testing
Aerospace
- Jul 19, 2013
- 127
I'm attempting to extract modes of interest from a model. I expected this to be somewhat straightforward: run a frequency analysis and downselect modes based on frequency and the modal participation factor/effective mass. My model is basically a constant thickness shell of one material tied to a variable thickness shell of a different material, essentially, a tube with insulation. When I looked at the participation factors, I expected to see the first bending, torsion and elliptical modes dominate the participation factor accounting for a larger portion of the response. This wasn't the case, in fact, the first bending mode's participation factor was basically in the noise compared to some higher frequency localized modes.
I did some more digging and found that the modal participation factors output by Abaqus varied depending on how my model was setup. Originally, I created an assembly of two parts representing each shell. I updated my model to create a flattened input file without parts/assemblies. The frequencies reported matched dead on, but the participation factors varied wildly with the largest value coming from wildly different modes depending on which model I was looking at. This was in 6.14-1. Running the same models in 6.13-1 gave the same mode for the largest modal participation factor, but other smaller, but still significant, values did not match up well at all. Running either model in both versions of Abaqus gave different values as well.
To dig down further, I created some simple models to try to figure out what was going on. I created a rectangular beam and extracted the mass matrix and mode shapes from Abaqus. I used an existing MATLAB routine on hand to separately calculate the same thing for both simply supported and free-free boundary conditions. I found that calculating the modal participation factors (for x, y and z-rotation for the beam in a plane) outside of Abaqus, the x-values calculated from the Abaqus matrix and shapes matched exactly with the values reported by Abaqus (no matter the magnitude, ie, the modes with a value O(10^-2) matched to the same number of decimals as those with values O(10^-18). This was not the case for the y and z-rotation values. The values with larger magnitudes matched reasonably well, but the smaller values were generally off by orders of magnitude. I did at least see the trends I expected as far as the first bending mode having a large participation factor, which then reduced considerably for higher modes.
I tried to replicate the issue I was seeing with strange/unexpected modes having high participation factors and expected ones have negligible values by creating a beam consisting of two shells tied at the centerline, but was unable to replicate the weirdness seen in my full model. Any suggestions on additional steps to try? I'd like to replicate the issue in order to send a model off to Abaqus support since I'm unable to send my actual model.
I did some more digging and found that the modal participation factors output by Abaqus varied depending on how my model was setup. Originally, I created an assembly of two parts representing each shell. I updated my model to create a flattened input file without parts/assemblies. The frequencies reported matched dead on, but the participation factors varied wildly with the largest value coming from wildly different modes depending on which model I was looking at. This was in 6.14-1. Running the same models in 6.13-1 gave the same mode for the largest modal participation factor, but other smaller, but still significant, values did not match up well at all. Running either model in both versions of Abaqus gave different values as well.
To dig down further, I created some simple models to try to figure out what was going on. I created a rectangular beam and extracted the mass matrix and mode shapes from Abaqus. I used an existing MATLAB routine on hand to separately calculate the same thing for both simply supported and free-free boundary conditions. I found that calculating the modal participation factors (for x, y and z-rotation for the beam in a plane) outside of Abaqus, the x-values calculated from the Abaqus matrix and shapes matched exactly with the values reported by Abaqus (no matter the magnitude, ie, the modes with a value O(10^-2) matched to the same number of decimals as those with values O(10^-18). This was not the case for the y and z-rotation values. The values with larger magnitudes matched reasonably well, but the smaller values were generally off by orders of magnitude. I did at least see the trends I expected as far as the first bending mode having a large participation factor, which then reduced considerably for higher modes.
I tried to replicate the issue I was seeing with strange/unexpected modes having high participation factors and expected ones have negligible values by creating a beam consisting of two shells tied at the centerline, but was unable to replicate the weirdness seen in my full model. Any suggestions on additional steps to try? I'd like to replicate the issue in order to send a model off to Abaqus support since I'm unable to send my actual model.
