You are on page 1of 9

Indiana Institute of Technology

2014 Summer Research Project


Structural Analysis using Multiple CPUs

Fort Wayne
August 11th, 2014
Leonardo Freire Marques
Leslie A. Grundman

Indiana Tech 2014 Summer Research Project


Structural Analysis using Multiple CPUs

Fort Wayne
August 11th, 2014

1
Table of Contents
Introduction .................................................................................................................................................. 3
Finite Element Analysis (FEA) .................................................................................................................... 3
NX Nastran ................................................................................................................................................ 3
Shared Memory Processing ...................................................................................................................... 4
Data and Results ........................................................................................................................................... 5
Conclusion ..................................................................................................................................................... 7
Future work................................................................................................................................................... 7
References .................................................................................................................................................... 8

2
Introduction
Finite Element Analysis (FEA)
To perform the finite element analysis is used the finite element method.(FEM) It
is a mathematical technique used to obtain approximate solutions to complex problems
that cannot be solved using basic theories[1].
FEM works by breaking down a real object into a large number of finite elements,
therefore it splits up a complex problem into a greater number of simpler problem
(elements are simple geometric shapes like triangles or quadrilaterals for 2D problems
and tetrahedrons and hexahedrons for 3D problems). All of the elements make up a mesh
which is an approximation of the problem. This method use mathematical equations to
predict the behavior of each element. A computer then adds up all the individual behaviors
to predict the behavior of the actual object.
Finite element analysis helps predict the behavior of products affected by many
physical effects, including[2]:
Mechanical stress
Mechanical vibration
Fatigue
Motion
Heat transfer
Fluid flow
The FEA share common procedure steps no matter how simple or how complex
the problem[1]:
Step 1: Evaluate the problem and make assumptions
Step 2: Describe how the finite elements will behave
Step 3 Build the finite element model
Step 4: Form element equations
Step 5: Assemble each element equation into s global problem equation
Step 6: Specify loads and boundary conditions
Step 7: Solve the global problem
Step 8: Evaluate the results.

NX Nastran
NX is a CAD software that allows you to work in objects, given them their geometry.
Nastran is a finite element solver from Siemens that helps to predict how the
structures will behave in a set environment.
The Nastran can be used in parallel processing, in this project was used the shared
memory processing.

3
Shared Memory Processing
Nowadays the volume and complexity of the analysis is going up, and the time
allocated to perform these tasks is under increasing pressure. In order to perform these
analysis is used the parallel computing. It is an approach that uses multiple computers or
processors working together on a common task. Each processor works on a part of the
problem and exchanges information with other processors.
The benefits of using parallel processing are:
Reduce solution time on large problems.
Decreased turnaround time for each solution; allowing to perform varies
analyses more designs under more conditions in a shorter amount of time.
In this research was used the Indiana Tech supercomputer (Godzilla) 12-core
workstation. The Figure 1 illustrate the Godzilla architecture.

Figure 1. SMP Architecture.

A SMP machine is defined as a single machine with multiple processors that share
a common memory and input/output. Multiple processors can operate independently, but
changes in a memory location affected by one processor are visible to all other
processors. The primary disadvantage is the lack of scalability between memory and
processors[3].
To request a SMP job in NX Nastran is necessary to use the command parallel=p
(where p is the number of CPU used by NX Nastran to solve the modeling).

4
Methods and Results
First was designed a model in the NX (Figure 2)

Figure 2. Model

After the model was finished, then was created a FEM file and applied tetrahedral
3D mesh. And the boundaries conditions were applied (Figure 3).

Figure 3. Model meshed with applied boundary conditions.

5
And then the model was solved, the Figure 4 shows the nodal displacement.

Figure 4. Nodal displacement

The Table 1 shows the data collected. Varying the number of processor working
in the solution and the time took to finish the task.

Number of
Time
cores
1 0:57:23
2 0:33:22
4 0:22:18
8 0:16:57
12 0:16:23
Table 1. Data

The Figure 5 shows how the speed of the solution behave while change the
number of CPU.

6
4

3.5

3
Solution speed up 2.5

1.5

0.5

0
1 2 4 8 12
Number of CPU

Figure 5. Solution speed up versus number of CPU.

In the figure above is observed that the increase of number of CPU do not is linear.
The graph behave as a logarithmic function. Doubling the number of CPU does not mean
the solution will take half of the time, as shown in the figure.
The time took when was used a number of 8 CPU is pretty close when was used
the maximum number of CPU, 12 CPU. The difference is just 34 seconds.

Conclusion
Complex models with great number of elements are becoming the norm today. As
analysis models continue to grow in terms of size and complexity, solver performance
becomes a critical factor. Using multiple processors in parallel can significantly reduce
solution run times compared to the more traditional serial solution that uses one
processor. Improving the companys ability to produce results in time to impact key design
decisions.
Increasing the number of processor the speedup will not be algebraic the same
magnitude. It is not a linear function.

Future Work
I recommend to those who will do a similar search to compare the results in this
projects with other computer.
Build a Distributed Memory Process (Figure 6) network to compare with the SMP
technology used in this project.

7
Figure 6. DMP Architecture

Distributed memory process (DMP) can achieve much higher levels of scalability
compared to shared memory processing (SMP)

References
[1] Mac Donald, Bryan J (2011). Practical Stress Analysis with Finite Elements. 2nd ed. USA:
Glasnevin Publishing. 386.

[2] AutoDesk. (2014). Finite Element Analysis. Available:


<http://usa.autodesk.com/adsk/servlet/item?siteID=123112&id=17670721>. Last accessed 9th
Aug 2014.

[3] Siemens (2011). NX Nastran 8 Parallel Processing Guide. USA: Siemens Product Lifecycle
Management Software Inc. 73.

You might also like