Announcement

Collapse
No announcement yet.

Full and reduced integration result obtain different results in hybrid model UMAT

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Full and reduced integration result obtain different results in hybrid model UMAT

    Hi Jorgen,

    I'm running a code analyzing the creep deformation of a compact tension specimen under a substantial dead load. I'm calling your hybrid model UMAT in Abaqus 6.7. I find that the evolution of stress and chain strain at the notch tip are not similar with CPE4 and CPE4R elements. Specifically, the CPE4 elements predict a chain stretch slightly above that required for failure (e~1.1) after 5000 seconds, and around that time the code aborts due to too many iteration attempts after reducing the step time in abaqus/standard. However, CPE4R elements do not reach such an extreme state even after 500,000 seconds, pseudo-asymptotically reaching about chain strain e=0.99 (e~0.875 at 5000 seconds).

    My question is: what do you think about the divergence in solutions between the various element types using your UMAT (I note some variance in your benchmarks runs)? Given that the difference is that between failure and survival, I can't ignore the discrepancy.

    Is there reason to believe that the creep behavior is better predicted by one element over another?

    I am currently running the simulation with a finer mesh and an incrememtal creep strain tolerance of 1e-6 to see if it improves.

  • #2
    In general, it is not uncommon to see differences between full and reduced integration elements. The difference that you describe sounds larger than perhaps anticipated. What hourglass control do you specify?

    Another thing to keep in mind is that since full integration point elements have more integration points they can sometimes resolve a higher max stress/strain than a simple reduced integrtion point element.

    Finally, I am curious, how distorted are your elements at the crack tip?

    -Jorgen
    Jorgen Bergstrom, Ph.D. PolymerFEM Administrator

    Comment


    • #3
      I am using the "enhanced" hourglass control, as it was the one that resulted in non-zero hourglass stiffness (I guess this is a problem in general with UMATs, right?)

      I thought about the number of integration points, but there was very little differential between IP in the full integration elements. I was concerned about strong gradients giving funny answers, but this appears to not be the case.

      I spent a lot of time messing with the simulation, and got a lot of different sorts of outcomes. There is some inexplicable instability causing large cutbacks, even at low strains in some cases.

      Which brings up another question: does the UMAT specify its own inelastic strain increment tolerance? The message file says that the UMAT is requesting time step reduction, and there is no creep strain reported to control with the creep strain increment tolerance in the *visco routine.

      My mesh is very regular and rectangular near the notch tip, and at high strain the elements are long but not distorted at all (still rectangular). Local LEmax is about 1.5-1.75.

      I think the dramatic stiffening near lock-out causes the cutbacks (most of the time), which is probably a real warning that the validity of the simulation is about to run out. Running in explicit and allowing element deletion will probably alleviate some of that nonsense.

      -Jevan

      Originally posted by Jorgen View Post
      In general, it is not uncommon to see differences between full and reduced integration elements. The difference that you describe sounds larger than perhaps anticipated. What hourglass control do you specify?

      Another thing to keep in mind is that since full integration point elements have more integration points they can sometimes resolve a higher max stress/strain than a simple reduced integrtion point element.

      Finally, I am curious, how distorted are your elements at the crack tip?

      -Jorgen

      Comment


      • #4
        The UMAT only asks for a time step reduction if a numerical problem (something like a NaN) has occured.

        Also, you should not use the *Visco procedure, use *Static.

        -Jorgen
        Jorgen Bergstrom, Ph.D. PolymerFEM Administrator

        Comment


        • #5
          Similar encounter...

          Hi Jorgan and Jevan,

          I had a similar encounter (different deformation results with Full and Reduced Integration). I am not using any Usermat, I use *Visco and coupled-temperature dispalcement axisymetric analysis (Quad elements) with large deformation. I found that if RI (Reduced Integration) is used, the displacement response (as function of time) reaches an asymptote after certain time period. I used default hour-glass control in Abaqus. While, it doesn't if I use Full Integration (material flows - deformation is monotonically increasing with time)

          While I am still looking into why this is happening.. any further remarks/suggestions on use of R/F Integration from your side is most appreciated.

          Thanks,
          Jay

          Comment


          • #6
            Reduced integration point elements are "stiffer" than the corresponding full integration elements. Perhaps that is what is causing a difference in your case.

            -Jorgen
            Jorgen Bergstrom, Ph.D. PolymerFEM Administrator

            Comment

            Working...
            X