Search
Close this search box.
Notifications
Clear all

[Closed] Getting stuck in local Minima

6 Posts
3 Users
0 Reactions
758 Views
Posts: 11
Topic starter
(@dvhudson08)
Active Member
Joined: 12 years ago

Hi,

I am using Mcalibration to curve fit BB model (ANSYS version and built-in both) on a stress strain data which has softening portion as shown below. What I have noticed in such cases is that both BB model option is getting stuck in local minim during optimization process. I tried to get parametric study to find the senstivity and then readjust the parameters. With this readjustment, BB (in-built) is doing very good job in finding the fit but ANSYS built in BB get stuck in local minima again.

Anyone else found such problem?

If I am curve fitting data which does not have softening portion, both work equally well.

Topic Tags
5 Replies
Posts: 3998
(@jorgen)
Member
Joined: 5 years ago

I have not heard of that problem.
Can you provide the experimental data or more images illustrating the problem?

/Jorgen

Topic Tags
5 Replies
Posts: 11
Topic starter
(@dvhudson08)
Active Member
Joined: 12 years ago

Hi Dr Bergstrom,

Thanks for the reply. Please find the data below. I used Bergstrom-Boyce (in built) and it did a great job in fitting this curve without any hassle. However I need to do lot of manual adjustments to get ANSYS-BB to fit. It looks that it was hitting local minima. Also the quality of fit was very different between ANSYS-BB and in built BB. This is the fit for ANSYS-BB:

I dont have figure for in-built BB but it was indistinguishable from experimental curve!

0,0

0.00564,11.03324

0.0116,22.18563

0.01725,32.27076

0.02166,39.27199

0.02547,43.54612

0.02927,47.82026

0.03245,50.67095

0.03629,52.81184

0.03917,54.23973

0.04303,54.72142

0.04787,54.1384

0.05175,53.31644

0.05627,52.37725

0.06176,51.55847

0.06629,50.73779

0.07081,49.56156

0.0763,48.86129

0.0818,47.56846

0.08633,46.51075

0.09085,45.69006

0.09763,44.87383

0.10408,44.41251

0.11149,44.19012

0.11859,43.96709

0.12664,44.10152

0.13566,44.23785

0.14339,44.49016

0.15176,44.86225

0.15982,45.11519

0.16723,45.24834

0.17592,45.62107

0.18494,45.99443

0.19138,46.00716

Posts: 11
Topic starter
(@dvhudson08)
Active Member
Joined: 12 years ago

[QUOTE=Jorgen,12633]I have not heard of that problem.

Can you provide the experimental data or more images illustrating the problem?

/Jorgen

Hi Dr Bergstrom,

I messaged sometime back with the files and the mcal file but I do not see it here. did you get it?

Posts: 5
(@twain544)
Active Member
Joined: 11 years ago

I see something similar regarding local minima, at least thats what I assume is happening. Using the build-in solver, I get a somewhat reasonable fit:

However, when I switch to the FE solver, Abaqus explicit in my case, the calibration just seems to get stuck and gives very poor results:

Ive tried randomly perturbing the solution, smoothing my stress strain data, and using very small time steps, but to no avail.

Posts: 3998
(@jorgen)
Member
Joined: 5 years ago

That sounds odd. I always get good convergence for simple material models.
One trick to keep in mind when working with linear viscoelasticity is that the characteristic times have to be of the right magnitude.

In some sense it would be better to use the logarithm of the characteristic times...

-Jorgen

Posts: 3998
(@jorgen)
Member
Joined: 5 years ago

That sounds odd. I always get good convergence for simple material models.
One trick to keep in mind when working with linear viscoelasticity is that the characteristic times have to be of the right magnitude.

In some sense it would be better to use the logarithm of the characteristic times...

-Jorgen

Share: