RISA 3D 16.04 Update 2

[TEDstruc]

Is anyone else experiencing much longer model run times with the latest update of RISA 3D?

I just updated from 16.03 to 16.04

I had a model that ran fine using 16.03 in less than a minute. Now with 16.04 the exact same model gets stuck in a seemingly endless loop of solving for P-Delta Shears. I probably let it calculate p-delta shears for 5 minutes before I ran out of patience, cancelled the solution, turned off p-delta, and it solved in about 2 or 3 minutes. I don't see any glaring p-delta issues when I look at the deflected shape, nor are there any instabilities that RISA is detecting.

I'm wondering if I had changed something in the global settings previously that made the program run faster, and after the update, maybe it reverted back to default settings?
I tried changing the number of member sections from 10 to 5 and internal sections from 100 to 10. Didn't seem to help.

 

[JoshPlumSE]

It's usually a good idea to take a look at the program's release notes. But, unless your model has thermal loads or analysis offsets, then I don't think anything mentioned in the upgrade could have caused the behavior you're experiencing. Certainly, the issue could be a "mirage".... meaning something changed in the model to cause the difference in behavior. Or, the issue could have been caused by what was thought to be an inconsequential change that didn't make its way into the release notes.

I should point out that when I was with the old RISA Technologies, I was pretty strict about trying to include any substantive change in the release notes. When an issue like you describe would happen, I just always wanted to be able to say definitely to any user who called in that I knew where the bug must have come from. There was always a little bit of resistance from sales / marketing folks.... not wanting to make the company look bad by listing too many bugs / corrections in a minor release. After all, no one likes airing their dirty laundry in public.

Since a good portion of the senior engineering staff has turned over recently (Roger, Matt, Mike, and myself), it's possible that there isn't anyone from an engineering perspective with enough authority to feel comfortable fighting for the old procedure for release notes. If it were happening to me, I would send them an e-mail (though I wouldn't be surprised if they're not allowed to respond to e-mails from me) to ask about what's causing this behavior. Then if it's something that should have been in the release notes I'd politely point out to them why it's important to show any substantive changes in the release notes.

 

[bootlegend]

Same issue here. We sent them a note and they are looking into it but had us roll back to 16.0.3 for the time being. Must be one of those "Oops!" moments.

 

[JoshPlumSE]

Any tension only members or compression only springs? Those can be very sensitive and small changes to the solution can cause iteration issues which lead to larger solution times.

 

[dik]

Josh... unless it was unstable, you would think that it would converge fairly quickly with only a few iterations.

Dik

 

[JoshPlumSE]

Dik -

You would think, but non-linear solutions like this are really tricky. Especially for a program like RISA which is (at its core) a linear solution.

The models that really kill you are the ones where the force in a tension only member are really low. Say 0.01 lbs. You and I would call that a 0 force member. But, to RISA it's either in tension or compression and can cause multiple iterations.

A couple of years ago (released in v16.0 I believe), I spent a lot of time with our lead developer coming up with a scheme to better converge models with tension only members. Essentially, when we found a tension only member in compression instead of turning it completely off for the next iteration we would use a reduced stiffness of some small fraction (1E-6) of it's normal stiffness. This really helped convergence. I should also point out that this changes is one of the first steps towards RISA bridging the gap between a mostly linear solution and true non-linear solution.

There's only one person left over there which truly understands what we had done and why. That being said, I did document it in the help files.

Note: If they make some minor changes to this (like going from 1E-6 to 1E-8 or something) then this would certainly be considered such an esoteric change that it wouldn't have been put in the Release Notes. Of course, sometimes these little esoteric changes have bigger ramifications than you would think. Or, sometimes they're not done correctly....

 

[TEDstruc]

I DON'T TYPICALLY USE TENSION ONLY MEMBERS, I MIGHT HAVE A COUPLE EULER BUCKLING ONLY MEMBERS. I'LL CHECK FOR ANALYSIS OFFSETS TO SEE IF THAT IS POTENTIALLY CAUSING ANY ISSUES.

HOW DO I REVERT BACK TO 16.03? DO I HAVE TO CONTACT RISA?

ONE MORE THING, IT'S GETTING STUCK ON THE FIRST ITERATION.

THANKS FOR ALL OF THE REPLIES.

 

[bootlegend]

I think we made sure we had no tension only or Euler buckling members and it wouldn't run. It also wouldn't solve simple DL only case.

They sent us a link to 16.0.3 to reinstall.

 

[JoshPlumSE]

I probably can't help too much here guys as I can only speculate what's happening based on past experience. For what it's worth, these are my thoughts:

1) Before I install any upgrades to RISA, I always go the folder where the program is installed and make a copy of the current executable renaming it something like risa3dw1602.exe. I do it because I always want to be able to test one version versus another any time. But, it really helps if you realize there is a bug and need to use an older version.

2) If you want to revert to an old version, contact their support group. They should be able to send you a link to the older install routine.

3) Euler Buckling members behave almost the same as tension only members. They should be less prone to iteration or convergence problems, but it could still be related to what you're seeing.

4) Gravity only cases are FREQUENTLY the worst Load Combinations for tension only convergence problems. Those are the ones where the braces have nearly zero force in them. Then if the program turns them all off the model becomes unstable. The configuration that caused the most head aches was X bracing where the two tension only X braces were connected at the center of the X.

 

[JoshH726]

I also had similar issues and emailed RISA. Here's the response:

I apologize about the delayed run-time you are experiencing in RISA-3D. We have had our developers look into this issue and been able to provide a resolution. I'm happy to let you know that we have just released an updated version 16.0.5 that has dramatically improved solution run-times. Be sure to fully uninstall RISA-3D version 16.0.4 before running this installation package.

http://downloads.risa.com/3d/install_3d_1605.exe

You should no longer have that same experience with the P-Delta solver that you described.

______________________________________________

Tried the same model I sent them, and it's significantly faster.

 

[JoshPlumSE]

JoshH -

Thanks you for the update! For what it's worth, their release notes say the following about the correction:

• Corrected an error in the iteration of member design which caused an increase in solution time.

I'm not sure why member design (which is post processing) would affect P-Delta iterations (which are part of analysis). But, at least it got resolved quickly! All software programs have issues with bugs from time to time. There is a non-RISA analysis program I've been using that crashes whenever I search the help file for certain terms. But, being able to resolve it correctly and quickly is the real test of quality.