Negative Jacobian at Random times

[nyelaura]

Hello,

I am looking for advice on optimizing parameters for a viscoelastic solid with hyperelastic elastic component. I am able to get models working but my issue comes from when I change parameters every run. Sometimes the model works and about half of the time febio does not reach the end of the analysis and gives a "negative jacobian" error. I tried uncoupled vs coupled viscoelasticity, refined vs course meshes, and large vs small time steps. I have tried multiple material models: viscoelastic with ogden unconstrained, viscoelastic with EFD neohookean, uncoupled viscoelastic Veronda westmann, and uncoupled viscoelastic ogden. All of these changes and changes to parameters have given errors at seemingly random times.

I am wondering if the problem is that the stress in my data are too large? Or have I not found the correct model yet? I am optimizing my parameters in MATLAB (attached picture shows my data in orange and FEBio model output in blue - first attachment results)

ViscoEFDNeohookeanOptimization.PNG

Working viscoelastic EFD Neohookean:
bladder_visco_EFDNH_working.feb

Adding a second relaxation strength and getting negative jacobian:
bladder_visco_EFDNH_error1.feb

Increasing k and getting negative jacobian:
bladder_visco_EFDNH_error2.feb

These are just two examples of the many combinations I have tried that have stopped before finishing at 14000 seconds. Any suggestions would be very much appreciated!

Regards,
Laura

Edit: Are there bounds that I should place on the parameters and is there anywhere I can find suggested bounds for each of the different materials?

[maas]

Hi Laura,

Sorry for the late reply. Have you been able to make progress on this? I ran these models and I confirm the problems you are seeing. As far as what's causing them, my best guess at this point is that you are running into an instability (some type of buckling mode). Looking at the non-converged states (by running the model in debug mode), I can see that the mesh is deforming quite a bit both during the sudden jumps as well as during the relaxation phase at larger strains. You could try ramping the displacements increments instead of a sudden jump, but if the material is indeed undergoing some transition due to the instability, then I'm not sure there is much you can do about this. If this is unexpected, then perhaps the material parameters that you chose are not yet in the correct range.
Placing bounds on parameters may indeed help. I'm not familiar with MATLAB's optimization capabilities, but the method implemented in FEBio can sometimes take large steps when the solution is far from the initial guess, which can cause convergence problems. When you see the optimization algorithm take large steps, it is often better to end the optimization and retry with different initial values. I think our online documentation lists bounds on material parameters, and FEBio will often not run when the material parameters are outside the valid range. But even if you stay within the theoretical range, it might help to narrow the intervals so the optimization algorithm doesn't try non-physiological values for the parameters. I hope this helps. Let us know if you need further assistance with this.

Best,

Steve

[nyelaura]

Hi Steve,

I wanted to reply just to let you know that I am grateful for your advice. I ended up using EFD neo-hookean since it was the only material that I was able to get to work. I lowered poisson's ratio to 0.35 and this allowed the elements to not invert on themselves. This project will be passed onto another student so I wanted to thank you and Gerard for all your help even though I will not have more questions on this!

Regards,
Laura