14 Replies Latest reply on Jan 25, 2010 11:54 AM by Joe Galliera

    Random Vibration Post-Processing

      I have a quick question about interpreting the stress results of a Dynamic Random Vibration study.  The values are RMS Von Mises stresses, which (as far as my understanding goes) are equal to one standard deviation, assuming a normal (Gaussian) distribution.  So, when subjected to PSD acceleration, the stresses in each part have a 32% chance of exceeding the stresses shown.

       

      The problem I'm having is determining a pass/fail criterion for this study.  Many of the studies I've seen claim to use the PSD curve (for a specific amount of time?), and then provide results in a damage or life plot.  I looked into using stresses created by the PSD curve and using them as inputs for a fatigue study, but it doesn't appear that SolidWorks has that capability.

       

      I also tried assuming that 3*RMS stresses would ensure that stresses in the part would only have a 0.25% chance of exceeding the displayed RMS values (i.e.: the pass/fail limit is set to 1/3 of the material yield strength), but this seemed unrealistic.  The RMS stresses in many of the robust parts we studied were coming fairly close to the limit (1/3 Sy).

       

      Can anyone give me some insight on interpreting these results?  For reference, I'm using PSD curves defined in MIL-STD-810G, Method 514.6C.

        • Re: Random Vibration Post-Processing
          Kirby Meyer

          Hello Bryson

           

          The best way of addressing this is to approach the system at "Qualification" levels.   If you are using Maximum Predicted Environment (MPE),  take the existing PSD map and multiply everything by 2, or equivalently adding 6 dB to the profile.    Then examine the results and see what the RMS stress values are (IIRC, the RMS stress is due to RMS forces, which are, in turn from RMS accelerations). 

           

          If the RMS stress values at Qualification are below yield stresses for the system, then I would consider it a safe system.

           

          What you are basically doing is doing what has been put into practice for physical testing but now at the simulation level.   Qualification levels are generally never met during the life of physical service,  but are used to effectively 'shake the system down'.  If it works at Qual it usually passes into production...

           

          Kirby

          • Re: Random Vibration Post-Processing
            Rick Nagel

            Not sure if this thread is dead or not but I'll post anyway as I'm suprised there aren't more posts regarding this topic. Maybe not that many people use COSMOS for vibration analysis?

             

            Anywho, I have performed random vibration studies using ProMechanica, NASTRAN/PATRAN, and just recently my current employer has upgraded the COSMOS software and we will probably be undertaking some random vibration studies. To date I have not attended the COSMOS training for advanced simulation so I'm also trying to figure out what results the software is providing.

             

            Your original comment or question seems to go along with the random vibration theory I'm familiar with. I am not familiar with the methodologies that have been advocated by the other posters. I'm not saying they are incorrect, I'm just not familiar with them. I base my analysis approach and interpretation of results on a methodolgy detialed in books by Dave S. Steinberg. Specifically I suggest you take a look at Vibration Analysis for Electronic Equipment. The book gives examples of hand calculations of a single degree of freedom vibration problem and predicts the fatigue life. The basic steps involved with the analysis and interpretation of results are:

            1. Run a modal analysis

            2.  Perform a dynamic random analysis (ie) put in your PSD curve data

            3. Review the predicted stress level (1 sigma)

            4. Create a table of the 1,2, and 3 sigma stress along with number of predicted cycles (cycles are determined by time at test (ie) a one hour test with a resonant frequency of 44Hz would be about 159,000 cycles)

            5. Calculate cumulative damage in accordance with Miners equation

            6. Compare these results on an S/N curve for your material of interest

             

            In a perfect world you would claim infinite life.

             

            I believe this method is very conservative but it is well documented. If this criteria is too rigid for your application, you have to decide as an engineer what analysis methodology is acceptable. Now with all that said, I haven't had alot of success with the post-processing and results tools in COSMOS. Seems like I will have to do a lot of work by hand for a final solution. The one example in the Help section of a circuit card is pretty poor.

             

            Anyway, thats my two cents. I will be digging into this more and hopefully getting some more guidance on exactly how to interpret the results next month at SWW 2010.

             

            Good luck