10 Replies Latest reply on Nov 27, 2013 3:27 PM by Jared Conway

    Fastest Contact Setup - Discrete Pairs?  Split surfaces?

    Shawn Mahaney

      I often have assemblies with a roller running in a C-channel rail.  The roller can contact the front, side, or rear of the rail.  Usually I have a good idea which will or will not make contact, depending on the loading, but have found that they all possible directions need to be defined to get a solution.  As I watch the solver pick through all the possible combinations I suspect that I coould help it along byt being more explicit and putting in everything I already know for sure.

       

      What is the best way to set up the contact pairs, for speed?  [see example screencaps]

       

      I could put all the surfaces in one set, the cylindrical and flat face of the roller and the three flat faces of the rail.

       

      The side-thrust and rolling motions could be separated, so the flat face of the roller is in one set and the cylinder in the other.

       

      All could be set up separately.

       

      But how about the cylinder?  Does the solver have to do anything special to deal with this shape that comes around on itself?

      In some cases I have split the cylindrical face and then had three separate contact sets with nothing in common.

       

      Another question not illustrated here - Since the rails are usually very long compared to the roller size, and I know generally where the roller might contact, does it help any to split them up so that the entire length is not considered?

        • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
          Bill McEachern

          My view would be the last case  above but with the split lines moved (and added to both surface sides) to the smallest region of contact you feel confortable in specifiying (ie you know where contact will occur and just specify those regions.For solve speed that would be fastest. The other thing is that for the round to flat surface to surface contact is likely better for results and node to surface for the flat to flat for speed.

          • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
            Jared Conway

            are we talking solve speed or are we talking setup speed?

              • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                Shawn Mahaney

                Solve speed.

                 

                In my case there is usually a completely free moving body, touching another body only at four rollers in a pair of rails.  The moving body is further restrained by chains or hydraulic cylinders, which I represent as "link" elements tied to yet other bodies or anchors in space.

                The compound load on the moving body can cause it to touch at one of many combinations of the roller faces.  I try to load the model as literally as possible and let the program figure out which contact condition is right.

                  • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                    Bill McEachern

                    is there any initial contact? If not how big are the gaps? Have you tried point to point springs with inconsequential yet finite values instead of the links or at least some of them as maybe appropriate for the case? Is this is a static study or an NL study? Have you tried a coarse mesh in the contact areas and what does a fine mesh get you? How long are we talking here for how many DOF?

                      • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                        Shawn Mahaney

                        Linear static.  I wouldn't use the small springs as these things really do move a lot.

                        There is no initial contact.  I've been modeling things so that the 'rollers' come out to 1/4 millimeter from the rail surfaces. [in reality this is about how we spec and shim them to sit]

                         

                        Here's a snapshot (displacement plot) of the most complicated one I've managed so far - a four stage fork lift mast, with carriage and forks.  There are four sets of rollers, four per set.  In addition there are two pairs of chain and a pair of hydraulic cylinders each of which bridge across three of the four sections.  Another pair of chains lifts the carriage in the top section and another par of cylinders controls the fore/aft tilt of the whole thing.  At full exetension and rated load, the top twists and bends to the side multiple inches, which is what we do see in real life.

                         

                        I've got this one running in just over an hour with a little under 300,000 second order elements, 1.5 million DOF.  But other models set up similarly take a lot longer to run, and I'm trying to narrow down why.

                        • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                          Shawn Mahaney

                          Also, I have found it good to do a modest mesh refinement near the expected contact, for stability and precision of results.  In particular bearing connections like having multiple elements across the inside diameter and a dozen or more elements around the outside diameter. 

                            • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                              Bill McEachern

                              What is the object of doing the analysis like this? By this I mean modeling these contacts this way in the context of the system level analysis shown. What are you after and why do think you need to approach it this way. You can certainly get the right load path in the mast without resorting to this level of complication at the system level. It depends what you are after though. Also the problem looks like it might be better handled in the NL solver given the large displacements of the mast. Why analyze the forks in this analysis when you could remove them and use the remote load as an example.

                                • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                                  Shawn Mahaney

                                  The ultimate goal is stress values and total deflection of the assembly.  Down the line I'd like to get some idea of the total stored energy, which impacts the stability of a maneuvering fork truck.  I also run the physical testing lab, and it's an all day job just to set up one loading, it's difficult to get just total deflection let alone twist, and right now we don't even have organic capability to tilt sideways (which is part of a standard stability test). 

                                   

                                  We've tried non-contact analysis a number of ways, such as modeling each piece one at a time and tranferring the resultant loads to the next piece manually, or tying the weldments together with skinny solid bars.  All we'd get out of that is stress results with a lot of uncertainty.  So one I day I started trying to be more and more literal about it.  It took a long time to get a stable run, but I seem to have that licked.  The last step was adding articulated pairs of chains over simulated sheaves (a rigid pivot on a bearing connector).

                                   

                                  The problem is not non-linear at all.  Stresses in each weldment are all well below yield.  The total displacement in a bad case is only around 1.5% of the extended length, and much of that comes from take-up of slack, even in my ideally shimmed unit with .25 mm initial gaps.  I just have to make sure I force it to 'solve with small displacements'.

                                   

                                  I put the forks in because it's one more step toward being totally literal, it looks better in the report, and it's actually easier than working out the moments for each case to apply loads to the carriage.

                                    • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                                      Bill McEachern

                                      OK well you seem to be happy doing it that way so have at it. By the way I was not suggesting removing the contact, though that is always a good idea when it can work. I was trying to get you to consider simplfing it so you get where you want to go more efficeintly say by replacig the rollers with flat surfaces backed by flat solid bits that allow the requisite flexibility and starting the calc with all that in initial contact. But if you think that sort of thing can't work then I am not about to argue with you. AS far as stability goes you might want to consider the motion program and the use of bushings to add the requisit flexibility but that may not provide sufficient fidelity for you either. Everything is an approximation - it comes down to understanding your situation and figuring out what is best for you. Speeding up the calc though is what you posted about and anything about selecting contact sets is going to be small beer in the big scheme of things on this one. You could reduce the DOF count by replacing section of the long solid bits with shells or beams to get it to go faster.

                                      A couple things to note:

                                      1) there are 3 non-linearities in FEA - contact, materail and NL geometry. So I hear you loud and clear on the material. The contact supported in Static is more or less small displacement contact but large displacements are supported in a limited sense - that being if it solves with the large displacement flag set. On NL geometry the general rule of thiumb for determining if non linearities are emerging is generally considered if the deflections approach 1/2 the structural depth resisting those deflections. So in your case that is the depth of the structural sections that make up the mast. You could argue that the bulk of the deflections is coming from the slop in the joints which is probably true. However the over all beam appears to be deflecting significantly so the stiffness of the beam is changing with the defelction and can no  longer be considered small. However it is up to the analyst to make those calls.

                                      2) Use of the remote load does not require figuring out the moments - you just apply the load it figures it out for you. But again up to you but it does add to the complexity of the calc and isn't helping your solution time issue.

                              • Re: Fastest Contact Setup - Discrete Pairs?  Split surfaces?
                                Jared Conway

                                in the long run the way to reduce solve time is not based on how you create them it is how many you create.

                                 

                                i haven't read the full back and forth between you and bill but i think he's saying the same thing that i'm going to say:

                                1. if you can simplify to bonded, do it

                                2. if you can simplfy your assembly, do it

                                3. use split lines if contact isn;t going to happen on the whole face to reduce the contact that is calculated

                                 

                                when you have your model down to a reasonable size, the solve time is what the solve time is.

                                 

                                to note, you could throw some hardware at this or try 2014 large problem direct sparse sovler but i think in the long run you'd benefit from working with someone one-on-one on your problem to see how it could be simplified..etc. we do offer that at hawk ridge if you're interested. jared@hawkridgesys.com