11 Replies Latest reply on Jan 15, 2015 5:33 PM by Mark Kamps

    contact friction force

    Mark Kamps

      Hello and thanks in advance for any help or suggestions!


      I'm currently working on a simulation with my two parts. The first part is a surface with an representation of a rough surface, while the other part is a rigid top part.

      These two parts are joined in an assembly, and the rigid top part will be pushed into the rough surface, causing it to deform. The goal is to extract the forces and the deformed bodies for further analysis. (See parts, attached)


      After some testing with coarser meshes and a different surface, I now was ready to go for my final assembly, however this failed.


      The results I got were somewhat strange, since the result/contact force response graph was showing weird values. Unfortunately, i didn't get a screenshot, however I did save it as an .csv file (attached). As you perhaps may see, the graph only has some values, while the most values are missing.. How is this possible? The deformation of the body, the stress, the strain all looked good, and I didn't get any errors..


      I will explain the simulation and its constrains. When you cant open the files, try to see the simulation as two cubes being pressed into another, where the upper cube is rigid and non deform-able. The parts are 1x1x1mm in size.

      - The lower part is fixed on its lowest face. The upper part is fixed in z and x movement (only allowing y movement). The upper part is fixed in y translation (0.04mm).

      - A contact set has been made (no penetration) between all the surfaces of the lower part (the rough surface itself) and the lowest surface of the upper part.

      - The lower part (surface part) has been divided with split lines to allow for mesh control, the upper part has a 0.015mm mesh, the middle part a 0.05mm mesh and the lowest part has a 0.1mm mesh.


      For the settings, the Large Problem Direct Sparse is selected. This best suits the non-linear simulation? The LPDS is better than the Direct Sparse since the computer doens't have that great resources? The meshing results in around 700.000 DoF. My computer is a i5-3570 with 8gb ram and a hd7950 GPU.


      So my questions are, how and why are these response graph values missing/incorrect? How can i solve this?

      Which solver is best to use and why, I couldn't really find a best answer.

      What does the solver Method has for influence? (the Newton-Raphson, or the Modified N-R?) Which should i pick?


      Any help is appreciated!



        • Re: contact friction force
          Jared Conway

          what is the application?


          is this an educational or commercial application? do you have access to tech support and further training from your reseller?


          what is your expectation regarding the results? do you have any benchmarks that you can check?


          and what is the further analysis that you're going to do?


          i haven't had a chance to look at your files but from what you describe it seems like an odd application for fea and that you're going to need a really high mesh count to get any good results for the contact and forces. i would be focusing on that before you worry too much about what solver to user and what solver parameters.

            • Re: contact friction force
              Mark Kamps

              Thanks for your reply.


              It is an educational application, therefore no training from a reseller or something like that.


              The goal with this simulation will be to verify certain models. For the prediction of forces on surfaces the currently used model is the Greenwood Williamson model, however a new model has been developed which is also taken some other values into account and it claimes to be more precise than the currently used GW models. As a project my goal is to replicate the situation described in the models, and give exclusion over which model is more reliable.


              With this analysis the forces will be extracted, just as the deformed body. These forces will be made dimensionless by some mathematical calculations. The deformed bodies will be turned into a PointCloud file using a custom MATLAB code, to see how the surface has changed (mean of the surface, height of individual asperities (spheres of the surface). With these values a graph can be constructed to compare this analysis and the models.


              My expectations are that i will get a proper response graph for the forces. I did the analysis on some other surfaces, which were coarser meshed and less precise surfaces, and they did give me the right results. I can't figure out why this doesn't work.


              Also, when i try to make the mesh finer, the analysis will fail. resulting in several errors:

              -Zero or negative pivot point, x at row x.

              -Solver has numerical difficulties.

              I can't properly find what those errors mean.


              To explain, I wanted to use adaptive meshing, however this is not available with non-linear studies?


              Also, the goal is not to create a 100% reliable result, which can be shown to a client or something. The goal is to give exclusion over the models. The difference between the models is so big, that even a give or take result from this model is enough to give some clue.

                • Re: contact friction force
                  Jared Conway

                  the errors indicate a setup and convergence issue. i'm still not sure i understand the reference to greenwood williamson. are you trying to validate the contact method used by solidworks simulation?


                  and no, adaptive meshing is not available in nonlinear. you will need to improve mesh manually.

                    • Re: contact friction force
                      Mark Kamps

                      Thanks for your reply.


                      Did you have a chance to look At the files? Did you notice things were setup wrong? The only things i can think of is a too complex surface, is that possible?


                      In regard to the greenwood williamson model, within my university a new surface contact model has been made. This model will account for surface interaction, so the surface deformation. The GW model does not account for this. This will result in much lower surface compression forces. My goal is to calculate these forces usinf FEA.


                      With these forces i hopefully can conclude which model best fits reality (GW or the new model)

                        • Re: contact friction force
                          Jared Conway

                          i'm still not following the greenwood williamson model


                          the contact that is calculated by solidworks simulation is based on their methodology. i don't see how the forces you come up with will translate to validation of some other methodology. they are all approximations and come with their own assumptions. IE, there is no way to say the FEA is exact and the correct number. only a physical test can tell you that.

                          • Re: contact friction force
                            Jared Conway

                            which study should i be looking at? nonlinear1 or nonlienar2?


                            any particular reason for using nonlinear dynamic vs nonlinear static?


                            to stabilize the problem, i'd set the faces of the moving block to fixed in the normal to direction. you're pushing downwards on the bottom face of it anyways.


                            the way you have defined your analysis you're asking the software a strange question. you have the bumpy surface but then have a rigid face that you're pushing into it. is that really what you want?


                            any particular reason for using the large strain option?

                              • Re: contact friction force
                                Mark Kamps

                                Sorry if I explained it unclear.


                                You are right in regard with the contact forces, those are based on the GW model. However, when using the model on a surface, to generate a contact force versus plane seperation graph, the GW model doesn't account for deformation of the surface. This causes the forces needed to compress the surface according to the GW model to be way higher than (hopefully) the FEA. The newly developed model does account for surface deformation in its calculations, therefore the force to be lower.


                                The reason I am using a bumpy surface (The rough surface) and a rigid plat, is to calculate the contact forces (which will be dimensionless) and the plane seperation (a value that shows the surface deformation, will also be made dimensionless). So yeah, this is excaclty what I want.


                                You should be looking at nonlinear1. The other one is merely for changes since 1 didn;t work.


                                For the large strain option. I didn't know whether or not it was needed. therefore I put it on. Does it mather that much?


                                Obviously this is no scientific prove that something works or not, that is not its goal. Its goal is to see if there is any truth to the claim.

                                  • Re: contact friction force
                                    Jared Conway

                                    large disp is for when you're using certain material models (help should elaborate on that), it changes what your stress/strain input is


                                    my opinoin would be to add more restraints so that the plate goes downwards


                                    start with a deformable plate (push it downwards from the top)


                                    and start with about 1/8th of the total surface to test your methodology

                                      • Re: contact friction force
                                        Mark Kamps

                                        Thanks for another of your replies!

                                        I changed two of the fixtures on the plane to roller sliders, so it can only move downwards. I still have the bottom plane of the top part as a translational fixture.


                                        It is undesirable to change it to a deform-able plate, this means that the plate itself absorbed some of the forces, causing the results to be inaccurate. The upper plate is supposedly 100% perfectly rigid, meaning it will not deform or change the outcome. That is the reason why this fixture is chosen. This fixture enables that perfect rigid effect. If there would be a way to run the simulation without this top part, it would be even better. As long as I can compress the lower part with a linear force. (the force has to be spread like that plate would, not something like pressure, which will act on the lower part evenly devided.)

                                        The force has to act on the top of the 'bumps' first, than it has to move downwards, after each step moving further down, touching and deforming more and more 'bumps'.


                                        The surface itself (so only the bumps, not the underlying 'body') is Gaussian distributed, meaning that it is save to say that the distance the mean of the surface (which is known) an the highest bump is around 3sigma. This sigma value is known. The goal is to compress the surface a little more than 3sigma (lets say 3.5sigma). In this case, the part is 1000um (micrometre) high, sigma is 12um. The compression I am willing to get is 40um.


                                        1/8 of the surface is 125um, that will deform not only the surface (the bumps) but also the underlying body.. How will that help me?





                                        Anyway, I tried some things myself, and I notice that whenever my mesh is to fine, the simulation fails.. A coarser mesh gives results.. Isn't that strange?