent of the plate is large in
comparison to the size of hole so as to pose a challenge
when sizing the mesh in the vicinity of the stress
concentration. The geometry used by benchmark 4 is
shown in Figure 11 with the analysis details and calculated
results available in Table 4. The target solution is the
maximum and minimum principal stress in the vicinity of
the hole. Many analytical solutions are available for this
configuration but care is needed when interpreting the
analytical solution as the traditional FEA approach takes
the through thickness variation in stress into account.

There is significant solution to solution variation in the

MeshFree results as the reference memory is increased.
As the solution does not coverage with increased cell
density confidence cannot be generated in the results
although, from inspection of the 5 solution iterations, the
predicted solution was within 14% for the maximum
principal stress, and 27% for the minimum principal stress. Figure 11: Geometry of the plate with hole benchmark

Table 4: midas MeshFree - Benchmark 4 Details

Figure 12: Deviation from target result against model size

9
 Figure 13: Principal stress results – Maximum (top), Minimum (bottom), Traditional FEA, Target Solutions (left), midas MeshFree (right)

Benchmark 5 – Stress Concentration the level of reference memory assigned to the solution
process. Viewing the midas MeshFree principal stress
Notched Shaft distribution it can be seen that the stress at the centre of
Benchmark 5 evaluates the ability of the designer the notch has a degree of variation with circumferential
oriented software to capture the stress concentration in a position. If the stress at the base of the notch is averaged
notched circular shaft loaded in uniaxial tension. The over the circumference of the shaft it is expected that the
model details and predicted results are shown Table 5. results would be a close approximation to the target
The maximum principal stress produced by the target solution. The variation in stress with circumferential
solution can be seen in Figure 14 with the midas position appears to present a challenge when predicted
MeshFree results shown in Figure 15 for Iteration 1 and the stress distribution on curved surfaces. From Figure
Iteration 5. 15 it can be seen that the circumferential variation in
stress reduces when increased reference memory is
The midas MeshFree solution overpredicts the peak assigned to the solution i.e. when a denser array of cells
principal stress by between 7% and 13% depending on is used.

Table 5: midas MeshFree - Benchmark 5 Details

10
 Designer Oriented Software – Is it Accurate? Part 2

Figure 14: Target Solution – Maximum principal stress

Figure 15: midas MeshFree – Maximum principal stress – 0.5GB Ref Memory Top, 8GB Reference Memory Bottom

11
Benchmark 6 - Natural Frequency Thin Cantilevered Plate
The sixth benchmark is designed to test the ability of the software to accurately predict the first five modes of vibration of a
thin square plate constrained to act as a cantilever. The solution details are provided in Table 6 with the mode shape
associated with the 5th mode of vibration shown in Figure 16.

The results predicted by the coarsest solution setting are a good match to the target solution with a discrepancy of
between 1.8% and 2.9%. Interestingly the predicted results do not vary significantly as increased reference memory is
assigned to the solution although the analysis diagnostic information indicates that the number of degrees of freedom
increase with the additional assigned memory.

Table 6: midas MeshFree - Benchmark 6 Details

Figure 16: midas MeshFree –Mode Shape 5, Analysis Iteration 1- Contours showing normalised displacement

Table 7: midas MeshFree - Benchmark 7 Details

 Designer Oriented Software – Is it Accurate? Part 2

Benchmark 7 – Cantilever Under End Load

Benchmark 7 considers a cantilever beam, fully built in at one end with the other end loaded with a force acting
perpendicularly to the beam axis. The target solution for benchmark 7 is the tensile stress found on the top surface, at the
constrained end of the beam. While this appears to be a trivial problem and analytical solutions to this problem are readily
available, a bending dominated problem has been included in this study as it can highlight deficiencies in both element
formulation and the refinement of the automatically generated mesh.

As the midas MeshFree solution converges quickly and only marginal changes in the solution quantities are
observed when additional reference memory is assigned to the problem only 3 solution iterations were run.

The simulation details for benchmark 7 can be found in Table 7. Directional stress is not currently an available output in
midas MeshFree but fortunately in this case the maximum principal stress (Figure 17) and axial directional stress in this
case should be equivalent. The predicted results for both stress and displacement are an excellent match to the target
solution.

Figure 17: midas MeshFree – Benchmark 7 – Solution Iteration 1– Maximum principal stress.

Benchmark 8 – Cantilever Under End Load – Stress Concentration

Benchmark 8 extends the geometry and loading used in the previous benchmark by building the geometry used in
Benchmark 7 into a larger structure. A fillet radius of 5mm is used to smooth the transition between cantilever and the
supporting structure. Benchmark 8 explores the ability of the designer-oriented software packages to capture the stress
concentration near the constrained end of the cantilever. The length of the cantilever and the relatively small size of the
fillet radius is expected to pose a challenge to an automatic discretisation process.

The target solution for this benchmark is the peak Von Mises stress in the model which should be found at the fillet
transitioning between the cantilever and the supporting structure. The solution details and predicted results can be viewed
in Table 8.

The convergence of the target solution as the reference memory assigned to the solution process is increased can be seen
in Figure 18. Here we can see that there is considerable variation in the result quantity of interest under the first two
solution iterations. The results range in accuracy from within 1% when a reference memory of 8GB is assigned to 18%.
From Figure 18 it can be seen that there is a considerable variation in the quantity of interest between solution iterations.
As the reference memory is increased, the results range in accuracy from within <1% when a reference memory of 8GB is
assigned to 18%.

Table 8: midas MeshFree - Benchmark 8 Details

13
 Figure 18: Deviation between peak Von Mises stress and target solution with increased degrees of freedom

Figure 19: Benchmark 8 – Von Mises Stress – Traditional FEA (right), Midas MeshFree, Iteration 6 (left),

Conclusion
A summary of the results from the eight benchmarks can solution even at the lowest accuracy setting (Reference
be found in Table 9. One of the appealing features of the memory set equal to 0.5GB).
meshFree tool is the ability to perform a simple
refinement study by increasing the reference memory While it was not possible to produce a converged solution
assigned to the solution. Where this approach has when running benchmarks 2, 4, 5 and 8, it should be
allowed a converged solution to be demonstrated, a noted that these problems were deliberately selected to
single value representing the converged results has been push the boundaries of a designer oriented software
included in Table 9. Where it was not possible to package.
demonstrate a converged solution, the range of results
produced under different grid densities has been The analyses run in support of this article have been
included. MeshFree produced results for benchmarks 1, produced using a maximum reference memory setting of
3, 6 and 7 that were an excellent match to the target 8GB. It should be noted that there is no limit for the

14
 Designer Oriented Software – Is it Accurate? Part 2

maximum memory that can be assigned to midas quickly as the midsurfacing preprocessing step is
MeshFree on a 64bit operating system and so the option effectively removed from the method workflow.
of allocating more reference memory is available if your
hardware is up to the task. NAFEMS members interested in further details of the
benchmarks are encouraged to read the "Designer
The analysis process is simple to use and requires no Oriented Software – Evaluation" white paper available at
specialist knowledge. The results for thin, flexible nafe.ms/designer and we hope to continue this series in
components appeared to pose no problem for the following issues of Benchmark.
numerical method and models can be turned around

Table 9: A summary of the midas Mesh Free results for all the benchmarks

References
[1]. I. Symington, "Designer Oriented Software Is it Accurate?", 16 January 2020. [Online]. Available:
https://www.nafems.org/publications/resource_center/bm_jan_20_1/. [Accessed 14 April 2020].
[2]. "MESH FREE makes you FREE," MIDASIT, [Online]. Available: http://www.midasmeshfree.com/. [Accessed 2020 April 14].
[3]. A. Kumar, S. Padmanabhan, Burla. R, "Implicit boundary method for finite element analysis using non-conforming mesh or grid",
Int. J. Numer. Meth. Engng 2008; 74:1421-1447
[4]. A. Düster, J. Parvizian b, Z. Yang, E. Rank, "The finite cell method for three-dimensional problems of solid mechanics", Comput.
Methods Appl. Mech. Engrg., vol. 197, pp 3768-3782 Aug 2008, 3768-3782
[5]. I. Symington, "Designer Orientated Software Evaluation – White Paper", 16 January 2020 [Online]. Available:
https://www.nafems.org/publications/resource_center/wp_jan_20/. [Accessed 14 April 2020].

About MidasIT
NAFEMS would like to thank Piotr Stepien and Sungwoon Woo for sharing their knowledge of the methods that underpin
the meshFree tool.

15