Inverse fitting viscoelastic properties

[NataCalc]

Hi Steve,

I am using the material (polycaprolactone) for FEA models which exhibits viscoelastic properties (figure of cyclic from 10% strain control shows viscoelastic behaviour from second cycle). I have already inverse fitted elastic properties of material with Ogden 2nd order. Now, I have to inverse fit viscoelastic properties of material. Do I have to start from 2nd cycle (which demonstrates viscoelastic properties while loading) followed by unloading? Or I have to inverse fit from beginning (loading) followed by unloading and then loading (2nd cycle), figure is enclosed?Fig1.zip

Regards,
Nataliya

[NataCalc]

Fig1.zipFor inverse fitting of viscoelastic properties I have chosen load followed by unloading. The analysis is static. For material model I am using uncoupled viscoelastic (block material) with Ogden elastic part. I want to optimise g1 and t1, respectively. For this purposes I am using
(1) 'levenberg-marquardt', FinDiffRelStep=1e-8 and Squared differences for optimisation function. I am getting following results:
Local minimum possible. lsqnonlin stopped because the relative size of the current step is less than
the selected value of the step size tolerance.<stopping criteria details>.
(2) Nelder-Mead simplex direct search:
Exiting: Maximum number of iterations has been exceeded
- increase MaxIter option.
Current function value: 4537.883245

Results of simulation does not perfectly fit those from experiments. How would be possible to improve it?

[maas]

Hi Nataliya,

It's difficult to say how to improve an optimization since there are many factors involved. Some things to consider are, is the constitutive model chosen the best one for what you are modeling and how accurately does the simulation reproduce the experimental approach? In addition, if your initial guesses are far from the actual solution, the numerical method might not be able to find it. You could try some manual runs (by which I mean, just run forward models with different material parameters) to see if you can find better initial guesses. Also, the experimental curve shows some interesting behavior that is not captured by the simulation: It has an inflection point at the point (the little dip) and a sharp transition in the lower branch. Neither of these are captured by the simulation. This makes me wonder if you'll need a more complex constitutive model, e.g. more than one viscoelastic term. But it could also be that the simulation is not capturing the experimental approach accurately enough. Sorry I can't be more precise, but I hope this helps.

Cheers,

Steve