An Incomplete Guide to

ANSYS Workbench
Small Satellite Research lab

The University of Georgia

Version 1.0

Casper Versteeg, Christine Yang, John Courtney

May 14, 2018

1 Introduction 3

2 Creating Models 3

3 Model Setup 4
3.1 Engineering Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Mechanical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4 Structural Simulations 7
4.1 Modal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2 Harmonic Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 Response Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.4 Random Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.5 Inertial Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

5 Thermal Analysis 17
5.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Transient Thermal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Steady-State Thermal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.4 Results Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 Conclusion 24

1
List of Figures
1 Geometry Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Workbench Start-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Modular Model Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Engineering Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5 Mechanical Model Start-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
6 Mesh Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
7 Structural Analysis Project Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 8
8 Boundary Condition Geometry Denition . . . . . . . . . . . . . . . . . . . . . . . . 9
9 Modal Total Deformation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
10 Harmonic Response Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
11 Component Shock Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
12 Response Spectrum Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
13 Random Vibration Directional Deformation Results . . . . . . . . . . . . . . . . . . . 15
14 Mass Acceleration Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
15 Inertial Acceleration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
16 Geometry for Thermal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
17 Transient Thermal Project Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
18 Transient Thermal Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
19 Command Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
20 Transient Thermal Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
21 2-σ Corrected Temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

List of Tables
1 Component Shock SpectrumTabular Data . . . . . . . . . . . . . . . . . . . . . . . 13
2 Random Vibration Qualication ProleTabular Data . . . . . . . . . . . . . . . . . 14

2
1 Introduction
This guide is meant to help those completely new to ANSYS get acquainted with its major features
and get started on building a project. By no means is this guide complete (hence the "incomplete
guide") but although it skips over a lot of features, it should be comprehensive enough to give you
the knowledge required to gure it out. The very rst thing you will learn about ANSYS, it that it is
incredibly picky, and can really make you regret life if you do not dene constraints correctly. While
this guide is meant to help put you on track, every problem is dierent, and dierent methodologies
might apply.

2 Creating Models
The meshing tool in ANSYS is incredibly powerful, and its solver methods are well optimized. For
this reason, creating models for an ANSYS simulation is reasonably straight forward. As long as
there are not any terribly extravagant details, like threads or tiny holes for pin connections, your
model will most likely be ne. Really, a rule of thumb should be, that if the feature is about
a millimeter in side length or diameter, it is probably safe to remove it, unless it is structurally
signicant.

(a) Not Ok (b) Probably Ok

Figure 1: Geometry Comparison

Figure 1 shows a comparison of acceptable geometry that will probably mesh well to a geometry
that will probably not mesh well. The trick, here, is obviously all those holes that are normally for
the PC/104+ connector pins. While they might be structurally signicant, the ANSYS solver will
have signicant problems converging to a solution, and the mesh will likely contain thousands more
nodes than are maybe necessary.
Structural signicance is usually determined by either high elasticity in material, brittleness of
material or low strength, but every instance is dierent and should be considered based on context
rather than rules.
Any 3D model can be created in Inventor, or other CAD software for simulation purposes. For
this guide, we will be using a Z-Magnetorquer board, as shown in Figure 1 (note: this component

3
is not actually used, but is good to show you how to operate ANSYS, simulations shown here are
for demonstration). The best practice is to export any CAD models to .stp, STEP le, which are
essentially text instructions for CAD software or interpreterslike ANSYSto recreate geometry.

3 Model Setup
When you open ANSYS Workbench for the rst time, it will look a lot like Figure 2. The project
schematic area is where models are built and set up. The "Analysis Systems" on the left may look
dierent for you depending on what license of ANSYS you are running, but the main components
discussed here should all be present.

Figure 2: Workbench Start-up

It is generally good practice to keep ANSYS models modular. This means, instead of just
dragging and dropping the analysis types into the schematic and working them all separately,
instead building the model with common elements going to dierent analyses. This would make the
model look something like Figure 3. The components A and B can be found by scrolling down to
the "Component Systems". Components can be connected after dropping by simply clicking and
holding the parts that need to be connected, and dragging the connection to the next component.

4
 Figure 3: Modular Model Building

3.1 Engineering Data

The "Engineering Data" component in ANSYS is where material and uid properties can be dened.
As opposed to all the other modules we will be seeing in this guide, the engineering data simply
opens a new tab in the workbench, named according to what parts are connected toand thus
dependent onthe engineering data (this name can get pretty obnoxiously long for large models).
By default, for a completely new Workbench model, the only material dened is "Structural
Steel". Just remove that. By clicking on "Click here to add a new material" you canyou guessed it
add a new material and dene its physical properties. We are trying to simulate the Z-Magnetorquer,
which, for all intents and purposes, is just FR-4. Because we will initially only be looking at
structural models, the only thing that needs to be dened here is density and linear, isotropic
elasticity. These can be found under the "Toolbox" menu to the left by expanding the relevant
menus. If you want to run a thermal model, you will need density, isotropic thermal conductivity
and specic heat, as well.
Fields that require data, under the "properties" will be highlighted in yellow. For now, just
dene a density ρ = 1850 kg · m−3 . As for the isotropic elasticity, because the various properties in
a linear material are linked, you can tell ANSYS to calculate any of them based on just inputting
two, just depending on what data is readily available. This can be changed by changing the input
in the drop-down menu next to "Derive from". For now, dene the Young's Modulus (or elasticity)
E = 22000 MPa and Poisson's Ratio ν = 0.118. The units can be changed by clicking on the
units next to the quantity for any of these, and selecting desired units. Once the quantity has been
entered, ANSYS will automatically do the conversion if the units are changed. Your window should
look like Figure 4

5
 Figure 4: Engineering Data

Of course, if desired, more material data can be dened here, but the analysis in this guide will
be fairly simple just to expose you to the various options available. Now, do not close the window,
this will exit out of ANSYS. Instead, close the Engineering Data tab, this should bring you back to
the project schematic.

3.2 Mechanical Model

While ANSYS has the ability for you to draw geometry, this generally does not work very well, and
it is recommended to just import CAD made in a program that is designed better than ANSYS'
DesignModeler. Trust me, don't hurt yourself, just import. Always import.
To import geometry, right click on the "Geometry" eld under Mechanical Model. Select Import
Geometry, and then "Browse...". Navigate to the .stp of the geometry you wish to analyze. The
icon next to "Geometry" will change, to indicate you have added an external .stp le.
Double click on "Model" in the project schematic. This will open the Mechanical Model. Some
dialogs will appear to tell you that ANSYS is importing the geometry you have previously attached,
and then the window should look something like Figure 5.

6
 Figure 5: Mechanical Model Start-up

Notice the question mark (?) next to "Geometry". This means that some user input is required.
If we expand the geometry menu, there will be a list of all the components that make up the
assembly that is being modeled. Currently, this is only one part, but there could be, like, a lot.
By clicking on the parts that have question marks next to them, this brings up the details of the
part in the bottom left. Under "Material", note it is likely missing "Assignment". This just means
the part needs a material to be assigned to it. If you click on the yellow eld, this will bring up a
drop-down menu of all the materials created in the Engineering Data section.
Once all the parts have materials assigned to them, the Geometry menu should have a nice,
satisfying green tick next to it. We can now move on to creating the mesh of the part. This is
really why simplifying the geometry is denitely recommended, because this can get seriously out
of control if you don't. Click on "Mesh"; under "Details" we can see a bunch of meshing options
that will either make your mesh more course or more ne. If the mesh is giving your trouble,
your geometry is probably the culpritANSYS is smart enough to determine the optimal meshing
parameters, so if errors pop up, it is probably your fault. Either right-click on "Mesh" and click
"Update", or nd the button in the top ribbon with the yellow lightning bolt that says "Update".
This will generate the mesh.
Your Mechanical window should look something like Figure 6(a). The view here is to show you
why simplifying the model is important. Figure 6(b) shows what the mesh would have looked like if
the original model (Figure 1(a)) had been used. For unsimplied models, for one, the meshing will
take signicantly longer. For comparison, the unsimplied model has almost 90,000 nodes (41,000+
elements), whereas the simplied model has less than 3,000 nodes and about 350 elements. While
more nodes and elements will give you a more exact result (which is why you can mess with the
mesh parameters), computing the results of the analysis will take your entire life. Here, speed is
preferred over accuracy, but exact settings should always depend on your model's requirements.

4 Structural Simulations
Structural simulations always come in four parts: Modal, Response Spectrum, Random Vibration,
Harmonic Response and Inertial. The modal analysis should always precede the response spectrum,
random vibrations and harmonic response analyses, as the dynamic results from the modal analysis

7
 (a) Simplied (b) Unsimplied

Figure 6: Mesh Results

are used in the proceeding analyses. The setup for each of these analyses will be dierent, so they
will all be covered separately. To create the overall model, simply drag the additional components,
and drop them on the "Solution" eld of the modal analysis. The inertial (which is a static structural
type of analysis) does not require the modal as a prerequisite, so it can be dropped on any "Model"
eld. You can always drag the analyses around to clean up the project schematic. For structural
analyses, the model setup should look something like Figure 7.

Figure 7: Structural Analysis Project Schematic

Note how all the analyses will either have the lightning bolts or question marks, denoting they
require additional setup. These will all be covered separately here. This is a good time to save your
project, and for the love of science, save frequently!

4.1 Modal

The modal analysis, as well as the response spectrum random vibration and harmonic response,
are vibration-related analyses. A mode, in continuum mechanics, is analogous to the natural (or
fundamental) frequency of a spring-mass system. In essence, all nite element analyses are, are

8
spring-mass-like analyses. The process of meshing pretty much reduces a continuous system to a
bunch of masses, connected by springs which the underlying solver will then use to compute specic
kinds of results. A mode, much like a natural frequency, is the forcing frequency at which your
structure experiences a local maximum in deection, at some point in the structure. You will see
what this means when we complete the analysis.
Open the modal analysis "setup". Another Mechanical window will appear, but now showing
all the dierent types of analyses that are also connected to the modal analysis. We can stay on
this window for a good while.
In general, the rst thing you should always be thinking about at this point, is "what are the
boundary conditions of this system", meaning, what specic kind of displacement, or temperature,
or current or ow or whatever do you know about the system at this point. When talking about
structural analyses, the boundary conditions are generally "Fixed Supports". A xed support is
essentially where the displacement of any node on the model gets set to 0 (zero).
The best way to add any kind of support, is to right-click on the analysis, and select "Insert" >
"Fixed Support". Now, this xed support is going to require a denition based on the geometry of
the part, essentially where do we set the displacement to zero. For this, we select the inside surfaces
of the four PC/104+ holes on the board, as shown in Figure 8.

Figure 8: Boundary Condition Geometry Denition

Sometimes, you might need to dene a "pre-stress", for something like fasteners, which are
already under tension before they are subjected to launch stresses, but we will not bother with that
now (you can generally get away with not modeling fasteners in ANSYS at all, there are equations
you can use for fastener analysis to give you some conservative estimates). Under the analysis
settings, you can select "Max Modes to Find". This defaults to six, which is a pretty good number.
A higher number of modes will take more time to solve, but will also make the subsequent analyses
give better results. Six is a happy middle-ground.
Realistically, we could call it done here, as the modal analyses will produce a bunch of dynamic
results mostly used for further analyses, but let's set a solution type here, too. Under modal, right-
click on solution, and insert a "Total Deformation" solution type. Now, this is important: the
results you get from this, are completely meaningless. They are what is called mass normalized,
which is a solution technique used in vibrations used to decouple degrees of freedom. If you want

9
to know more, take the class. Suce to say, that this will tell you the shape of the structure's
deformation under vibration. You can add multiple "Total deformation" solutions, and make them
correspond to dierent modes in the details menu, up to as many modes as you have the analysis
nd (i.e. if you are solving up to the sixth mode, you can only have six total deformation solutions).
Beware: you will only need to extract the frequencies from this analysis!
You can add whatever kind of solution types you want to this analysis to ndalthough the
most important ones are coveredright-click on the solution under modal, and click "Solve". This
will evaluate the modal results for this structure. Depending on complexity, number of modes, and
number of elements, this can take a while. When it nishes, in the bottom right, there will be a
table with the mode number, and what frequency it occurs at. For this particular model, the rst
mode is at 684.67Hz, which is fairly highin modal analysis, high frequencies are good. It means
that you will need to excite the structure with very short period, which takes a lot of energy, and
is hard to do. Depending on how many total deformations you have the analysis nd, you can see
what the deection would look like. The results for this model are shown in Figure 9.

Figure 9: Modal Total Deformation Results

Note that, in the ribbon directly above the project tree, there are several drop-down menus next
to "Result". This is where you can change the way solutions are displayed. You can change the scale
of the deformation shown, change the type of contour, and add max/min markers to the model.
Also note, that the "min" deection marker is at one of those holes we set as xed supports, which
makes sense because the deection will be zero there, which is the lowest magnitude of deection
in this model. With this done, we can move on to the next solution, which is related to the modal
solution: the response spectrum.

10
4.2 Harmonic Response

The harmonic response will give you a much better idea of how your structure is going to be behave
under uniform vibration than the modal analysis, and the two should be complementary. The
harmonic response does require the modal analysis as a prerequisite, but it returns deections of
your structure that are not mass normalized. Rather, it will use the dynamic behavior prole from
the modal as input data, and returns the actual deection of the structure under vibration. What
the harmonic response does, is force excitation at the xed supports dened in the modal analysis
at a range of frequencies. Essentially, it uses the equation f (t) = F sin(f t) to vary the forcing
sine wave over a range of frequencies. You will see, over dierent frequencies, you will get dierent
responses.
The rst thing you should see in your harmonic response, is that the modal data is being
transferred in. There is then also a question mark by the analysis settings menu. The only input
required here, is a range of frequencies you want to see evaluated in your harmonic response. For
demonstration purposes, the range here will be set from 0 to 1000Hz, to be able to show you at least
the rst three modes. This analysis can be quite slow, so you should evaluate how many solution
intervals (in Herz) you want. Here, it is set to 10000 (more points will make this take longer, but
will give you a smoother curve).
The next thing is to dene a forcing boundary condition. Basically, we will add an inertial
acceleration here. Go up to the environment bar to the top left, and add Inertial > Acceleration.
Under the denition for your loading, set "Dene by" to "Components", and we're going to set the z
component to 1mm · s−2 , however, you should perform this analysis on all axes separately (x, y and
z ) to get a complete picture of all possible responses. The act of setting this to 1mm · s−2 , essentially
normalizes our response to the forcing amplitude, meaning that the actual response amplitude A
will increase linearly with the amplitude of the forcing sine wave. We will see how this aects our
results later.
Under solution, add a "Frequency Response" deformation solution. This will require a piece of
geometry to parse for deformation, you can select all geometry in your part by using Ctrl+B to
enable the body select tool, and then Ctrl+A to select all bodies in your analysis. Or, alternatively,
you can just select the ones you are interested in, but since your entire body will be under excitation,
a good rst step will be to select all bodies, and trim it down from there if it's taking too long.
Apply this geometry, and your simulation should be ready to go. Right-click the solution, and click
Solve. This is going to take your structure, and evaluate it with a forcing amplitude F = 1mm · s−2 ,
at the range of frequencies f you have specied.

(a) ANSYS (b) MATLAB

Figure 10: Harmonic Response Results

11
 The results of this analysis are presented in Figure 10. The left is produced directly by ANSYS,
which has the y -axis on a log-scale (you can turn this o, but presentation of graphs using MATLAB
is always preferred, and less messy), and the right is produced with MATLAB on a linear scale.
From the left graph, you can clearly see a spike in amplitude at just under 700Hz (684.67Hz, to be
exact). This is the response amplitude from the rst mode. The other spikes you can observe at
slightly higher frequencies, are the second and third modes.
Note from the y -axis label on the MATLAB curve, that the response amplitude looks like | FA |.
This is what "normalized by the forcing amplitude" means. Essentially, because the frequency
response "transfer function" is a linear operation, the response amplitude scales linearly with the
forcing amplitude. So if this system were forced with 2mm · s−2 , the response amplitude would scale
with the same factor 2. If you can quantify what the acceleration of the system is, you can quantify
what the response amplitude will be.

4.3 Response Spectrum

The response spectrum analysis is somewhat hard to explain. Essentially, it is a type of shock
analysis, and shock is a type of vibrational excitation. Essentially, given an excitation prole, this
will give you the peak deection your structure will see in the environment described by your prole.
It is not like modal deection, but its results depend on what the modal analysis looks like.
To begin the setup, there is a pre-determined shock spectrum supplied to us. Start by adding
a "RS Acceleration" boundary condition to the analysis. Because the analysis should be linked to
modal, it will take the xed supports from the modal analysis. In the RS Acceleration details, next
to "Boundary Condition" select "All BC Supports". This essentially forces the system at the bases,
which is conventionally how a shock analysis is set up. Next, the "Load Data" eld is also yellow,
this means we will have to supply a table of values of frequency and acceleration. These values are
given by the graph in Figure 11. For convenience, the data for the table is also provided in Table
1, where the acceleration in G's from the graph has been converted to m · s−1 .

Figure 11: Component Shock Spectrum

12
 Frequency (Hz) Acceleration (m · s−2 ) Acceleration (mm · s−2 )

100 539.55 593.55 · 103

Table 1: Component Shock SpectrumTabular Data

Lastly, the details for RS Acceleration require a direction in which the forcing load is applied.
Because launch orientation is not generally known beforehand, the analysis (including all other
analyses that are orientation dependent) will need to be run for forcing in all directions, x, y and
z . For now, this example will be run with the forcing direction along the z -axis.
Under the solution for Response Spectrum, add a "Total Deformation" solution, and an "Equiv-
alent Stress (Von Mises)". The stress will allow us to determine whether the part is likely to fail
under shock loading. No other input is required at this point, so you can solve the model.

(a) Deformation (b) Equivalent Stress

Figure 12: Response Spectrum Results

The results are shown here in Figure 12. Note how the deformation gure has the exact same
shape as the modal analysis' rst mode result. This should tell you, that the structure responds
at a frequency lower than the rst mode. Interpreting the data from this analysis can be quite
tricky. It is not as straight-forward as saying that this is exactly what is going to happen to your
structure. You have to remember, shock will only really occur a handful of times on your structure,
and neither you, nor ANSYS, know at what frequency this will occur. If your modal analysis is
predicting a high frequency for the rst mode, you can rest assured that the response predicted by
your response spectrum analysis will be an absolute worst-case scenario; your structure will likely
not see these kinds of stresses, or deections. The big take-away, is that response spectrum results
will give you large numbers, but it should be on you as an engineer to interpret how much you ought
to worry about that.

13
4.4 Random Vibration

The phenomenon of random vibration is a statistical one. Think about a series of random numbers
in some interval, plotted against time. If you were to extract the frequency information from that
(using something like a Fourier analysis) you can't. All frequencies are present in a randomly
generated signal. This means, that if your structure is in a scenario where it is not being uniformly
forced, but instead has some random forcing component (very common in rocket launches), your
structure is going to respond to every single forcing frequency at once. Let that sink in for a second.
In random vibration, your system responds to every forcing frequency simultaneously, including the
modes.
The way to quantify this interval of forcing amplitudes (the range of random numbers), we use
Power Spectral Density, or PSD. This can be given in terms of acceleration, displacement or velocity,
but essentially quanties the amount of average power present in the frequencies (don't worry about
what that means). In your analysis, add a "PSD G Acceleration" boundary condition, and set it to
all xed supports. Then, add the data specied in Table 2 in the tabular data section, and set the
direction to the axis you want to evaluate (again, you will need to do all of these separately).

Frequency (Hz) G Acceleration (G2 · Hz−1 )

20 0.026
50 0.16
800 0.16
2000 0.026

Table 2: Random Vibration Qualication ProleTabular Data

The random vibration analysis, because of its statistical nature, produces statistical results,
evaluated along a single axis. So, even though you have set the forcing axis to one, you will have to
add three "Directional Deformation" solution types. Each of these, will be set to one of the three
axes. So the rst will be set to x, the second to y and the third to z , or whatever order you want, as
long as all axes are represented. Because we are dealing with probabilities, you will have to dene
your condence interval. To assure the structure is going to behave well under random vibration,
set the "scale factor" to "3 Sigma". This means that you have 3-σ condence, or 99.73%, that the
actual deection will be lower than predicted by this analysis.
Once you are set up, run the analysis. The results for this demonstration are shown in Figure 13.
Note how there are three separate gures, each representing an axis. Generally, random vibration
results are rather small if the rst mode is very high. For a dierent system with a low rst mode,
this could be in the order of several millimeters, so it depends entirely on the modes. Also note, the
z -axis deformation shown here, has the exact same shape as the modal analysis showed. These are
good checks to make sure your results make sense, and you have not made a mistake somewhere.

4.5 Inertial Acceleration

The last type of structural simulation is probably the simplest one, plain old acceleration. Since we
know that accelerating applying a force on a mass (like thrust from a rocket) causes it to accelerate,
and vice versa, we can quantify the acceleration from the rocket on our structure, and determine
the deection.

14
 (a) +X (b) +Y

(c) +Z

Figure 13: Random Vibration Directional Deformation Results

We will use the Static Structural analysis type for this, since inertial acceleration is also called
static acceleration. Add the same boundary condition as for the harmonic response, and set it to
dene by components. Now, if you thought this was straight-forward, you would be wrong. As
it turns out, structures of dierent mass will experience dierent accelerations in the same rocket.
Meaning, your 4kg CubeSat will accelerate dierently from the Hubble Space Telescope, which has
a mass of about 11 tonnes. This acceleration is specied by the Mass Acceleration Curve (MAC),
shown in Figure 14.

15
 Figure 14: Mass Acceleration Curve

However, the MAC can be very inaccurate at low masses. It is recommended to use the accel-
eration corresponding to 8kg to anything that is lighter (or approximately 33G = 323.73m · s−2 =
323730mm · s−2 ), even though a 4kg satellite should accelerate closer to 42G's (42G = 412.02m · s−2 =
412020mm · s−2 ). Whatever the satellite mass, as predicted by the Mass Budget, should be used
rst to evaluate the acceleration per the MAC. The larger acceleration should be used as an initial,
more conservative benchmark, although both can be checked if necessary.
Because this is not a vibrations analysis, the xed supports from the modal do not carry over,
and will need to be redened. You are free to copy the xed supports from the modal, and paste
them in the inertial. Then, all there is left to do, is add a total deformation and equivalent stress
solution type, and run it (again, remember you have to run this analysis with the acceleration along
each dierent axis, and present all results).

(a) Deformation (b) Equivalent Stress

Figure 15: Inertial Acceleration Results

The results are presented in Figure 15. Note the shape of the deformation is completely dierent
from the vibration analyses, which should make sense. The inertial deformation and stresses are a

16
component on top of everything that happens due to vibrations, but combining all these results is
not as straight forward as just adding them. It should be up to you as the engineer to determine
whether these results are acceptable. The usual way to do this, is by using a factor of safety. this
is given by the equation nd = losspredicted stress
of function stress . The loss-of-function stress, is usually the yield
stress of the material, although the ultimate strength is also used. A minimum nd of 2.0 is required
for yield, and 2.6 for ultimate. Material strength specication can be readily found online.

5 Thermal Analysis
Thermal analyses are somewhat dierent from their structural counterpart. In structural models,
we can see things deform and have a much better intuition for something being correct. Thermal
models are quite dierent, since it deals with the more abstract quantity of temperature. Especially
in satellite applications, where convection is not present, results can be somewhat counter-intuitive.
We must begin, by adding the thermal material property data to the engineering data. As
discussed in Section 3.1, we will need density (which we have from the structural model), isotropic
thermal conductivity (technically FR-4 is anisotropic because it is a composite material, but isotropic
is a good enough approximation for now) and specic heat. Once these have been added to
the material data, set the conductivity to k = 0.294W · m−1 ·◦ C−1 , and the specic heat cp =
1150J · kg−1 ·◦ C−1 . With the material data in place, we can begin the thermal analysis.

5.1 Model Setup

Besides the steps needed in model setup dened for structural analyses, some more might have to
be done for thermal analyses. For structures that have enclosures, meaning any surface of material
that "sees" another surface that is not at the same temperature as it, these must be properly dened
so that all the view factors are accounted for. Your working geometry might not require this, but
in the case that it does, this tutorial will go over the setup of how to work with it.

Figure 16: Geometry for Thermal

17
 Figure 16 shows the new working geometry, which simply consists of a new piece of material
(could be another board, who knows) oset from the Z-magnetorquer. They will have the exact
same thermal properties, for simplicity.
Once the mesh has been created, we will need to create named selections. To do this, go into
the model where you would normally be creating your mesh, right click on "Model" under Project
in the Mechanical window, go to Insert > Named Selection. From here, select all surfaces that
make up a single enclosure. If there are multiple enclosures, you will have to make multiple named
selection.
If you have more than one enclosure, you will have to do some fancy stu. Namely, you will
have to add a few lines of APDL command code. This can be accessed by right-clicking on your
analysis, and inserting a "Commands" environment. There is a supplementary document, provided
by ANSYS, that tells you how to do this, but it is summarized here. Each enclosure gets a number.
Since we only have two surfaces that see each other (therefore one enclosure), and we are assigning
them the same emissivity (the fraction of thermal radiation a body emits relative to a perfect emit-
ter, i.e. a black body), we can do this with two lines:

SF, <named selection>, rdsf, <emissivity>, <enclosure number>.

The SF command species surface loads on nodes (i.e. a surface-to-surface type radiation, and
the rdsf label species the surface-to-surface condition. By default, the named selection will be
called "selection" but can be named whatever you want. With a single surface-to-surface condition,
you do not have to do this, because it is automatically designated enclosure 1.
Once all enclosures have been designated with the command line above (all of them have to live
in the same command environment as separate lines), you can specify radiation boundary conditions
as surface-to-surface using the appropriate enclosure number. Do note, that each surface may have
only one radiation condition assigned to it. This means that a surface-to-surface condition cannot
overlap with a "to ambient" condition, since the surface-to-surface condition should be specied as
"open" if it also sees the environment.

5.2 Transient Thermal

A transient thermal analysis essentially tells you what the temperature prole of a structure looks
like over time. Unlike structural models, which use structural mechanical boundary conditions and
forces as loads, thermal models will have temperature-specic boundary conditions and heat ow as
loads. The governing equations for structural and thermal models, are quite similar, but dierent
in certain subtle ways that warrant separate explanation in this guide.
Ignoring the structural models, and making the edits to the geometry and engineering data
outlined in section 5.1, the model should look something like Figure 17. Your geometry might not
look exactly like Figure 16, but this guide should serve as a general setup procedure for similar
project.

18
 Figure 17: Transient Thermal Project Schematic

Having followed along with section 5.1, and done the named selection setup portion, your screen
should look something like Figure 18. The rst thing you should be thinking about, is what is the
initial temperature of your simulation. The initial temperature is going to determine how much
time it is going to take for your model to converge to a steady temperature. For this model, we will
set the initial temperature at 20◦ C.

Figure 18: Transient Thermal Setup

Even though this model will only have one enclosure, this will be a good opportunity to show
you how this might work. Right-click on the transient thermal project, and insert a "Commands
(APDL)" setting. This should bring up a text editor window for the scripting. It should look
something like Figure 19. Since the default name for a named selection is "Selection", and this was
left as default for the purpose of this demonstration, enter the following line in the command win-
dow (again see Figure 19, if you have multiple enclosures, you have to do this, just keep adding lines):

19
SF, Selection, rdsf, 0.9, 1.

Figure 19: Command Window

Next, we will think about the boundary conditions before we work on the analysis settings. The
reason for this, is that it is likely easier to quantify your loads rst, and come up with a prediction
of what your temperature prole might look like marching though time. This will then make it
easier to set your step and solver controls to the proper settings. It is unlikely that you will get the
analysis settings right the rst time, but this will minimize the amount of time you might spend
messing around with it.
The thing about ANSYS, much unlike Thermal Desktop, is that it will come up with conduction
models for you. Thermal Desktop requires you to specify all the contacts, and give them some inter-
face conductance, which ANSYS does not. ANSYS basically takes the mean thermal conductivity
value (k ), and assigns that as the interface conductance. This might be inaccurate, and so you
might have to change this in the "Connections" menu to the left.
The other kind of boundary condition that exists, is radiation (since convection is not present in
space, although adding convection conditions is very easy if you get the general setup of boundary
conditions). Add in a radiation boundary condition, by either clicking the "radiation" button
in the environment bar, or right clicking on the transient thermal project, and inserting it that
way. Let's start with the surface-to-surface radiation condition. Since we already made the named
selection earlier, change the "scoping method" under scope in details, to "Named Selection." Then,
simply select the named selection you want this to apply to. Change "Correlation" from ambient to
"surface to surface", and make sure the enclosure number and emissivity are the same as assigned
in the script line. If the enclosure is "open" (meaning some of the surfaces can still "see" the
ambient temperature, make sure the enclosure type is set to "open" as well. This will require
you to set an ambient temperature, since the net radiation heat transfer depends on this. The
ambient temperature is a dicult parameter to quantify correctly. A lot of back-and-forth between
simulations might need to happen for you to manually converge this to a single number (or set of
numbers). The recommended way to do this, is make a relatively simple, but representative model

20
in Thermal Desktop (or MATLAB, if you are not yet in this stage of modeling), and take a low
bound, and high bound of your temperatures, and run this simulation twice, once for each bound.
This will be similar to the way your hardware should be tested in thermal vacuum, but can be time
consuming. It is what it is. For the purposes of this model, the ambient temperature will be set to
0◦ C, which is fairly common for a low-bound temperature extreme.
Since we do not have any more enclosures, the rest of the surfaces will be set up as radiating
to ambient. It should be noted, that the entire satellite structure is not easily modeled in ANSYS,
and should therefore be reserved for Thermal Desktop (i.e. only model components inside the
frame in ANSYS thermal simulations). Similarly to the surface-to-surface radiation, the ambient
temperature will be taken as 0◦ C with 0.9 emissivity (specialized black paint). When selecting
the geometry this applies to, select every surface on the model, except the ones you have already
assigned to another radiation boundary condition. You are not allowed two boundary conditions
on the same surface, just like you cannot have a xed support and a displacement on the same face
of a structure. Your solver will not know which one to chose, and it will yell at you if you try to do
this.
Lastly, we will add a load to our structure. Heat loads come in a few dierent avors. There
is "heat ow" which is generally the simplest to use, you basically apply some amount of Watts
(or heat power) to a surface, and it will equally distribute it (this is very nice for electronics and
components), there is "perfect insulation" which basically sets the heat transfer on some area to
zero (you might think of radiation shielding for this), "heat ux" which is heat power per unit area
(think of the solar heat ux incident on the satellite when it is not in eclipse, i.e. heating rates
depend on the prole of the satellite shown to the sun), and "internal heat generation" which is
heat power per unit volume (you might think this is easiest to use for electronics, but it can be
obnoxious to keep track of all the volumes of your parts to normalize them, and they are usually so
small that it does not matter if you use heat ow). Go ahead and add a "heat ow" do the bottom
of the second board (i.e. not the Z-magnetorquer). Make sure that the magnitude of the heat load
is set to "constant" (so that it says "step applied") and not tabular (unless you somehow wish to
do this, but that will be slightly more advanced). Specify the value to 3W.
There is another type of boundary condition, which is a xed temperature. This is not commonly
used for real applications, but could be useful for troubleshooting your model. If you were to draw
a box around your model, and gave that a xed temperature, and included it as a surface-to-surface
radiation type, instead of to ambient on your model, the results should be the same.
In transient thermal, there are a few solution types that might make sense for you to add.
Temperature is the obvious one, but for design and possibly troubleshooting reasons, you might
also want to add a "total heat ux" solution type.
Now that all the boundary conditions have been set, the model is almost ready to go. Remember,
that we skipped over the analysis settings earlier, so we need to return to that. The main thing
you should be doing here, is messing with the amount of time you are running the solution for.
You can change the solver controls and convergence criteria, but the default should be adequate for
most models, unless you are trying to speed it up while sacricing some accuracy. For this model,
we will specify the "Step End Time" to be 2000s (which was previously determined through some
trial and error). You can decide how the solver marches your model forwards in time by setting
auto time stepping on or o. If you set it to "on", ANSYS will determine the time step based on
the rate of change of the temperature (if the temperature's "slope" starts to change, then it will
start to use more points to evaluate your model, whereas it will use fewer points if the temperature
slope is somewhat linear), but we will see what this means. If you want your temperature and heat
ux returned, say, every second, you can turn the auto time stepping o, and specify the time step
manually. We will leave this on for now. Your model is ready to go, so go ahead and solve it.

21
 (a) Colormap Plot (b) Time Response

Figure 20: Transient Thermal Results

The results of this model are shown in Figure 20. On the left is the colormap plot generated by
ANSYS. You will notice, at the bottom of your solution screen in ANSYS, there is the exact same
curve as it plotted in MATLAB here, along with some tabular data. The tabular data was used to
create the MATLAB plot shown to the right. You can use the graph to control which time step is
shown in your colormap plot, and you can play this out as a movie, and save it as a motion-picture
le format if you so desire.
Some important things to think about here: what would have happened if we did not specify
the surface-to-surface radiation condition? ANSYS would basically not know that the two pieces
of material were able to "see" each other. So the Z-magnetorquer would basically cool down to
the ambient temperature, which is clearly not the case in this model (it is kept slightly warmer
by the radiation coming from the other board). What would have happened if we specied the
initial temperature to be dierent? Say we set the initial temperature also to 0◦ C, then we would
basically see both parts temperatures increase, since they will always reach their steady-state at
the same temperature, regardless of initial conditions, it might just take them longer to get there.
You are encouraged to play around with all the dierent parameters here (just as you are with
Thermal Desktop), to get a feel for how a certain value changing one way or another, changes the
temperature prole you observe.

5.3 Steady-State Thermal

Once you understand the gist behind transient thermal models, setting up and running a steady-
state thermal model is quite straight forward. Actually, it is completely the same. You can use the
exact same boundary conditions in a steady-state model, as you can a transient model.
However, there are some subtle dierences in steady-state models you should be aware of. The
main one being that you will not get to see how your model's temperature changes over time. As you
might be able to tell from the name, a steady-state model basically just tells you the temperature of
your model as time goes to innity (or goes suciently large that your solver stops seeing variations
in temperature and will call it steady-state to within some convergence criteria).

22
 With your model going to steady-state, your initial temperature condition does not matter.
Since time is a removed component (it will not even tell you how long it will take for this to happen,
it just returns temperatures) all the transient eects, like initial temperature, will no longer be
present, rather it will simply be whatever your very nal temperature will look like.
The main reason for running a steady-state thermal model over a transient, is speed. A steady-
state model will give you less information about your model than a transient simulation, but steady-
state models can be orders of magnitude faster than transient. A transient simulation basically has
to solve the same equations as the steady-state model, over a specied time-step, just tens to
hundreds of times over.
A few notes about the steady-state results. If you let your transient model run for long enough,
you should get the same prole, and temperatures as a steady-state model. However, the actual
numbers might dier ever so slightly. You have to keep in mind, that ANSYS is a numerical
solver, and is thus subject to error and conversion intervals. This might mean that your transient
temperatures are actually slightly higher, than the steady-state predicted temperatures. As long
as this is within a hundredth of a degree (higher or lower), this is likely just that numerical error
rearing its ugly head, but if it is in the order of degrees, you might need to check you models.

5.4 Results Interpretation

It was mentioned earlier in this guide, that quantifying your ambient temperatures can be quite
dicult. In general, thermal models are quite hard to run, and specify your boundary conditions
correctly. If you are taking data from programs like Thermal Desktop, or even MATLAB, the actual
temperatures are very sensitive to the time-step used in your model, as well as as a multitude of
other factors.
To account for the things you might not be certain about, it is common to "pad" your temper-
ature data with some empirically determined interval to make sure that your data will fall within a
certain condence interval. Yeah, you didn't expect statistics to pop up here too, did you...

Figure 21: 2-σ Corrected Temperatures

23
 The extra step here, is rather simple. Essentially, per this paper from the Aerospace Corporation
on testing for space vehicles, the interval you should use for your temperature data to bring it
within acceptance-test levels, is 11◦ C. This means, take the highest temperature predicted in your
model (easily acquired from steady-state thermal), and add 11◦ C to it. Then also take the lowest
temperature predicted, and subtract 11◦ C from it. This will be the 2-σ condence interval corrected
temperature range you might expect your hardware to see, based on the models you have run. This
interval is to account for errors in the model at every step of the way.
The corrected data will look something like Figure 21, as applied to the transient thermal results
from Figure 20. There are other ways you might present your data, based on what you need. If
you are running steady-state thermal models, you can get away with simply adding or subtracting
11◦ C. Data interpretation is an important aspect of doing good engineering work , so
always ask yourself if your data makes sense, and if the way you have set up your simulation makes
sense.

6 Conclusion
If you made it to the end of this guide, congratulations. I hope you are slightly more comfortable
with ANSYS because of it. As with the Thermal Desktop guide, this guide is by no means designed
to introduce you to every single option available in ANSYS, but it should expose you to enough
stu, and give you the tools to go and explore for yourself.
ANSYS will most likely yell at you for doing something wrong at some point in your time using
it. This happens. ANSYS is nicky and particular about how it wants things. If it yells at you, ask
someone how to x it, or toss a quarter into the Google-o-matic, and hope to science someone on
the internet will tell you how to x it.
You should beware. ANSYS is an incredibly powerful tool. It will solve any system you give
it. However, whether that model you give it is correct to begin with might is up to you to determine.
If you give ANSYS an incorrect model, it will give you incorrect results. Use your intuition. If your
results seem wrong, try to evaluate why they are wrong, and take careful, procedural steps to try
to x your results. Talk to experts. Do what you need to do to make sure your models are correct.

24