I am running a rotation check on my model. I am rotating the entire model via one node through one radian, about the x-axis. When I look at the results, there are a few nodal locations where stresses are extremely high for no reason, like 500000 Pa. This high stress only shows up though when I have "use corner data" turned on. When I turn it off, the stress is around 2 Pa. This seems to be the result of some unstable numerical issue, but I cannot find how to solve it. Does anybody have any insight into this behavior?
I think when you swicth off the corner the conotur only plots the stresses at centroidal element. Not sure if in your model are you applying elements like rbe2 to those elements leading to high stresses .
Jon, oddly there are no RBE elements attaching at this location.
I changed the contour plot to elemental, then turned on "use corner data", and check the "no averaging" option in the element discontinuities option. This should (supposedly) give me the "raw" values calculated in the solution. When I do this, I see multiple other elements that have a random high stress. (see attachment)
When I turn off "use corner data", then the stress drops back down to 2 Pa. Just for my knowledge, does anybody know how the centroidal stress is calculated. As I understand it, gauss point stresses are calculated by 1. calculating displacements at the nodes. 2. Extrapolating displacements to the gauss points. 3. Calculating strain at the gauss points. 4. Calculating stress at the same gauss points with contituitive equations. 5. Extrapolating stress to the nodes using shape functions. (Please correct any of the previous steps if they are not correct.) What I don't understand is how centroidal stress is calculated...through the same process of extrpolating displacement, then calculating strain and stress at the center? Is the centroidal stress then extrapolated back to the nodes if "use corner data" option is turned off? The reason I ask all of this is because it is strange that the centroidal stress can be so low (2 Pa) while the corner data can be dang high (~500000 Pa) for the same element. It is baffling. The displacements all look reasonable.
The attachment is a generic panel (looking normal to the plate). The gap at the bottom is where two components are attached via beams and RBE2s, but the high peak area on the bottom component does not have the RBE2 interfacing with it. This is one side of a box structure, so there is depth to this component. (I'm just obscuring the context just in case I am not able to show the entire model in full context.)
Can you post some part of model. If you plot stress components istead of equivalent stress then you can make error in definition of element csys.
The process to calculate the stress should be like you are telling in a linear analsys:stresses at gauss points and extrapolating . In non linear they are copied I think
As Karachun says maybe you can attach part of the model because the stress distibution looks "strange". What I see is that the region at the top with high stresses shows lack of connectivity between elements. One edge has 3 nodes whereas the other 2 nodes? Without model diffcult to say.
Thanks for everybody's responses. Unfortunately I was not able to post the model due to restrictions. I tried a goodly number of trial and error solutions. I got the stress down to 23000 Pa, or roughly 3 psi...within an acceptable range for me. I wish that I could exactly pinpoint what it was that reduced the stress, but I think it had to do with a bolt pattern not spidering out on one of the beams to the surrounding material. I was wondering if it had to do with hourglassing effects, so I increased the order, but realized it wasn't the issue (as I believe FEMAP does not have a single-integration quad plate element, which is usually where hourglassing is manifest). So I changed it back to linear elements, and in rerunning the stress decreased. Somewhere in between all of that I added the RBE2s to the ends of the beams, so I'm not entirely sure if a combination of these issues worked. It seemed to me that I tried each of these solutions individually and neither worked, but then somehow it worked. What I'm saying is I can't pinpoint exactly what caused it to stabilize I did try out a handful of other minor modifications that I didn't list here, so perhaps that is where the solution improved. Either way, I double checked the translations and rotations and things appear to be in order, with a low stress resulting.
I was discussing this issue with a coworker this morning, and he mentioned that at a previous job, he was performing a rotation model check. Try as they did, their team couldn't get their model's stress below a couple ksi during a rotation check. The rotations checked out fine, but the stress was still appearing. Several experts at the company had no answer except to say it was close enough if the rotation results checked out, and if the stress results were several orders of magnitude lower than the actual working stress. Whew. That was killer.