Gentlemen

I am modeling an pipe system for an underground gold mine. The system consists of a chilled water pipe (5 celsius) and a warm water return (35 celsius) running parallel to each other. These pipes are u-bolted every 3m to a structural beam, which is in turn "hung" from the "roof" of the mine by chains on either side.

Some sizes to gain perspective:
Pipe diameter: 500mm NB
Total pipe length: 3709m
Chain length: 800mm
Number of elements: 4900 (approx)
Rod Increment (degrees): tried all
Rod Tolerance (degrees): tried all
Caesar version: latest

The system is shown in the schematics below. The model is attached, and the config file is uploaded as "caesar.cfg.txt".

My questions are as follows:

Q1. Why will the model not converge? (+Y supports have been investigated, but too many assumptions need to be made...)

Q2. The first schematic shows the anchors as red dots with points in the model labelled 1 to 4. If I split the system into 3 parts (ie: 1-2, 2-3 and 3-4), then part 1-2 converges, but not the other two parts. Part 1-2 is more complex than the other two parts - so why won't they converge?

Q3. This model takes approximately 30 minutes to run (Intel Core2 Duo @ 2.2GHz, 4GB ram, Win7-64). Why is it taking so long?

Q4. Caesar allows 512MB of ram to be allocated to it. Whilst this model is running, I have some 2.3GB of ram free. Can more ram not be allocated to speeding up the run?

Q5. The C-Noded Lim, Gui, Y restraints disappear from the input and output display. Is this memory related (tested on both ATI Mobility Radeon HD 2400 XT and Nvidia Ge-Force 9600M GT, in all Caesar II graphic driver modes)?

Many thanks for your attention.

(zoom in to schematics if not clear)

Q1: Every model is different and there is no specific "one answer fits all" response. You can search this forum for "convergence" and find a great deal of discussion.

Q2: See the response to Q1 above.

Q3: If a particular load case doesn't converge, another iteration is performed. This continues, potentially for a very long time, until all of the non-linear boundary conditions converge. Again, every job is different.

Q4: More RAM won't speed things up. For a given job, the stiffness matrix and load vector are a given size. As long as these fit in core, you don't need to allocate more memory.

Q5: No this is strictly a graphics issue. It could be related to the amount of video RAM on your graphics board, but this has nothing to do with the amount of RAM you allocate to CAESAR II.

We'll take a look at your job to see if we have any suggestions.

I can get the job to converge. Set these values in caesar.cfg:

ROD_TOLERANCE = 0.3000000E+02 59
ROD_INC = 0.5000000E+01 58

You'll have to do this outside of CAESAR II. Use an editor to adjust the values. Be sure to set the filename to caesar.cfg (not caesar.cfg.txt).

The tolerance of 30 degrees seemed to do it. That means if the change in rod angle from the previous iteration to the current iteration is less than 30 degrees, the answer is close enough. That's pretty rough.

I can't say the numbers are correct. But my quick look at Load Case 1 didn't show outlandish displacements.