Learning LISA on Linux

[vturbine]

After reading some of WebPilot's tutorials, I decided I wanted to try to learn FEA methods using LISA, a low cost full (and free trial version) program that runs natively on Windows, and so far from what I can see, on Linux in WINE.

This thread will, I hope, document my progress in trying to first learn basic FEA methods using LISA, and eventually modeling and comparing results with WebPilot's Dynajet valve analysis. I hope that the learning process may help others, as well.

Here are the specs for what I'm using in software:

Puppy Linux 4.1.2
Wine Version 1.1.7
LISA Version 6.1.2 (full version)

LISA 6.1.2 running in Wine

[vturbine]

I downloaded the LISA software last night from:
http://www.lisa-fet.com/download.htm

I first tried the 6.2.3 version, but found that WINE 1.1.7 did not work well with it because of version 6.2.3's dependence on Microsoft's DirectX. I even downloaded the Winetricks script and installed d3dx9 which should have given WINE DirectX capabilities, but repeatedly got an error message saying oleaut32.dll was not a recent enough version. Playing with winecfg and specifying a native dll for this one still didn't work, despite the fact that LISA brings a version of this dll into its own directory.

Giving up on DirectX, I downloaded the 6.1.2 version of LISA which uses ActiveX, and this was installed without problems and opened right up in Wine 1.1.7.

I clicked on Help-Lisa's User Manual and found that it couldn't be located. Apparently the Manual directory was missing. I then noticed that in the dropdown Help menu, there was an entry for Unzip Manual. Clicking on this did indeed install a manual directory and several other files in the LISA directory.

I clicked on Help-Lisa's User Manual again and found that the manual now couldn't be displayed without downloading Gecko. I did this, but still had problems displaying the manual.

Since during manual unzipping I noticed that the manual files were all html files, I navigated to the manual directory, and, on a hunch, did a search for "index.html" I found it and opened it in my regular browser and YAY, it opened and it was a table of contents with links which worked to all the other manual pages. No need for running Gecko in WINE, why not just use my native Linux browser for the manual?

I set a bookmark in my browser for that page, and now it's easy to open the LISA manual. Here's what it looks like:

[vturbine]

I navigated to the Mechanical Analysis -- Quickstart lesson, and followed the steps there to analyze a simple cantilevered beam. I won't give the steps here, since they are covered so well in the manual.

Here is a picture of the mesh I created showing the nodes, elements, constraints, and point load. Everything worked according to the manual except one VERY important thing! I was unable to save the drawing. At this point I was using the free trial version of the software and should have been able to save up to 1300 nodes. But found I couldn't.

[vturbine]

Doing a little online research, I found out the reason I couldn't save was that I hadn't filled in values in the LISA License Editor (under Help).

These values are to be found in Help-->Help-->Activating the LISA License. NOT in the manual...A little tricky to find and non-intuitive for those who RTFM, like myself

Anyway, after properly entering the license values, and repeating my earlier work (which did help reinforce the process mentally), I was able to save and then perform the analysis of the beam.

EDIT:

Because I'd tried to add DirectX via Winetricks earlier (not guaranteed to work properly with WINE), and was using an older version of WINE, I started to have some emulator problems. Deciding it was time to do a fresh install of WINE, I upgraded to WINE Version 1.1.21 and reinstalled LISA. Everything has been running smoothly since.

I also found that the 1300 node trial edition licenses had been automatically entered this time around, so the above license entry problem was probably not typical.

Here's a plot of the beam I was working on:

[WebPilot]

Looks good ... and complicated. PM me if I can be of some help.

[vturbine]

Thank you WebPilot. I will if I get to a point where the problem is conceptual understanding, rather than software installation and operation.

Tonight I've input a manual example for:

Beam, is of length 12 in, and cross-section 2 in x 0.5 in. It is rigidly fixed at it's left end. The material properties are Young's modulus (E) 3e7 psi, Poisson ratio 0.3, density 7.22e-4 lb/in**3. The lowest natural frequency of this beam is required to be determined.

Unfortunately it didn't work out, I followed the steps twice, so the problem is likely to be either the manual steps are incomplete or incorrect, or the program is not working properly in WINE (likely). I suppose it could also be a program bug, or a mistake on my part, repeated twice.

Tomorrow I'll boot up Windows, and try it again on that OS to see if it solves properly. If not I'll contact the program author.

Note: this program has not been advertised as working with WINE, but since it opens in WINE, and successfully solved a cantilevered beam problem so far, I still have hopes for it. I guess we'll find out tomorrow.

[tufty]

Have you tried Code Aster? Not sure if it does all you want, but it's linux-native so you don't have to tit about with WINE and all that crap. Oh, and it's free :)

[vturbine]

Nope, never heard of Code Aster. I got an error when attempting to download documentation, and the rest of the docs were in French. It looks like overkill for what I'm interested in. But you're welcome to open a parallel thread for that one if you are familiar with it.

Back to Learning LISA on Linux.

The same problem showed up in the Windows XP installation as the Linux/WINE install. So I wrote to the program author.

He looked at my data file and said it was missing constraint information. But I knew I had in fact added it.

The problem turned out to be a mistake in the manual. LISA, as I learned, has a modeling interface, but the solver portion of the program operates on a text data file (.DAT) created by that interface, and manually saved or updated by the user.

There was an instruction to "save" data in the user manual example I was trying, but it came at too early a stage in the hand data entry. Other data (such as constraints, and global preferences) were entered later, but there was no "save" instruction in the example after they were added. So the solver had tried to work on an incomplete .DAT file.

I repeated the example, this time saving the .DAT file before starting the Solver, and everything went smoothly in XP.

Mesh

Partial screen shot of output file

Graph

[vturbine]

Unfortunately, it didn't go quite as well in Linux/WINE. Almost, but not quite. However I was able to work out what the problem was, and come up with a workaround.

The problem:
In Wine, LISA produces a slightly different datafile. Specifically, it adds some leading spaces in front of a few lines of data towards the end of the file. It's easy to delete these spaces, and then the LISA SOLVER engine will accept the data and solve the analysis. If you don't correct the spaces, the Solver will still produce output, but the eigenvalues will be missing the first digit. They will be replaced by a DC1 character (hex 11) Everything else will work fine.

I notified the author of the program of this discovery, and hope that this will be changed in the future. LISA is not promoted to work with WINE, however, it seems that with some very small changes, it will indeed be compatible out of the box. I'm not discouraged by these initial few minor problems. This is a learning exercise for not only FEA analysis, but also for making things work, including software. I enjoy hacking (in the old sense) and making things mesh together on a shoestring -- I've done that with computers and software ever since DOS days.

[vturbine]

So, having worked around most cross plarform problems I decided to tackle the 2" by 1" by 0.01" cantilevered plate problem that WebPilot posted in the Mode 1/0 thread. Typing the coordinates of the nodes into LISA and following the LISA manual revealed my first mental blunder -- I tried solving a 2D quadrilateral with thickness, for a 3D problem.

In fact, using LISA, I needed to create 3D hexahedrons to properly model torsion in the cantilevered plate. This process took awhile to learn, but eventually I was able to construct 3D elements, scale them, subdivide them, sweep away duplicate nodes, etc. LISA's user manual was good in this regard, with worked out examples and steps.

But my problems had really only just begun. Just because you can create a mesh model doesn't mean you know how to assign it physical values.

Plot of the cantilever torsional mode of WebPilot's example problem in LISA on Linux

[vturbine]

This is what really tripped me up.

LISA Properties...NOT.png (6.88 KiB) Viewed 2384 times

I used what I thought were the correct figures for steel:

Young's Modulus of 30000000 lbs/sq. in.
Density of .283 lbs/cu. in
Poisson Ratio of 0.3

I checked these in lookups on the web and in WebPilot's examples.

However the results of running the model were all wrong. I went through a lot of unnecessary hair pulling over this. I ran the same model, using SI units for everything, and got good conformance with WebPilot's results. I tried doing an alternate set of calculations by hand, also using SI units. But every time I tried a model in English units, the numbers were all wrong. I was sure I'd used inches throughout -- no mixed units. What was wrong with LISA?

WebPilot pointed me in the right direction -- the lb figure needed by the program for density is lbm, not lbf, and the way to achieve that is to divide the standard english unit density of .283 by 32.2 ft/sec^2, the acceleration due to gravity, and then dividing by 12, because we are using inches throughout this model, not feet.

The final density of steel for entry into LISA would then be .000733. When I changed the Global properties to that figure, I was finally able to get a good run.

After a half hour run, it was great to see numbers appear that looked like WebPilot's results, and to be able to see a graphic plot of the plate's shape in Mode 1/0.

[vturbine]

Here are the model and results:

[vturbine]

Comparing LISA's results with those posted by WebPilot

In summary, for an a/b ratio of 2:1, Barton's analytical work gives values of

* mode 1 (or mode 0/0), 846.1 Hz
* mode 2 (or mode 1/0), 3638 Hz
* mode 3 (or mode 0/1), 5266 Hz

Calculating the frequencies from the LISA program outputs gives:

* mode 1 (or mode 0/0), 892.1 Hz
* mode 2 (or mode 1/0), 3526 Hz
* mode 3 (or mode 0/1), 5512 Hz

Variation from the Barton's calculated values is:

* mode 1 (or mode 0/0), +5.4%
* mode 2 (or mode 1/0), -3.1%
* mode 3 (or mode 0/1), +4.7%

We just squeaked past the 5% cutoff in mode 1. Perhaps a better model would get it closer.

One problem with LISA is that it doesn't support much in shell element operations, so I had to use 8 node bricks. To optimize the mesh for the solver portion of the program, the model was made from a mesh of cubes. This meant a .05" spacing between all nodes. Yet it only yielded two layers in the Z direction. More layers there would make for a more flexible model with lower frequency results. But it would also create geometrically more elements and processing time to compute.

I might try experimenting with more Z layers, and non cubic elements to see if results are any better.

But so far, it's very satisfying to master something simple like this in an engineering modeling system I've never used before.

[vturbine]

Elements come in a variety of types, and it is a little bewildering whan starting out to try to understand what they are, what they mean, and when to use each type. The LISA manual gives a good start with the "quickstart" examples. But you are basically following steps by rote here, without a lot of explanation or definition.

I found a fine set of explanations for elements here, and it really helped clarify the terminology and meshing strategy for me:

http://www.biomesh.org/elements.phtml

[vturbine]

Well, the learning was great, but the conclusion is no go for LISA.

I successfully learned how to model a finite element in vibration, using LISA, and managed to get reasonable results for WebPilot's example. Extremely glad I did.

But I finally learned enough to realize that the LISA program won't be able to handle pulsejet valve modeling, at least in its present level of sophistication.

Because I can't use shells to model a valve, I have to use bricks. But bricks work best when they are cubic -- with all sides equal. They become less accurate as they deviate from that.

Also, in a sheet material, the mesh should be divided, when using bricks, into at least one half the thickness, and preferably one third or even less. In other words, with say a valve of .006" thick material, the mesh should be .003" or preferably finer to adequately model the valve's flexibility. That means one side of every brick must be .003" or smaller in height. If we want to keep the bricks cubic, it means all three dimensions are .003"

Theoretically this would be possible. But practically, it isn't

For a half inch wide rectangular strip valve 1" long made from .006 material it would take (0.5 x1 x .006)/ (.003^3) brick elements to make up an accurate model. That's about 100,000 elements.

Even if it didn't crash or freeze up trying to display that many, my computer would probably take over a week of continuous computation to solve that problem, and LISA display and menu/mouse program function would crawl to a standstill.

LISA worked better on WebPilot's example because it was thick -- .1" is huge for a strip valve, but it was only intended as a test lesson. With 1600 bricks and a half hour LISA solved as many as 5 modes (loadcases) for that problem. And for thicker objects much simpler models would produce accurate results quickly.

But pulsejets are difficult to model, because they use thin materials -- just as they are difficult to weld.

So, learn and live.

Glad for the learning. I'll be looking for another FEA program, free or low cost, which can handle thin flat shells in dynamic vibrational modeling, and be run on Linux. I might have to learn engineering French following tufty's Code Aster suggestion, after all.

I've sent for a LiveCD of CAELinux 2008 (the last 32 bit proc version) to explore that possibility.

Où est la bibliothèque?