Discussion Forum

Posted: 1 decade ago 12 sept. 2013, 17:49 UTC−4

Hi, Without looking at your model in detail, I have the following comments: 1. Stress-strain curves with downgoing slopes will in general give models which cannot be solved once the strain has passed the peak. The material is then (locally) unstable. It may work on occasions, depending on the how the material is confined. 2. When an elastoplastic analysis has problems to start, it is often so that the initial load step takes the material far into the plastic regime. This can be checked by disabling the plasticity node and just run an elastic analysis and check the stress state. 3. You could also try by running only the heat transfer problem and see that it solves properly. I hope this can help you to track down the problem. Regards, Henrik

Posted: 1 decade ago 13 sept. 2013, 04:20 UTC−4

Dear Henrik, Thanks for the response. But does that mean for stress-strain curves like mine where stress reaches a peak then comes down, it is not possible to solve in COMSOL? Regards Anuj

Posted: 1 decade ago 13 sept. 2013, 05:11 UTC−4

[QUOTE] But does that mean for stress-strain curves like mine where stress reaches a peak then comes down, it is not possible to solve in COMSOL? [/QUOTE] Yes, except for some displacement controlled special cases. This is a limitation on the physics rather than on the numerics. The material model is unstable. In order to be able to measure those curves in the lab, the machine must be displacement controlled. Most engineering problems are however load controlled. When being on the downgoing slope, the material actually releases energy. Unless the strain energy in the surrounding material increases faster, the situation is unstable. Note also that the downgoing part of tensile tests is more of an effect of test specimen geometry than a real physical property. So for a multiaxial stress state, you can really only trust the part of the curve that is recorded before the 'necking' of the test specimen. Regards, Henrik

Posted: 1 decade ago 13 sept. 2013, 05:26 UTC−4

Thanks Henrik for sharing the knowledge. I have to think a lot. Regards Anuj

Posted: 1 decade ago 13 sept. 2013, 05:31 UTC−4

Dear Henrik, Can we implement true stress-strain curve in COMSOL? Whether it is possible to do my problem by that? Anuj

Posted: 1 decade ago 13 sept. 2013, 09:50 UTC−4

Dear Anuj, You are applying a very high load compared to the low yield stress and you are applying it instantaneously. In a transient analysis it is frequently better to ramp up the load with time. When I significantly reduced the load it worked. Also if I understood your model correctly your maximum yield stress is around 550 Pa (not MPa) which is very low. Nagi Elabbasi Veryst Engineering

Posted: 1 decade ago 13 sept. 2013, 10:09 UTC−4

Dear Anuj, As Henrik stated materials with a negative stress-strain slope are unstable, are better with displacement controlled loading, and may not be trusted if the softening is due to geometric effects (like necking). I just want to mention two more issues: 1. It’s important to understand the reason why the material is softening and the region of the stress-strain curve that you can trust. In addition to necking, a material can softening due to damage. Modeling that requires a different material formulation that keeps track of a damage variable especially if your model involves unloading. 2. In some cases mild softening in the engineering stress vs. strain curve actually corresponds to mild hardening when plotting true stress instead. With large strains (as in the plot you provided) it’s important to specify the measures used for stress and strain. Nagi Elabbasi Veryst Engineering

Posted: 1 decade ago 13 sept. 2013, 11:34 UTC−4

[QUOTE] Can we implement true stress-strain curve in COMSOL? Whether it is possible to do my problem by that? [/QUOTE] If you use the large strain plasticity option in COMSOL, then the hardening curve should be given as true stress-true strain, i.e Cauchy Stress vs. Logarithmic Strain. Regards, Henrik

Posted: 1 decade ago 14 sept. 2013, 02:15 UTC−4

[QUOTE] Dear Anuj, You are applying a very high load compared to the low yield stress and you are applying it instantaneously. In a transient analysis it is frequently better to ramp up the load with time. When I significantly reduced the load it worked. Also if I understood your model correctly your maximum yield stress is around 550 Pa (not MPa) which is very low. Nagi Elabbasi Veryst Engineering [/QUOTE] [QUOTE] Dear Anuj, As Henrik stated materials with a negative stress-strain slope are unstable, are better with displacement controlled loading, and may not be trusted if the softening is due to geometric effects (like necking). I just want to mention two more issues: 1. It’s important to understand the reason why the material is softening and the region of the stress-strain curve that you can trust. In addition to necking, a material can softening due to damage. Modeling that requires a different material formulation that keeps track of a damage variable especially if your model involves unloading. 2. In some cases mild softening in the engineering stress vs. strain curve actually corresponds to mild hardening when plotting true stress instead. With large strains (as in the plot you provided) it’s important to specify the measures used for stress and strain. Nagi Elabbasi Veryst Engineering [/QUOTE] Dear Nagi, Thanks a lot for pointing out the mistake in defining the yield stress, indeed it was not in MPa. After correcting it, it is running up to 14 seconds. I also applied the load with ramp in 20 seconds. My queries: (1) Can we check somehow that the stress-strain curve has been followed correctly at least up to 14 seconds, i am worried because for the curves with negative slope (as in my case) there will be two strain values for the same stress value. (2) As you (and Henrik) pointed out that "materials with a negative stress-strain slope are unstable, are better with displacement controlled loading, and may not be trusted if the softening is due to geometric effects (like necking)", shall i take the stress values up to the peak only (and corresponding strain) in engineering stress strain? or may i take the stress values up to peak in true stress-strain by selecting "large plastic strain" in plasticity node? (3) Could you plz refer me to some example for "Modeling that requires a different material formulation that keeps track of a damage variable " as suggested by you? (4) It would be appreciated if i could get some example or reference for your suggestion " it’s important to specify the measures used for stress and strain" (5) I just plotted the stress-strain data, the plot is attached. I understand that x axis is the temperature and the color band is showing the stress values, but what is along the z axis as 100, 200, 300..., i expected it to be strain which i have not defined up to 300? Regards Anuj

Posted: 1 decade ago 14 sept. 2013, 02:19 UTC−4

[QUOTE] If you use the large strain plasticity option in COMSOL, then the hardening curve should be given as true stress-true strain, i.e Cauchy Stress vs. Logarithmic Strain. Regards, Henrik [/QUOTE] I will ask to you also that as you (and Nagi) pointed out that "materials with a negative stress-strain slope are unstable, are better with displacement controlled loading, and may not be trusted if the softening is due to geometric effects (like necking)", shall i take the stress values up to the peak only (and corresponding strain) in engineering stress strain? or may i take the stress values up to peak in true stress-strain by selecting "large plastic strain" in plasticity node? Regards Anuj

Posted: 1 decade ago 14 sept. 2013, 16:54 UTC−4

Dear Anuj, There are several references for the relationship between engineering and true stress including page 1 of this link: http://academic.uprm.edu/pcaceres/Courses/MatEng3045/EME8-2.pdf . Note that the conversion equations provided apply to incompressible materials, which makes them applicable to plasticity. I would convert the stress-strain data to true stress-logarithmic strain if it’s not in that format already, use the range of data up to the peak slope provided that necking has not physically occurred at that point in the experiment, and select large strain plasticity in COMSOL. I am sure you use COMSOL to create a plot of Mises stress versus plastic strain at different points to check where you are on the stress strain curve. You brought up an interesting point of the non-unique relationship between stress and strain when there is softening. For a specific stress value you may have two or more stress values as you noted. You should therefore use small load increments (or time increments in a transient analysis) to solve these problems. The concept of non-unique stress strain relationship is not restricted to softening materials however. Consider even “regular” plasticity with positive tangent modulus. For a specific value of stress you can be on the yield curve, or unloading from a higher point on the yield curve. Easier to demonstrate with a figure but don’t have one handy. Nagi Elabbasi Veryst Engineering

Posted: 1 decade ago 14 sept. 2013, 17:03 UTC−4

You also asked about damage models. There are probably COMSOL based models involving damage that users have implemented but I did not search for them. You can certainly implement one since COMSOL is very flexible in that regard, though the formulations are not trivial. I haven’t looked at the references in that area for a while but I believe this was a very good paper but too theoretical (JC Simo, JW Ju, On continuum Damage-Elastoplasticity At Finite Strains, Computational Mechanics, 1989). Just because your material softens does not mean it is experiencing that form of damage though. Our company also develops a material library called PolyUMod that has damage based material models (http://www.veryst.com/what-we-offer/polymer-modeling) . This material library will hopefully be available for COMSOL in the near future. Nagi Elabbasi Veryst Engineering

Posted: 1 decade ago 14 sept. 2013, 23:24 UTC−4

Thank you Nagi for the references, of course the paper "On continuum damage-elastoplasticity at finite strains" is very good (too theoretical aslo), but as you suggested that may not be applicable in my case, i have to study that in detail. Please clarify one more doubt- isn't it true that the peak point in Engineering Stress-Strain curve is point where necking starts? If so, then i will take data points up to peak only and convert them to true stress-strain and use in my model. Warm Regards Anuj

Posted: 1 decade ago 16 sept. 2013, 02:02 UTC−4

Dear Anuj, Yes, necking occurs when the force reaches its maximum, that is the peak on the engineering stress-strain curve. The instability criterion for a tensile test in terms of true stress and true strain is [MATH] \frac{d\sigma}{d\epsilon} = \sigma [/MATH] In situations like this, I usually just make a continuation of the hardening function from its peak value with a very small positive slope. Then it is of course necessary to do an a posteriori check on the strains and try to use some engineering judgement on if the solution can be accepted.or not. Best regards, Henrik

Posted: 1 decade ago 16 sept. 2013, 03:33 UTC−4

Thanks Henrik for the comments. For each temperature, i will continue the stress-strain curve with almost constant stress. Could you plz just tell one thing, what maximum strain shall i define for each temperature (as i will have to make a 2D text file)? Just in case if required, my stress strain curves are there on this forum. Regards Anuj

Posted: 1 decade ago 16 sept. 2013, 11:43 UTC−4

[QUOTE] Please clarify one more doubt- isn't it true that the peak point in Engineering Stress-Strain curve is point where necking starts? If so, then i will take data points up to peak only and convert them to true stress-strain and use in my model. Warm Regards Anuj [/QUOTE] Anuj, Henrik, It may not be consequential for you Anuj's model but I just want to point out that when the stress reaches a peak point and starts dropping with strain it is most likely due to necking but that is not the only cause. If your material is a metal then it’s necking, but for other classes of materials like polymers or concrete the drop in stress may be for a different reason. Here is a reference for example for polymers showing experimental stress-strain data (Figs. 2 and 3) that soften in compression and sometimes even harden again at higher strains: http://www.sciencedirect.com/science/article/pii/S0020768305003677. That softening in compression cannot be due to necking. Nagi Elabbasi Veryst Engineering

Posted: 1 decade ago 16 sept. 2013, 11:52 UTC−4

Thank you Nagi for discussing in so much detail and the reference. It is really very interesting to know these. Could you plz also comment something regarding Henrik's last comment and my reply to that? Warm Regards Anuj

Posted: 1 decade ago 16 sept. 2013, 11:58 UTC−4

I think you should extend the stress-strain curves the way Henrik suggested. Regarding the maximum strain values for your curves, sorry I can’t get to that level of details for Forum questions.

Posted: 1 decade ago 16 sept. 2013, 12:06 UTC−4

Thank you Nagi, Whatever you have shared, that is woderful. Regards Anuj