5 Replies Latest reply on Jul 25, 2008 8:24 PM by Anthony Botting

    Frequency analysis - Calibration?

    Scott Pletz
      Hello all,

      A little history first- We (Dometic, LLC) mfg Rv Air Conditioners. For various designs we have many copper tubing paths. When designing a new copper tube we have samples made and then installed in an A/C in the lab for testing. Part of these test consist of frequency analysis for each tube using a scope. As one can imagine this is a very tedious job.

      I have undertaken the task of evaluating CosmosWorks for replacing the lab work. Unless of course findings in Cosmos further warrant scope analysis.

      I have taken the Certified Material Properties supplied by our vendor and applied them to my model.

      The frequencies from the scope differ from those of Cosmos in some cases as much as 20%. In a few cases only 2-3%. I can live with the 2-3%.

      My next approach will be to vary my material properties until Cosmos results come closer to those seen from the scope. This is the only current way I can come up with for "calibration" i.e. calibration of Cosmos to the scope. (B.T.W- the scope just came back from it's yearly check-up/calibration)

      What I am looking for from this forum is others experience in this area. Thoughts, ideas would be greatly appriciated.

      Also, Is there an plug-ins available for connecting the scope directly to the computer and using that data to "calibrate" a specific material.

      These thoughts may be way out there but I decided to take a chance.

      Thanks in advance!

      Scott Pletz, CosmosWorks 2008 SP3.0
      Scope- Tektronics TDS 2024
        • Frequency analysis - Calibration?
          Vince Adams
          Hi Scott, I've been involved in several modal correlation projects in the past and have to caution you against tweaking material properties to force correlation across a 20-30% spread. I'd only consider this for highly uncertain materials (plastics, odd alloys) to cover a range less than 5-10%.

          A 20-30% error is most likely due to a mismatch between the test and the simulation. Restraints and weight variation are common suspects. Depending on the length of the tube, small wall thickness variations from spec can add up to a large % difference in weight. You could tweak density blindly to get correlation but if weight wasn't the culprit, you'll be fooling yourself. Weigh your test samples and make sure any unsupported fixture weight is accounted for.

          Along these lines, minor variations in tube form vs. CAD could stiffen some modes and weaken others. In one project a few years ago, we were performing a similar correlation on a wireform. We had a the actual sample laser scanned and overlaid it on the CAD geometry. Night & Day difference!! When we build a new solid model to represent the actual test specimen, everything dialed right in. Similarly, if you introduce any pre-stress in the assembly of the tube to the test fixture, this could shift your modes.

          Another common cause of test correlation error is the orientation of the excitation and/or coupling of multiple modes. Remember that a Freq analysis in COSMOSWorks provides a "perfect" and "clean" modal response. You may have a slight variation in excitation angle that causes the mode shape to be slightly different.

          How does your mode shape compare to the shape calculated in COSMOSWorks? This should be your first check. It is usually easier to correlate to a mode shape than the actual frequency. If the shapes don't match, no sense in proceeding. Use high speed video or a strobe to see this.

          Hope this helps! Let me know what you learn!

          • Frequency analysis - Calibration?
            Anthony Botting
            HI Scott: we did things like this for spacecraft structures of metal matrix composite (aluminum, with pieces of other materials inside the metal). We actually calibrated the FEA models using test data of simple bar and beam samples of the material, then we could 'play' with the FEA simulation models of actual structure shapes, performing 'what if' studies with great confidence. Since frequency proportional to sqrt(K/M) so getting the density and stiffness of the structures was primary. We conducted lab tests in static deflection on beam samples (as a way to isolate, or extract modulus, E), to get the density we conducted weight and water displacement tests on samples of the material. Once all that was done, as vibration frequency is very, very sensitive to boundary conditions, we attempted to create truly-fixed (or clamped) boundary conditions on simple test bars, and used that to calibrate the FEA model from beam-bending vibration testing. From these tests we 'tuned' the effective material modulus in the COSMOS. We were then able to use COSMOS to simulate spacecraft bus structure vibration. These simulations matched test data within about 10%, which was about the measurement uncertainty level in the lab anyway. The end results was, our customer got what they wanted. I don't know about scopes! Tony
              • Frequency analysis - Calibration?
                Scott Pletz
                Thank you both!

                This will help our engineeering department better decide what direction to take next!

                I will update you in the future regarding all of this.

                Thanks a bunch!

                Geeezzzeee I love this forum...lol
                  • Frequency analysis - Calibration?
                    Scott Pletz
                    Vince and Anthony,

                    Upon further reading your replies may I come to the understanding there is almost no way to completely eliminate variables to get a a consistant scope vs CosmosWorks results? At least no way in a timely manner?

                    Based on what Anthony had mentioned:

                    "A 20-30% error is most likely due to a mismatch between the test and the simulation. ......


                    Along these lines, minor variations in tube form vs. CAD ....."

                    I would need to have the model match test piece exactly. When a tube is formed the original I.D. / O.D. becomes oblong. Now the model does not match the test piece therefore giving a different frequencing results between the two test.

                    If this is the case we could only use Cosmos as a a quick test in the very early stages of the design but would ultimately need to perform a scope test anyway.

                    Is it fair that my interpertation is correct?

                    Again! thank you for any help.... :-)
                • Frequency analysis - Calibration?
                  Anthony Botting
                  Yes. Vince is right. CAD must match test pieces - you should be able to create variable wall thickness geometry in CAD using sweeps with guide curves.