10 Replies Latest reply on Jun 4, 2013 11:25 AM by Jared Conway

    Linear vs. Non Linear Discrepancies

    Adam Chevalier

      Introduction:

      I've got an analysis that involves a rotating turbine wheel and a shrink-fit shaft/bushing combo.  A few years back I completed a linear analysis on the components where von Mises stresses peaked around 85% of yield.  After running the turinbe for a few years my customer came back and basically asked "just how fast can we run this before we get into trouble?".  My goal was to analyze the turbine beyond the point that linear analysis predicted yield would occur utilizing a non linear analysis.  Step 1, in my mind, was to run the non linear analysis at the same speed the linear analysis was run in order to verify that the non linear was set up and working correctly.  Getting the analysis to run was fairly difficult but with a loose enough mesh it seemed to cooperate.  Unfortunately the results between linear and non linear don't match well.

       

      Components: (locations of shrink-fits shown)

      assy.jpg

       

      Differing results:

      stress.jpg

       

      Tabulated results:

       

      Locationvon Mises Stress, Linearvon Mises Stress, Non Linear
      Bore, near nose14789
      Bore, near backface15076
      Blade root, exit13371
      Blade Root, inlet10838

       

      Any thoughts on what could be going on here?

        • Re: Linear vs. Non Linear Discrepancies
          Bill McEachern

          What version of the software are you running? Did you sequence the application of the loads in the NL study? They would be sequenced automatically in the linear study.

            • Re: Linear vs. Non Linear Discrepancies
              Adam Chevalier

              Version: SW Premium 2013 x64 Edition, SP3.0

               

              The model currently has a centrifugal load and two thermal loads applied to it.  How would/should I go about sequencing them?

                • Re: Linear vs. Non Linear Discrepancies
                  Bill McEachern

                  actually on second thought I think the shrink fit would get done first in the NL study as well.

                  I would do just the shrink in the linear study and compare to what you get in the NL study and then do the same for the last step.

                  In my expereince, the shrink fit in NL has had issues in the past. For me it was in relation to subsequent steps and converging.

                   

                  Due to the equilibrium iterations in the NL studies stresses at any sort of idealization/contact get smoothed out a bit and hence lower stresses in localized areas. You should not necessaily look at maximums. I would look at both studies with the same scales (stress magnitudes) and see if they look much different. They should be close. If not then maybe more dianostics should get done. Probe averages over certain peices of geometry as well.

                    • Re: Linear vs. Non Linear Discrepancies
                      Adam Chevalier

                      Unfortunately the stresses (averages, minimums, maximums) don't match very well at all.  The blade root stresses for example are on the outside of the turbine and will be minimally effected by the shrink-fit but show drastically different stress values/distributions.

                       

                      Knowing that you've had trouble running shrink-fits in NL before is a tad discouraging.

                       

                      Am I understanding you correctly that you'd suggest running the NL analysis with the only "load" being the shrink fit and compare that to the linear analysis with the same, shrink-fit only scenario?

                        • Re: Linear vs. Non Linear Discrepancies
                          Bill McEachern

                          Yes, I would do that: compare shirnk fit to shrink fit only.  There are other ways to avoid using a shrink fit so no need to panic. I would compare all loads separately to the best approximation you can to sort out where the issue may be.

                           

                          To avoid the shrink fit you can makethe shaft and or bushing "smaller" and then let them expand to form the shrink fit. This can be done via temperature cycle, load cycle or restraints (a bit trickier). The temperature load you specified seems unrealistic but it isn't illeagal and maybe it represents your situation sufficiently well. I would see what you get by removing it in both studies. Also you can ge the NL study to run faster by adjusting the initial load step to 1 and set the maximum step size to 1 as well. If these values are too large the program will adjust them for you assuming the defaults are what you are running. You can sequence events through "time" in an NL static study. Obviously in a static study there is no time so it is called "pseudo time" - it is just a mechanism to sequence things. You might want to read up in help on this is relation to NL studies if the shrink fit does turn out to be the issue. You will have to sequence things to get where you want to go.

                  • Re: Linear vs. Non Linear Discrepancies
                    Jared Conway

                    on the results plot, the chart is the same on both?

                     

                    same element type and size used on both?

                     

                    and you mentioned that you ran the linear in a previous version, you've rerun it in the current version and it came out the same right? what was the previous version in use?

                     

                    could you elaborate a bit more on the setup of the problem. what restraint are you using, where is it applied. what exactly are the centrifugal loads and thermal loads. for the thermal loads how are they applied? faces..edges..etc.

                      • Re: Linear vs. Non Linear Discrepancies
                        Adam Chevalier

                        Jared Conway wrote:

                         

                        on the results plot, the chart is the same on both?

                        Yes, same scale.

                         

                         

                        same element type and size used on both?

                        Type, yes.  Size, no.  I had to reduce the mesh density in order to get the run to finish.  For what it's worth though I've been doing this for years and would expect the differences in meshes to result in a 5% difference in results, max.  (That would be for linear analysis though)

                         

                         

                        and you mentioned that you ran the linear in a previous version, you've rerun it in the current version and it came out the same right? what was the previous version in use?

                        The original linear analysis was run years ago (2009?).  I did re-run it using SW 2013 and got the same results I had gotten back in 2009.

                         

                         

                        could you elaborate a bit more on the setup of the problem. what restraint are you using, where is it applied. what exactly are the centrifugal loads and thermal loads. for the thermal loads how are they applied? faces..edges..etc.

                        Sure thing.  I'll add a picture below, but here's a descripotion:

                         

                        The turbine shaft (central most piece) is restrained axially (left/right on the cross-cestion above) on the right-most face.  It is also restrained circumferentially about the portion of the shaft on the right.

                         

                        Shrink fits were shown in images earlier.

                         

                        The centrifugal load is shown below and is applied about the axis of rotation.  70,000 RPM.

                         

                        Thermal loads are applied at the edges of the turbine blade inlet (OD) and on the face of the turbine blades at the exit (barely visible in the picture below).  820F inlet, 620F exit.

                         

                        fixtures.jpg

                         

                        Message was edited by: Adam Chevalier for clarification

                          • Re: Linear vs. Non Linear Discrepancies
                            Jared Conway

                            i like the idea of trying to isolate whether the shrink fit is the issue or the loading is the issue

                            that will help you nail down what's going on. it sounds like you're on subs, at that point i'd get it over to your technical support team at your reseller.

                             

                            the one thing to think about on your loads is the temperature BC on the face. in a static analysis that can make for some funky results because the rest of your body won't have a temperature applied to it.

                              • Re: Linear vs. Non Linear Discrepancies
                                Adam Chevalier

                                Thanks for the input.  In an attempt to justify my temperature load methodology I offer up the following explanation:

                                 

                                We use ANSYS CFX to run fluid dynamic analysis on the turbomachinery we design.  From ANSYS we're able to get plots of the temperature profile through the turbomachinery but, to the best of my knowledge, there's no way I can take that information directly from ANSYS and put it into a SW Simulation run.  As such I have to apply temperatures at different points on the component until the result is a profile that closely matches the profile from ANSYS.  Generally I do this in a separate Simulation run and feed the results into my linear analysis.  This time, in an attempt to simplify the analysis ironically enough, I applied the appropriate temperature loads in the non linear run.

                                  • Re: Linear vs. Non Linear Discrepancies
                                    Jared Conway

                                    if you can, i'd go with temp loads applied to the body or run a temperature study first and get the temperature loading at every node.

                                     

                                    i can't think of anyway to import the ansys temp to simulation unless it can make an input file the same as flow simulation. you'd have to talk to the guys at ansys to see if they;ve cracked that code.