1999-01-2126

Integrating Thermal and Structural

Analysis with Thermal Desktop™
Tim Panczak
Mark J. Welch
C&R Technologies, Inc.
Littleton Colorado, USA

Copyright © 1999 Society of Automotive Engineers, Inc.

ABSTRACT the analysis models will be changed without the other engi-
neers’ knowledge, invalidating the data mapping. Shortcuts
Structural and thermal engineers currently work indepen- are sometimes taken in which a few temperatures are
dently of each other using unrelated tools, models, and mapped to the structural model, and then the remaining
methods. Without the ability to rapidly exchange design temperatures computed by performing a steady state solu-
data and predicted performance, the achievement of the tion using the structural mesh. These approaches are labo-
ideals of concurrent engineering is not possible. rious, time consuming, and prone to error, which may lead
to artificial structural deflections.
Thermal codes have been unable to exploit the geometric
information in structural models and the CAD design data- Attempts have been made to use the radiation model
base, and do not facilitate transfer of temperature data to geometry ( TRASYS [1], TSS [2] ) as a basis for automating
other discipline’s analysis models. This paper discusses the mapping of temperatures. A fundamental problem with
the key features in Thermal Desktop for supporting inte- such attempts is that the model is only concerned with the
grated thermal/structural analysis. Approaches to thermal surfaces of the model that participate in radiation
modeling in an integrated analysis environment are dis- exchange. Temperature gradients within 3D solid regions
cussed along with Thermal Desktop's data mapping algo- are not represented, and are not available for interpolation.
rithm for exporting temperature data on to structural model For example, predicting thermal distortion for optical sys-
grid points. tems requires analysis of many solid features such as the
complex webbing and mounting structures on the back of
INTRODUCTION mirrors, and the structures that support the optical compo-
nents. The restriction of current radiation tools to use only
Tighter coupling between thermal and structural analysis conic surfaces with regular boundaries also creates signifi-
has long been sought, but has been prevented by incom- cant differences in the structural and thermal representa-
patibilities in existing tools. Each discipline makes abstrac- tions, making interpolation difficult or, in some areas,
tions in order to reduce the problem into one that is efficient impossible.
for calculations on a computer, and that most clearly repre-
sents the physics being modeled. These different represen- Using the geometry of the radiation model, temperatures
tations often make it difficult to transfer data from one may only be interpolated within individual surfaces at best.
model to another. The radiation model does not contain rules for interpolating
temperatures throughout the entire domain as does a finite
For example, a thermal model might consist of a geometric element model (using the element’s shape functions). The
surface model for radiation calculations, and an arbitrary radiation model does not contain information on how tem-
set of nodes and conductors for other portions of the peratures vary from surface to surface. Temperatures must
model. A significant part of the model may not be repre- therefore be extrapolated to the edge of each radiation sur-
sented by graphical entities, requiring data from the ther- face, ignoring any connectivity between them.
mal model to be mapped to the structural model by hand,
or by generating a custom program to map thermal nodal Another common but fundamentally flawed approach to
temperatures onto structural FEM nodes. integrating structural and thermal models is to use struc-
tural FEM codes for building thermal models. Codes based
Changes in the design or in either of the analysis models on these methods sometimes (but not always) recognize
usually renders the custom program invalid. Often one of the fact the matrix of terms produced by FEM is fully com-
patible with SINDA/FLUINT, and that ad hoc generation of A unique resolution to the “model mapping problem” was
conduction and capacitance terms using element centroids developed and added to Thermal Desktop: interpolation
can be avoided. Such finite difference centroid methods information is taken directly from the thermal model,
may introduce additional errors when returning tempera- thereby eliminating the requirement for the thermal and
tures to the structural program. For example, averaging structural models to be the same. This means that thermal
element temperatures to produce nodal temperatures artifi- models and structural models can be developed indepen-
cially smooths temperature variations, reducing the pre- dently using modeling tools and methods honed for each
dicted structural displacements. specialty, while preserving the ability to pass accurate tem-
perature data from the thermal analysis tool to the struc-
Yet even when centroid conversions and their associated tural tool. This approach is consistent with the philosophy
problems are avoided, the resulting tools fail to gain wide- of the development of Thermal Desktop: the key to inte-
spread acceptance. Such simple approaches have been grated design that is acceptable to end users is not to force
reinvented many times and have existed for years in vari- the use of a single compromised tool, but rather to allow
ous forms, and yet have failed to address the thermal/struc- each specialty to work with their existing fully-featured tools
tural integration problem for a variety of reasons [3]. and concepts while providing seamless pathways for
exchanging data and for maintaining common design con-
One reason for the failure of the FEM-translation approach figurations.
is that if FE methods are used exclusively, they usually
result in intractable thermal models due to excessive run Postprocessed data (temperatures, heat loads, etc.) calcu-
times for radiation calculations. Even simplification of the lated for a thermal model may be mapped to any set of
structural model into a suitable thermal model typically arbitrary point locations. The thermal model may consist of
results in prohibitive run times, the main reason being the TRASYS/TSS-like finite difference surfaces (cones,
fact that curved surfaces must be modeled using flat finite spheres, paraboloids, etc.), finite element primitives (trian-
elements. gles, quads, tetrahedrons, wedges, and bricks), and arbi-
trary nodes and conductors. For each point for which data
A spherical surface that can be modeled with a simple is to be mapped (e.g., a NASTRAN grid point), Thermal
TRASYS surface subdivided into a few nodes must be Desktop checks each thermal object for proximity to the
modeled with many flat elements to maintain geometric point. If the point lies on or in the object, data for the object
fidelity. Often the thermal representation is grossly simpli- is interpolated at the point location. The thickness assigned
fied; cylinders are converted into four sided tubes and to surfaces for conductance/capacitance calculations is
spheres are simplified into boxes. Such simplifications may used to define the 3D space for which the surface occupies
in some cases produce reasonable thermal results, but and includes the effect of surfaces with different nodes on
such vast differences in the geometric representation each side.
between the structural and thermal models make tempera-
ture mapping back onto the structural model extremely diffi- Finite difference surfaces interpolate linearly in their para-
cult. Brute-force mappings such as “nearest node” are metric space (i.e. [angle, height] for a cylinder). Finite ele-
often used, which produce artificial distortions. ment primitives use their corresponding shape functions.
True bilinear interpolation is done on quad elements; trilin-
Thus, current approaches have forced thermal engineers ear interpolation is done on brick elements. Maximum
into two unpleasant choices. One, use a purely finite ele- accuracy is obtained in mapping thermal data since the
ment based approach in order to aid in translating data same functions used to define the temperature field in the
from the thermal model to the structural model and suffer thermal solution are used for interpolating data points.
from excessive turn around times for analysis, or use
present methods and suffer from excessive turn around To provide a generalized capability, locations for which
times in the effort to translate data to the structural model. thermal data is to be mapped may be input as an ASCII
Both approaches slow the iterative process characteristic text file specifying a name (for example, a structural FEM
of integrated design and optimization in a multi-discipline node ID) and an [x,y,z] location. Output is generated as an
environment, destroying the concurrency that is trying to be ASCII file consisting of the name and the mapped thermal
obtained. data at the location corresponding to the input name.

THERMAL DESKTOP APPROACH In addition, more automated integration with common

structural analysis packages is provided. For example,
Thermal Desktop™ is a CAD-based geometric front-end to locations to be mapped may also be specified using a
SINDA. It is a key part of the approach because its design NASTRAN input deck. Grid points are read directly from
overcomes many bottlenecks of thermal-structural design the input deck and temperature data is output in NASTRAN
integration [4]. One such bottleneck is the accurate map- load set format.
ping of temperatures produced by SINDA back to similar,
but not identical, structural models.
DEMONSTRATION MODEL telescope model, while Figure 2 shows a cut-away side
view. There are 6 optical elements in the light path in the
A simplified optical telescope model is presented to dem-
onstrate the end-to-end procedure for a combined thermal/
structural analysis. The level of detail in the model is typical
of a conceptual design model.

The model demonstrates the following key points:

• Efficient use of combined finite difference and finite ele-

ment methods for the thermal model.

• Independent abstractions of the telescope for thermal

and structural models.

• Interpolation of temperatures from thermal model to

structural model.

STRUCTURAL MODEL DESCRIPTION

The telescope is representative of current remote sensing

satellites with a 1 meter resolution capability. It is a 3 mirror,
Casagrain telescope with a 28” diameter primary mirror.

The metering structure is an invar, stiffened shell which is

36” long and 30” in diameter. A 3-legged spider supports
the secondary mirror assembly. The aft metering structure
is a truss-like structure of invar, protected by a thermal
enclosure. Overall, the structural model contains approxi-
mately 1000 grids and 1500 elements. Figure 1 depicts the Figure 2: Side View Showing PM and Aft
Metering Structure
following order:

1. PM = Primary Mirror (28” OD light-weight)

2. SM = Secondary Mirror (6.5” OD solid)
3. FM1 = Fold Mirror #1 (solid)
4. TM = Tertiary Mirror (solid)
5. FM2 = Fold Mirror #2 (solid)
6. FP = Focal Plane

The PM is modeled as a 3D shell structure with front plate,

back plate, and internal core structure. The mount pads,
adhesive and bipod flexures are included in the model. The
SM is modeled as a 2D shell with edge flexures. The other
optics are represented as lumped masses supported on an
aft metering structure.

THERMAL MODEL AND ANALYSIS

The thermal model and orbit used for the analysis are
shown in Figure 3. A thermal enclosure has been added
around the metering structure and solar arrays have
located on the exterior. The solar arrays are driven by artic-
ulators that automatically track the sun during the orbit.
Temperatures presented here are for the illustrated point in
the orbit (just before the subsolar point). This orientation
produces uneven heating on the telescope body and drives
Figure 1: Demonstration Telescope Model gradients in the mirrors.
 top, a few mouse clicks is all that is required to generate
the cylinder. In a similar fashion, rings used for the meter-
ing support structure and for the ends of the bus section
were modeled using disk surfaces, again with a nodaliza-
tion appropriate for thermal accuracy requirements. Simple
finite difference rectangles were sufficient for accurately
modeling the effect of the solar arrays.

A significant difference between the thermal and structural

models exist for the spider and primary mirror metering
structure. In the structural model, beam elements and
mass nodes are all that are necessary to simulate the
desired mechanical behavior for stress and dynamics cal-
culations. This structural representation, however, is miss-
ing necessary information to carry out the thermal analysis.
For the thermal model, 2D planar elements were used so
that radiation exchange could be properly modeled and so
that gradients could be calculated to map onto the beam
Figure 3: Thermal model and orbit used for optical elements. Like the structural model, the thermal represen-
telescope demonstration model tation also contains simplifications of the real part that
make it efficient for thermal analysis.
Radiation exchange factors, orbital heating rates, and the
conduction and capacitance matrix were computed using This portion of the spacecraft illustrates the need for each
Thermal Desktop/RadCAD. The radiative analysis was car- discipline to be free to generate computer models indepen-
ried out using a full Monte Carlo raytracing process includ- dent of one another, so that only the salient details are cap-
ing the use of specular surface properties. Nearly all of the tured in each representation. Not only does this maximize
thermal model was generated automatically, even so, in a the efficiency of the model for each discipline, but also the
few places it was desired to create arbitrary nodes and data that is generated matches more closely with the goal
conductors, for example to provide a simple model of the of the analysis. Even if we had infinite computer resources
heat loss through wire bundles to the solar array drives. All such that a micro-detailed model could be generated that
modeling methods are supported simultaneously in Ther- simultaneously satisfied all engineering disciplines, simplifi-
mal Desktop: arbitrary network connections, finite differ- cations would still be desired in order to reduce the burden
ence geometry, and finite element geometry. of extracting useful information from the enormous amount
of generated data. Simplifications are made not only for
Temperatures were solved using SINDA/FLUINT, and tractability, but also to most clearly represent the physics of
results mapped onto the NASTRAN structural model. Ther- the problem at hand.
mal Desktop’s Case Set feature [5] was used to automati-
cally launch radiation calculations, conduction and The thermal model of the primary mirror is shown in Figure
capacitance generation, SINDA/FLUINT calculations, and 4. The top figure shows the mirror mounted on the metering
post processing from within Thermal Desktop. The full up structure, and the bottom figure shows the detail used for
thermal model consisted of 746 thermal nodes. the mirror core. Unlike the spacecraft body and metering
structure, no geometric simplifications were appropriate for
Thermal Desktop communicates with both CAD and FEM the thermal model of the primary mirror. Both the structural
systems. In this demonstration case, thermal and structural and thermal model use a detailed finite element mesh. In
models were generated directly by the analysts, although analyzing optical performance, accuracy of the gradients
CAD drawings could have been used as the basis for inter- present in the mirrors are paramount. Influences of the core
active model creation. The thermal model was created by structure and the mounting points simply cannot be accu-
first importing the NASTRAN (FEMAP and I-DEAS also rately predicted by replacing this mirror with simple TRA-
supported) structural model into Thermal Desktop and then SYS or TSS surfaces. Supporting detailed thermal/
making changes appropriate for computational efficiency structural/optical analysis was the primary reason for add-
and thermal accuracy requirements. Extensive use of ing direct support of FEM to Thermal Desktop.
SINDA/FLUINT submodels were employed to facilitate the
organization of the thermal model. The thermal model illustrates three important modeling
aspects:
It was felt that the number of elements used to model the
tubular body of the telescope was excessive for thermal 1. Complexity was reduced without affecting desired
purposes. Rather than analyze the tube as consisting of a accuracy by using thermally efficient, mathematically
collection of many flat elements, they were replaced by a precise curved surfaces for the spacecraft body.
single Thermal Desktop cylinder with a coarser nodaliza-
tion. Using the interactive “snap” method of Thermal Desk- 2. Independent thermal and structural abstractions were
 Figure 5: Temperatures displayed on
thermal model of demonstration telescope

Figure 4: Thermal representation of primary

mirror and metering structure
used for the mirror metering structures.

3. Detailed finite element thermal models were used

where accuracy would have been unacceptable with
further simplification.

The effect of the solar loading in producing gradients in the

spacecraft body is shown in Figure 5. Local warming of the
spacecraft is also present in the regions close to the solar
arrays. Gradients can be seen in the spider locating the
secondary mirror due to conductive contact with the space-
craft shell. Temperature distributions in the face and body
of the primary mirror are shown in Figure 6. The effect of
the mounting points can be seen on the back face of the
mirror, most notably the cold spot near the lower left edge.
A slight increase in temperature can be seen due to the
other two mounting locations. Figure 6: Temperatures for primary mirror
DATA MAPPING not contained by a thermal object, the closest object must
be found. This test is specific to each type of thermal object
In an integrated engineering environment required to (cone, paraboloid, quad, brick, etc...). If the point is not
design complex thermal/structural systems, improvements directly contained by an object, all objects must be tested
to thermal tools have little consequence if the results can- to determine the closest.
not be accurately transmitted to the structural engineers.
Previous approaches to automating the temperature map- The effort in correctly developing the algorithms to test for
ping to structural models have been ineffective, most based proximity for the variety of thermal modeling primitives is
on taking a collection of thermal points with associated data the main reason simpler approaches have been taken.
and attempting to map this to a new set of structural points. However, previous work performed in the development of
RadCAD was utilized for the data mapping algorithm. In
Temperature mapping has suffered for two reasons: addition, the speed of the model mapping is increased by
orders of magnitude due to the unique incorporation of an
1. The thermal model is not built with modeling primitives oct-cell partitioning algorithm to weed out surfaces that are
conducive to interpolating results everywhere in the distant from the structural point to be mapped. The time
domain of the model. required to compute temperature data for 980 NASTRAN
grid points was approximately 30 seconds on a 200 Mhz
Using TRASYS/TSS modeling primitives, arbitrary, non- Pentium Pro processor for this demonstration model.
conic shapes must be modeled using one-node polygons,
which cannot be interpolated. Low nodal density surfaces Another significant improvement has been made to the
suffer from inaccuracy due to extrapolation to the edges. data mapping algorithms. All thermal objects use the maxi-
Indeed, had the core of the mirror been constructed with mum amount of information to determine the temperature
typical one-node polygons for the thermal model instead of at a given point. Bilinear interpolation is used for finite dif-
quad finite elements, no information of how the tempera- ference surfaces and all planar finite elements. Trilinear
ture varies in individual core segments would be present in interpolation is used for all solid finite elements.
the model, only the bulk temperature of the webs.
Only the orthogonally gridded finite difference surfaces, the
2. Algorithms for mapping temperatures have been too triangle and the tetrahedral solid, admit a direct solution to
simplistic and introduce additional error. mapping a given x,y,z point into local parametric coordi-
nates that can be used with the object’s shape functions to
A common approach is to map a handful of temperatures interpolate temperature. Iterative algorithms were devel-
by hand and then perform a steady state solution using the oped to map an x,y,z point into parametric u,v space for
structural mesh. This often leads to strange temperature quadrilateral elements and into parametric u,v,w space for
distributions consisting of local hot spots at the temperature solid wedges and bricks. The alternative approach is to
defined locations. Another simple approach is given an decompose quadrilaterals into triangles, and solids into tet-
x,y,z location, find the closest thermal node and use that as rahedrals. Although much better than a closest node
the temperature of the structural node. An attempt to approach, the original isoparametric contours are not pre-
improve this scheme is to use some sort of inverse dis- served and artificial discontinuities can be introduced. The
tance weighting to average the results of a collection of algorithms used by Thermal Desktop to map temperatures
thermal nodes near the given point. This approach can pro- to structural points use the same shape functions used to
duce incorrect results when the thermal model consists of compute the temperatures, resulting in the maximum fidel-
closely spaced surfaces (like a honeycomb panel or mirror ity in mapping data onto the structural model.
core). A structural point may lie exactly on a thermal sur-
face which should be used to compute the temperature, The data mapping command may be invoked by selecting
however, the closest thermal node may not lie on this sur- a toolbar icon or selecting from a pull down menu. The
face containing the point, but actually be on a different sur- FEM input file is specified as well as a file that will be gen-
face located near the point. erated with temperatures in the FEM code’s format. An
optional tolerance may be specified if the structural points
The correct approach is to take the collection of x,y,z loca- do not exactly lie in or on thermal modeling primitives. The
tions for each structural node and interpolate using the Thermal Desktop model mapping input form and an exam-
geometric information contained in the thermal model. This ple of temperatures mapped to the NASTRAN model of the
becomes more difficult than dealing with a collection of telescope are shown in Figure 7. The effect of the stiffening
thermal points, since proximity tests must be developed for rings near the base of the telescope in equalizing the tem-
each type of thermal modeling primitive. The algorithms are peratures by conduction is apparent as is the influence of
very similar to those used by RadCAD for ray/surface inter- the warm solar arrays.
section when calculating radiative quantities.

The first step is to find which thermal object (arbitrary node,

finite difference surface, 2D finite element, or 3D finite ele- CONCLUSION
ment) the structural point lies on, near, or in. If the point is
 CONTACT

C&R Technologies, Inc.

9 Red Fox Lane
Littleton CO 80127-5710
303 971 0292 Voice 303 742 1540 (FAX)
[email protected] www.crtech.com

User’s manuals, tutorials, and training notes for software

discussed are freely available in PDF format at
www.crtech.com

REFERENCES

1. Lockheed Engineering and Management Services Com-

pany, "Thermal Radiation Analysis System Users Manual,"
JSC 22964, Contract NAS 917900, October 1991.

2. Panczak, T., Welch, M., "Thermal Synthesizer System:

An Integrated Approach to Spacecraft Thermal Analysis,"
SAE paper 911582, 22nd ICES Conference, July 1992.

3. Panczak, T., "The Failure of Finite Element Codes for

Spacecraft Thermal Analysis," SAE paper 961450, 26th
ICES Conference, July 1996.

4. Panczak, T. and Ring, S., "A CAD Based Tool for FDM
and FEM Radiation and Conduction Modeling", SAE paper
Figure 7: Temperatures mapped to NASTRAN 981577, 28th ICES Conference, July 1998.
structural model
5. Welch, M. and Panczak, T., "Automating Thermal Analy-
Many problems were identified and solved to improve inte- sis with Thermal Desktop", SAE paper 1999-01-2156, 29th
grated thermal/structural analysis. Thermal Desktop is the ICES Conference, July 1999.
first system to efficiently combine CAD, FE, FD, and arbi-
trary thermal network methods to allow the engineer to
build both efficient and accurate thermal models. All types
of modeling methods may be used simultaneously so that
appropriate choices may be made regarding accuracy,
model mapping needs, and efficiency.

By integrating FE methods into the traditional FD/network

environment, new capabilities are provided without sacrific-
ing present modeling approaches. A consistent control vol-
ume view of FE coupled with nodal based radiation and
orbital heating rates allows FE and FD methods to coexist
in the same model and be used with familiar solvers such
as SINDA/FLUINT. These improvements combined with
new algorithms employed in the data mapping process,
greatly improves the integrated thermal/structural analysis
process for complex systems.

ACKNOWLEDGMENTS

Development of Thermal Desktop was made possible by

the support of NASA Marshall Space Flight Center, with Mr.
Bill Till serving as contract monitor.

The structural telescope model was developed by Dr. Victor

Genberg of Sigmadyne, Buffalo New York.