Help with multi-body 2nd order analysis in WAMIT

Discussion in 'Hydrodynamics and Aerodynamics' started by kattenbach, Jan 19, 2016.

  1. kattenbach
    Joined: Jan 2016
    Posts: 5
    Likes: 0, Points: 0, Legacy Rep: 10
    Location: Houston, TX

    kattenbach Junior Member

    Hey, everyone.

    Have you guys ever done a multi-body 2nd order analysis in WAMIT, using the direct method for calculating the full QTF?

    I'm running one right now, and I'm a little surprised, to say the least, with the time necessary for the computation of the force solution.

    My model has a total of 12864 panels over two bodies (no symmetry allowed for multi-body), 32 frequencies and 3 headings, and I'm computing only the difference-frequencies and for only one heading. WAMIT took 1.5h to process the first 3 pairs of frequencies, and it's taking more than 2h to get to the 4th pair. At this pace, it would be over a week before I can get the results.

    Past 2nd order analyses I have here, with 10,000 panels and single-body would take only 2.4 minutes for each pair.

    Do you think there can be something wrong with the analysis/model, or do you think that's simply the expected behavior because of the multi-body?

    Appreciate any input on the subject.
     
  2. jehardiman
    Joined: Aug 2004
    Posts: 3,762
    Likes: 1,148, Points: 113, Legacy Rep: 2040
    Location: Port Orchard, Washington, USA

    jehardiman Senior Member

    Have you looked at all the output (weights, volumes, CGs, displacements, velocities, etc) for a sanity check? Any leaks or sinkers/balloons?

    Also since you are looking at 2nd order, have you constrained the bodies adequately to prevent panel collision? Getting the two bodies in contact can break the solution.
     
  3. CDBarry
    Joined: Nov 2002
    Posts: 819
    Likes: 55, Points: 28, Legacy Rep: 354
    Location: Maryland

    CDBarry Senior Member

    DualSPHysics is an open source smoothed particle hydrodynamics code that runs on multiple GPUS to get fast runs on relatively low end machimes. (Thus eliminating my idea of building a Beowulf cluster out of a pile of Raspberry Pi's in a used deli fridge case.)

    You may want to try that code because SPHI is not sensitive to panel issues.
     
  4. kattenbach
    Joined: Jan 2016
    Posts: 5
    Likes: 0, Points: 0, Legacy Rep: 10
    Location: Houston, TX

    kattenbach Junior Member

    Yes, I checked. The first model got a different volume in X and Y compared to Z, and I found it was due to the gdf file being symmetric, and WAMIT somehow messed up the reflection of the panels. When I ran again using a gdf without symmetry, I would get satisfactory results.

    Panel contact is not a problem, as the bodies are far away.
     
  5. kattenbach
    Joined: Jan 2016
    Posts: 5
    Likes: 0, Points: 0, Legacy Rep: 10
    Location: Houston, TX

    kattenbach Junior Member

    Thanks for the tip, I'll take a look at it.
     
  6. kattenbach
    Joined: Jan 2016
    Posts: 5
    Likes: 0, Points: 0, Legacy Rep: 10
    Location: Houston, TX

    kattenbach Junior Member

    So, I was suspicious about the ILOG and IDIAG options in POTEN/CFG files.
    Those two were turned on on the first analysis, because I used WAMIT files generated by WADAM, and WADAM created the files with those options, so I decided to leave them turned on.

    The manual basically states that the potential solution does take much longer if those are turned on, but says nothing on the FORCE, so I figured it could be the case.
    I don't really have any reason to use those options, so I turned them off and re-ran.

    My suspicions were right, as I was able to get the solution time for the first period pairs of around 40 minutes, half of what I was getting before.

    Now the deal is, I left that analysis running, and after 5 days I only had one third of the pairs analyzed. That means this model would take 15 days to run. Comparing to what I had before, it would have taken 30 days!

    I don't know if this is just as expected and my regular computer simply does not have the power to get me that done quickly or if there is still something wrong, but the 1st order results seem good enough, so we could not locate any diagnostic for it yet.
     
  7. fastwave
    Joined: Jan 2007
    Posts: 128
    Likes: 24, Points: 18, Legacy Rep: 36
    Location: europe

    fastwave Senior Member

    A common problem with panel codes is that they run fast when you can fit the problem in ram. If you go over a size then it swaps to the disk and it is a lot slower.
    You said with 10000 elements it run ok. I would check the cpu usage and disk swapping to make sure it is not swapping.
     

  8. kattenbach
    Joined: Jan 2016
    Posts: 5
    Likes: 0, Points: 0, Legacy Rep: 10
    Location: Houston, TX

    kattenbach Junior Member

    That was one of the first things I checked. I estimated the size of the MAXSCR according to my RAM capacity, but the wamit process never got to the limit I used. I assumed that as all the solution was being used in the RAM already.

    I found one more thing that was increasing the simulation time:

    Using a local CS for the second body.

    Before I was using a GDF file for the second body that used the global CS to define. My second body is translated in the vertical only. After I used the same geometry, but with a local CS at the keel of the second body, and made the 2nd body translation vector non-zero, the simulation time for FORCE dropped an additional 4 times.

    After these two changes (the first one I talked on another post above), my simulation time per pair of period went down from 1.5h to 10 minutes.

    I think that's reasonable now.
     
Loading...
Similar Threads
  1. Ousmane
    Replies:
    22
    Views:
    1,827
  2. Dave_S
    Replies:
    4
    Views:
    1,543
  3. Wally2
    Replies:
    3
    Views:
    2,121
  4. ross whitaker
    Replies:
    32
    Views:
    6,530
  5. CasparvdBroek
    Replies:
    30
    Views:
    5,950
  6. massandspace
    Replies:
    33
    Views:
    4,890
  7. TwistedSister
    Replies:
    9
    Views:
    4,515
  8. flboats
    Replies:
    20
    Views:
    5,811
  9. samar87
    Replies:
    12
    Views:
    3,044
  10. quequen
    Replies:
    10
    Views:
    3,715
Forum posts represent the experience, opinion, and view of individual users. Boat Design Net does not necessarily endorse nor share the view of each individual post.
When making potentially dangerous or financial decisions, always employ and consult appropriate professionals. Your circumstances or experience may be different.