17 Replies Latest reply on May 26, 2015 8:43 PM by John Willett

    No-Penetration Model Without Initial Contact?

    John Willett

      I don't know how to search for this topic, so I'm just going to ask it:

       

      I have a model where gravity must "settle" one part against other parts (initially with very small gaps) that will support it in a precise position, orientation, and load distribution that are to be determined.  I'm getting mixed results -- either the solver runs happily but gives answers that don't see reasonable, or the solver doesn't run at all and says the model is insufficiently constrained.

       

      I'm running SW Premium 2014 SP5.0 (thought 2015 SP2.1 is available).  The contacts are set up for zero gaps and surface-to-surface, and the (static) study properties are:

       

      Gap/Contact: Improve accuracy for no-penetration contacting surfaces (slower)

      Incompatible bonding options:  More accurate (slower)

      Compute free body forces

      Solver: Automatic

       

      I'm beginning to wonder if this problem fundamentally requires a non-linear solution or if it can be done with a static simulation, perhaps with the large-displacement option.  A possibility is that I must figure out in advance the approximate contact points between the supports and the supported part so that the study can be started close to it's final solution.  (Obviously this would be inconvenient, since the problem must be solved in various orientations wrt gravity.)  Any suggestions would be greatly appreciated! -- John Willett

       

       

      P.S. -- Here's some background that might help you understand what I need to do:  I have a sensitive instrument that must be supported on three bolts in an arbitrary orientation wrt gravity.  Conventional wisdom is that one bolt hole should be a close sliding fit, one a short slot pointed at the sliding bolt, and one a somewhat oversize hole.  (The gaps here are only 2.5 mil or less.)  That way strains in the supporting structure (small transverse relative motions of the bolts) will not be transmitted to the instrument.  The problem is that the instrument's weight is shared among the bolts.  For different orientations, different combinations of transverse loads will be applied to it through the asymmetric bolt holes.  I need to compute those forces and calculate the resulting strains in the instrument. -- J.W.

        • Re: No-Penetration Model Without Initial Contact?
          Roland Schwarz

          I've been running a few contact simulations lately. Nothing works if the mesh at the contact regions is too coarse.

            • Re: No-Penetration Model Without Initial Contact?
              John Willett

              >>Nothing works if the mesh at the contact regions is too coarse.<<

               


              Thanks.  I think I've got that one covered.  The mesh is now so fine that the job takes a looong time to run.  Done any runs with incomplete constraints like this one?  Have you learned any other tricks? -- John Willett

                • Re: No-Penetration Model Without Initial Contact?
                  Bill McEachern

                  try the soft springs under solver options. It can be helpful for this type of problem. small amounts of friction can also be helpful. Hopefully the only direction that is free is the one normal to the contact surface. If you have more than one direction free and assuming u r in a static study you could apply "elastic foundation" restraints with small stiffnesses. The static solver is pretty robust for this sort of thing. I would debug using draft element before you blow a lot of cycles with the 10 nodded test. I would go coarse mesh as well Till you get it to robustly solve repeatedly.

                  conflicting BC's can also be a source of convergence difficulties.

                  • Re: No-Penetration Model Without Initial Contact?
                    Bill McEachern

                    one more thing, I use an initial gap of about 0.001" depending on the scale of things of course.

                      • Re: No-Penetration Model Without Initial Contact?
                        John Willett

                        >>...Hopefully the only direction that is free is the one normal to the contact surface. If you have more than one direction free and assuming u r in a static study you could apply "elastic foundation" restraints with small stiffnesses...

                        ...I use an initial gap of about 0.001" depending on the scale of things of course.<<

                         

                         

                        Bill -- Thanks for the quick, helpful response (as usual).  I'm not sure I understand your last comment, however:

                         

                        I'm already checking the Properties/Gap (clearance) box under the No Penetration Contact Set and entering zero (because my clearances are so small, as describe above).  But do you mean that I should also take a guess as to where the "free" part might "settle" and move it very close before the simulation starts?  (Of course my guess could be wrong...)

                         

                        As to the others, there are two directions free, but both are perpendicular to the contact surfaces.  (The part has a slider on the back perpendicular to all contact surfaces.  In SW Premium (static study is the only one available) I'm not finding any elastic connector available (although it is shown in SW Help.  Maybe it's a feature only available under the full Simulation package... -- John Willett

                          • Re: No-Penetration Model Without Initial Contact?
                            Roland Schwarz

                            Maybe try using SW Motion to get a better idea of what will happen before executing FEA.

                            • Re: No-Penetration Model Without Initial Contact?
                              Bill McEachern

                              Hey John,

                              Sorry I went on a holiday rafting down the grand canyon. the elastic support is under fixtures. You should put the parts in as close as a position to where they will initiate contact as you can work out with SWX. If this case has large motions then you need SWX NL and by large I mean it moves more than say 1/2 an elements face length in the area of contact.

                              I meant the gap should be proportional to the size of things but small - if the whole thing fits in say a cubic meter then the 0.001" thing is likely fine. The elastic supports are really helpful here. However, you can get non realistic answers if things move very far - then you need the equilibrium iterations in the NL solver to get things physically accurate.

                                • Re: No-Penetration Model Without Initial Contact?
                                  John Willett

                                  >>the elastic support is under fixtures.<<

                                   


                                  Thanks Bill! -- I couldn't find anything useful in SW Help under "elastic foundation."  What magnitude of stiffness would you consider "small?"  For that matter, what amount of friction would you consider "small?"  (I've used .01 -- perhaps too small? -- and the default .05, but I'm not really sure what I'm doing...) I hope you had a great trip! -- John Willett

                                   

                                  P.S. -- I also wanted to give a thoughtful answer to Jared's question, but I haven't finished experimenting with the options I already had on the table from previous helpful suggestions.  Now I have another to try... -- J.W.

                                    • Re: No-Penetration Model Without Initial Contact?
                                      Bill McEachern
                                      1. small spring rates would be say <1/1000 of the structures stiffness or applied forces - something insignificant when you have a solution.
                                      2. Small friction I would consider 0.05 +/- a bit....not so clear here - whatever makes it work but friction is tricky. Sometime no friction works best and sometimes its a small amount and sometimes its a lot. You have to look at the specific problem and see what contribution it might be making to stabilize the problem. It can also destabilize a problem - maybe more often then otherwise.
                                      3. I'm not really sure what I'm doing...well on the up side you can't learn how to ski powder unless you get in it...same for this stuff. You might want to consider getting help if things are getting expensive or there is significant risk in relying on the results. Otherwise leaning by doing is a good way to learn how this stuff works.
                                      4. Posting an image as Jared suggested would be a good idea.
                                        • Re: No-Penetration Model Without Initial Contact?
                                          John Willett

                                          Bill, Jared, and others -- Here's an image of a (mostly solved) sample problem.  I'm also trying to attach a P&G file in case anyone wants to try running it.  (There are a bunch of unnecessary studies in the assembly, but it was easier to include them than to selectively delete the less interesting ones.):

                                          Sample Run Elastic Support.png

                                          Details:  This sample problem (related to the one described in my OP) has a steel plate with three different holes, each penetrated in the center by identical circular-cylindrical steel rods:

                                           

                                          1) The upper left hole is round and oversize by 5.0 mil (0.005") on the diameter relative to the rod.

                                          2) The upper right is a horizontal slot that's oversize by 0.5 mil in width and by 5.5 mil in length.  (Previously I had tried a slot inclined only 30 deg CW relative to the vertical, which seemed an easier problem and could apparently be solved simply by checking "Use soft spring to stabilize model" in the "Direct sparse solver," perhaps because less free motion was possible.)

                                          3) The lower right is round and oversize by only 0.5 mil.

                                           

                                          I'm now running SW Premium 2015 SP2.1, so these are strictly static (often large-displacement) solutions.  All of these attempts have the node-to-surface no-penetration contacts set to zero gap.  Some have friction and others don't.  All have study properties set to "Improve accuracy for no-penetration contacting surfaces (slower)," "More accurate (slower)," and usually "Automatic Solver Selection" (which almost always ends up being "Sparse").  Many also have the "Large displacement" box checked, and some have the "Use soft springs to stabilize model" box checked.  All of these tests were done with relatively low resolution, but even so some took quite a while to run.

                                           

                                          The idea is that I know the plate should descend (-Y direction) and rotate CCW, but I don't know exactly where the it should "settle" on the three rods in response to gravity, flexure, etc.  Here's what I think I've learned.  Comments would be very much appreciated:

                                           

                                          A) It's necessary to look at the contact and reaction forces in order to determine which solutions are completely bogus.  (This does not guarantee that those satisfying these basic force balances are correct, of course, but it helps rule out a lot!)

                                          B) A straight direct-sparse solution yields garbage, and inclusion of soft springs and/or friction results in failed solutions, even with the "Large Problem Direct Sparse" solver.  An "iterative" solution also yields garbage, even with an "Elastic Support."

                                          C) The (apparently) best results, like the one illustrated above, were obtained with the "Sparse" solver (usually automatically selected; sometimes forced), "Large displacement" manually checked (even though it didn't ever seem to be recommended by the software), and decreasing amounts of "Elastic Support."  As the support was made weaker, more load steps were included in the solution, and it also became necessary to "restart" the solution during the 2nd or 3rd load step, after it complained that "Equilibrium satisfaction was not achieved."  I got up to 24 or more load steps with very small "Elastic Support."

                                          D) It is necessary to compare the reaction forces from the "Elastic Support" to the weight of the object being supported in order to determine that the "Elastic Support" is weak enough to allow accurate answers.  (Again, this is only a self-consistency check and does not appear to guarantee correct answers.)

                                          E) I tried hand-dragging the plate down and to the left with "Move Component"/"Physical Dynamics" to get it closer to where I thought it should end up and then starting the simulation from there.  (Of course nearly all of the mates had to be suppressed and the three rods "Fixed" in order to do this.)  This did not lead to any good solutions, however, and "Large displacement," "Elastic Support" attempts invariably failed.  I don't understand this, but maybe I'm doing something wrong...

                                           

                                          Tentative conclusions (again comments would be much appreciated):

                                           

                                          i) "Elastic Support" combined with "Large displacement" appears to work because it gives some immediate resistance to gravity-induced motion, forcing the solver to add more load steps and thus helping it to get past the sudden collisions that occur along the way to equilibrium.  Sound reasonable?

                                          ii) Even in this simple case the time consumed for a reasonable solution is prohibitive for my application, where there are much more subtle flexures going on in the structure of the instrument to be supported that must be accurately analyzed.  It appears I will have to find a way around directly modeling this "gravity settlement" issue as part of the larger problem.  Any suggestions?

                                           

                                          Thanks to everyone for all the helpful suggestions! -- John Willett

                            • Re: No-Penetration Model Without Initial Contact?
                              James Riddell

                              I'd suggest you select the Direct Sparse solver.  I've seen some issues with the FFEPlus solver especially with contact surfaces.  You might need to directly go to the Large Problem Direct Sparse solver.

                              • Re: No-Penetration Model Without Initial Contact?
                                Christopher Schaefer

                                The whole issue of "settling" is rather.. well.. um, unsettling.  If one components needs to move in to place first, then a purely static solution isn't realy the best choice.  If the time factor can be excluded, and the relative RBM (rigid body motion) is still necessary, than a non-linear static solution is where you would start from.  And initially test the solution with the absence of gravity and, instead, apply prescribed displacement for the component that needs to settle.

                                  • Re: No-Penetration Model Without Initial Contact?
                                    John Willett

                                    >>If the time factor can be excluded, and the relative RBM (rigid body motion) is still necessary, than a non-linear static solution is where you would start from.<<

                                     


                                    Christopher -- Thanks for your input.  If, however, you mean something other than the Large Displacement option in the static solver, then I don't have it.  I'm working with SW Premium only -- no full Simulation add-in available. -- John Willett

                                  • Re: No-Penetration Model Without Initial Contact?
                                    Jared Conway

                                    can you post a picture?

                                     

                                    can soildworks simulation (without nonlinear) handle this? > yes, as long as you're ok with all the loads being applied at the same time. ie gravity to settle it plus whatever external forces.

                                     

                                    what it won't be happy with is if there aren't enough restraints. ie if there are rigid body modes during the settling like others have described then it doesn't matter if you're using simulation or sim prem, you don't have a static problem to start. nonlinear can help by stabilizing the problem and then turning off those restraints but otherwise you've got yourself either a motion problem or a dynamic problem.

                                     

                                    using soft springs or making your own might help. but would need to see what you're working with here to offer any further advice. but it sounds like a stability problem.

                                      • Re: No-Penetration Model Without Initial Contact?
                                        John Willett

                                        Here's another attempt to attach the pack-and-go file mentioned in the previous post.

                                         

                                        For those who haven't already discovered it, SW 2015 (at least the Premium SP 2.1 version) includes simulation results in its P&Gs whether or not the corresponding box is checked.  My VAR tells me this is a known bug.  If you want the part and assembly files only, the recommended work-around is to check the box for "Include simulation results" and then uncheck all the .CWR files that appear in the listing before saving. -- J.W.