A flexible Python interface to FEAP

A flexible Python interface to FEAP
Computer Science Master of
Engineering Project
PyFEAP – A Python Interface
to Finite Element Analysis
Program (FEAP)
Name: Chen-Yi (Charlie) Chen
NetID: cc2275
I. Introduction:
This M. Eng. Project is about PyFEAP, which is a Python interface to the Finite Element Analysis
Program (FEAP) server. By utilizing this PyFEAP module, users can access both Python's and FEAP's
features simultaneously, and can share data between two environments. In brief, PyFEAP allows users
to perform the following tasks:
• Write a script in Python to perform relevant finite element analysis through FEAP.
• Use Python scripting ability to run parameters studies in FEAP.
• Extract mass, damping, and stiffness matrices for experiments with new linear solvers in a finite
element system.
• Evaluate reaction forces for a given displacement to experiment with new solvers.
• Plot FEAP structure using Standard FEAP's X11 graphics
• Debug FEAP features by directly examining and manipulating FEAP's internal data structures.
The PyFEAP module helps FEAP users to script more easily since Python is marvelous when it
comes to scripting. In addition, Python's stability and version backward compatibility makes it a good
platform to create an interface on.
II. Background:
This project idea stemmed from Professor David Bindel, who has worked with finite element
analysis and FEAP for years. Before this project started, Professor Bindel developed some extensions
to FEAP to help him and others perform more complicated tasks that are difficult in FEAP. Two such
extensions are MatFEAP and FEAPMEX; both of them provide an interface from MATLAB to FEAP
and extend some basic functions that FEAP can perform in FEAP environment. FEAPMEX, which is
the predecessor of MatFEAP, uses the MATLAB MEX files feature to extend MATLAB with external
functions written in C and Fortran. Therefore, FEAPMEX is written in a mix of C, Fortran, and
MATLAB, and the MATLAB functions in FEAPMEX provide a high-level scripting interface to FEAP.
FEAPMEX was a tightly integrated code that relied on low-level interfaces in both FEAP and
MATLAB. As FEAP and MATLAB evolved, this organization caused maintenance problems, and it
became difficult to maintain compatibility across multiple versions of FEAP and MATLAB. In order to
address the maintenance problem, MatFEAP decouples the system into a FEAP-based server and a
MATLAB client the communicate via a socket. In this approach, MatFEAP eliminates the dependence
of Mex files and minimizes the dependence between FEAP and MATLAB.
MatFEAP works pretty well as long as users have MATLAB installed. However, MATLAB is a
licensed software so users need to buy it. Thus, Professor Bindel had the idea to create a Python
interface that is similar to MatFEAP. Python is a great platform for such tasks. It is open-sourced and
free to install; it has great backward compatibility between versions (except the revolutionary Python
3). Python is also a very popular scripting language that supports scientific computing through different
third party libraries. All these great features make Python our choice of platform to create this project—
PyFEAP.
III. Architecture:
PyFEAP is built around two processes: one for Python, and one for FEAP. The processes are
effectively cooperatively multi-tasked. Both processes have to be active in order to make the whole
system runs. The FEAP processes starts a FEAP server and acts as a daemon to listen to the client
request. Thus, when we initialize PyFEAP process, namely calls to pyFeap.start, we create a TCP
socket connection between FEAP process and Python process. In addition, when issuing different
PyFEAP commands such as pyFeap.cmd, commands use this socket connection to transfer to FEAP
server for calculation, then PyFEAP retrieves the computation results from FEAP to display to the user
in Python console. Similarly, the variables are sent and received between processes using standard TCP
socket connection as well.
As stated above, the communication method between PyFEAP and FEAP server uses the standard
TCP socket built-in Python native library. PyFEAP uses the AF_INET and SOCK_STREAM parameters
as the socket connection and socket type. AF_INET in a Python socket represent a tuple of connecting
to (host, port), where the host is a string domain name or a IPv4 IP string, and the port is the port
number that socket will connect to. SOCK_STREAM specifies the socket type PyFEAP is using here as
a binary stream socket. As default, PyFEAP uses blocking socket which means Python blocks the
routines when socket is still waiting on sent information or has nothing more to return. Therefore, it
would be dangerous to "block" a the PyFEAP internal socket connection to FEAP server by directly
calling PyFEAP.socket.send or PyFEAP.socket.recv, as it might block the socket connection and hangs
the PyFEAP object. It is always best to use the PyFEAP defined _send() and _recv() functions, as they
have the buffer size determination that would prevent PyFEAP internal socket from blocking.
Often, when PyFEAP encounters errors, it will raise custom exceptions and kill the connection to
FEAP immediately. So the FEAP server would still be in a consistent state. However, there might be
times when PyFEAP's lower level functions encounter an error that is not handled. In those cases,
Python would still throw standard Python exception generated by PyFEAP, but the state of FEAP
server is indeterminable. Even though most of the time when this case happens, the restarted
computation script using PyFEAP seems to be okay, it would still be safe to restart the FEAP server
instance with the corrected computation script. In that way, we can guarantee the FEAP server would
be in a consistent state thus guarantees correct computations.
In general, FEAP server has its own signal handler Ctrl-C, which allows the user to stop the
daemon to close everything. Similarly, Python has its own signal handler for Ctrl-C. The user can use
Ctrl-C to break out any running script or computations to return to Python console or terminal prompt.
IV. Examples:
In this report, I provide two examples to show how to generate python scripts that utilize PyFEAP
interface to communicate to FEAP:
1. Modal analysis of a block
Our first, and simplest, example involves computing the first few modes of a 2D block of material
which is anchored on two sides. In this example, we use the PyFEAP interface to set mesh parameters
and to script graphics display.
The FEAP input deck
The FEAP input deck for this example describes a simple Cartesian mesh, n elements on a side:
feap
0 0 0 2 4 4
block
cart n n
1
2
3
4
0
1
1
0
0
0
1
1
ebou
1 0 1 0
2 0 0 1
mate
solid
elastic isotropic 10 .1
density mass 1.0
end
inte
stop
The Python script
The first Python script uses the PyFEAP interface to load the deck with n = 10 elements on a side; run
a modal analysis; and plot the first four modes.
#!/usr/bin/env python
# Example of FEAP / Python interface using PyFEAP:
# Run a modal analysis in FEAP and plot the original and displaced
# node coordinates for the first mode using FEAP graphics.
from pyFeap import pyFeap
import numpy
# create an interface object with the path to all the example files
obj = pyFeap.pyFeap("/home/charlie/matfeap-0.8/python/example")
# Build a 10-by-10 block of material
params = {'n': 10}
obj.start('Iblock', params)
# Run a modal analysis
obj.cmd('mass', 'tang', 'subs,,10')
# Plot the first four eigenvalues
obj.cmd('plot,deform,,,0.1')
for i in range(1,5):
obj.cmd('plot,wipe')
obj.cmd('plot,eigv,%d' % i)
print 'plotting eigenvalue of %d: ' % i
print 'Press any key to continue.'
raw_input()
# Wrap up
obj.quit()
Detailed description
The import in the beginning of Python script declares what Python modules to import into this script.
We import our PyFEAP module as well as numpy here.
from pyFeap import pyFeap
import numpy
We need to first create a PyFeap object in order to start everything. The constructor of PyFEAP takes at
least one argument, which is where the example FEAP blocks and Python scripts located.
obj = pyFeap.pyFeap("/home/charlie/matfeap-0.8/python/example")
The start command runs FEAP on the specified input deck. Notice that the parameter n is not defined
in the FEAP input deck; we pass it in as a parameter set. The parameters we passed in is a Python
dictionary structure.
params = {'n': 10}
obj.start('Iblock', params)
After this mesh is constructed, the input deck tells FEAP to enter interactive mode with the inte
command. When using the PyFEAP interface, interactive mode returns control to Python console.
Now we want to run an actual modal analysis. In the first script, We use FEAP's solvers for the analysis
by running mass, tang, and subs commands as we would from the command line interface. The
command to feed commands to the FEAP command line interface is cmd:
obj.cmd('mass', 'tang', 'subs,,10')
After computing the mode, we want to display the first four modes. We can accomplish this through a
for loop and pause between each plot's display.
obj.cmd('plot,deform,,,0.1')
for i in range(1,5):
obj.cmd('plot,wipe')
obj.cmd('plot,eigv,%d' % i)
print 'plotting eigenvalue of %d: ' % i
print 'Press any key to continue.'
raw_input()
When we are done with this input deck, we need to leave interactive mode and shut down FEAP. In the
usual FEAP interface, we would type quit and then answer n to the question Continue with interactive
input options for control? From the PyFEAP interface, we use the command:
obj.quit()
Plots from the script
2. Linear solves in Python
Our second example involves deformation of a block of material under a load. The effect is the same as
running tang,,1 in FEAP, but we do all the computations with Python, numPy and sciPy solvers.
The FEAP input deck for this example describes a simple Cartesian mesh, very similar to the mesh
used in the previous example. The block of material is anchored on the left edge, and loaded along the
bottom edge. This Python script uses the PyFEAP interface to load the deck with n = 10 elements on a
side; compute the tangent, residual forces, and deformation; and plot the deformed structure.
#!/usr/bin/env python
# Example of FEAP / Python interface:
# Run a simple small-deformation static displacement problem.
from pyFeap import pyFeap
import numpy
# Build a 10-by-10 block of material
obj = pyFeap.pyFeap("/home/charlie/matfeap-0.8/example/")
params = {'n': 10}
obj.start('Iblock2', params)
# Form the tangent matrix
K = obj.tang()
# Form the residual force vector
R = obj.resid()
# Compute a Newton update
du = numpy.linalg.solve(K.todense(), R)
# Set the displacement vector
obj.setu(du)
# Plot the results
obj.cmd('plot', 'defo', 'mesh', 'boun', 'load', 'end')
# Quit
print 'Press any key to exit.'
raw_input()
obj.quit()
Detailed description
As before, we begin by loading the input deck and changing the output and plotting parameters. We
then extract the tangent stiffness K and the residual R from FEAP. These commands form the tangent
and residual; there is no need to execute separate tang and form commands. Once the tangent and
residual are constructed, we solve the linear system, and then write the resulting displacement vector
back into FEAP's memory.
Once we write the displacement vector into FEAP's memory, it is available for other routines from
within FEAP. In particular, when we run the plot line, the deformed mesh reflects the updated
displacements.
As before, the last line shuts down the FEAP system.
Plot from the script
V. Future Work:
Possible future work for the PyFEAP module is to further develop functionality from FEAPMEX that
was never implemented in MatFEAP. It would be ideal to implement the utility C functions in
FEAPMEX to PyFEAP, just to complete the whole functionalities of this Python interface. Ideally, the
translation from C functions to Python module would be compatible since we used Cpython for
everything; it would make perfect sense that those compiled C functions can be imported to Python and
thus extend the functionality of current PyFEAP.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement