Request for FEA Analysis

[WobblyHand]

Bruce, have you played with allocating more virtual memory for Freecad to use? I would not want to do it on my old machine because it would probably wear out my HDD but I have a small laptop that has 8gb ram and a 256gb ssd, mostly empty. Might try allocating an extra 8 or 16 gb.

Stan, basically I allocated an additional 21GB of SSD to act as virtual memory, this is on top of 32GB of RAM, that is not shared with the video card.  My video card has 6GB of its own high speed RAM.  Bought this computer 2 years ago, there's lots more capable machines out there.  One of my crazy sims used all 32GB RAM and all 21GB of virtual memory.  It finished, but couldn't write the results to SSD, (for some reason) so the sim was useless.

Even with SSD, once you get thrashing on disk, performance goes way down.  No matter how much RAM you have, it's real easy to use it up with higher and higher fidelity simulations (or even to check if they are good enough).  For people with less capable machines, its really essential to exploit symmetry as much as possible to save RAM.  That's why I originally posted a 1/4 model on GTA, so people could simulate with moderate machines.

Definitely don't want to run out of RAM on a HDD system - you could wait hours or days, and you might trash the disk.  Consumer grade HDD's are not meant to be used as RAM.

[WhatUPSbox?]

Thanks. Fortunately the gamers keep pushing the consumer capabilities forward. I might see what's out there.
But yeah, cleverness in model building and partitioning goes a long way.

[WobblyHand]

But yeah, cleverness in model building and partitioning goes a long way.

Even when I was working, the EMag guys were doing this all the time (exploiting symmetry) because they were always running out of memory.  They had 256GB RAM in their server back then, and they still had to use symmetry whenever possible.  They were modeling large objects at millimeter wave frequencies.  Same problem, too many cells, not enough memory... 

[rsterne]

OK, so I sectioned the model again, so it's a 90 deg. revolution, cut in half length ways (1/8th of a complete cylinder).... I used a 1mm radius on the fillet, and I ran it with the standard 1.00mm max. and 0.00mm min. Mesh, and I got 905 MPa for the stress in the radius.... The mesh had 40609 cells....



I used a 100um mesh in the radius, which then contained 120261 cells, and got 896 MPa for the maximum stress in the radius....



I then used a 50um mesh in the radius, which then had 276408 cells, and got 893 MPa for the maximum stress in the radius....



I then tried using the "Mesh Size from Curvature" in GMesh, and it is easier to use and doesn't make as many cells in the adjacent areas, but concentrates them in the radius.... I tried 24 per 2PI radians (standard is 12), which gives 6 cells in 90 deg. instead of only 3, and got 54724 cells and a max. stress of 906....



I then used 50 per 2PI radians, which gives 12 cells in 90 deg. and got 94064 cells and a max. stress of 896....



In practical terms, the default of 12 cells in 360 deg. seems to be OK, and gives about 4 rows of cells in the radius.... I don't see a problem with the accuracy using that, but if you want to double the resolution in the radius, or at an angle change, you can change that to 24 or 50 without running out of memory....  I tried using 36, which should give a cell every 10 degrees, but for some reason could not get a mesh.... The hoop strength with this model is only 390 MPa (3mm wall at 800 bar), so even the 1mm radius has less than half the strength, not a practical thing to do....

Bob