Hi everyone, I met one problem and really need you guys to help me.

I used H-adaptive to attempt to get accurate results for static analysis. After 5 loops' calculation software said that the calculation has achieved 98% accuracy, so I thought the stress calculated must be accurate enough. But it turned out that my maximum stress is much greater than material's yield strength, which tells me I should consider using Non-linear and the maximum stress value may not be correct.

here is the problem, since stress value will not reach convergence if the singularity exists, those values will grow bigger with bigger increment every time I refine my mesh. But using h-adaptive is like having an automatic mesh refinement controlled by the software. Then reaching 98% similarity between calculated values from the closest two loops should give me an answer that stress has a tendency to convergent. But how could a static study using h-adaptive tell me the maximum stress of a yielded part? Can H-adaptive help us tell stress concentration or singularity if the stress is greater than yield strength?

Singularities are something one needs to get comfortable with because if you use linear analysis they happen all the time. I feel the best approach is understand what really happens. When metals (all materials really but let's stick to metals as most people seem to understand them) go over yield the stiffness drops and they turn to mush to the portion of the loads that results in post yield behaviour. When the stiffness drops this fraction of the loads gets locally redistributed. If the surrounding structure has the capacity the load bleeds out and all is fine. If the surrounding structure does not have the capacity to redistribute you get fracture if the load persists. In an elasto plastic NL analysis in SWX sim you don't get all the way. The best you can get if you are lucky the entire section goes above the true ultimate And the analysis fails. Chances are element distortion may cause the solution to diverge prior to that. If the solution completes chances are the load can be sustained with a good (I.e. Conservative) approximation of the post yield behavior. If the analysis fails best to assume failure. What to do With a linear analysis? Remember everything is an estimate. Singularities are nodal And typically only at the node on the singularity. Try plotting the elemental stresses, they are not subject to the influence of other elements in the projection of the integration point stresses in the nodal averaging. In Swx sim, as far as I know, the integration values within an element are averaged to get the elemental stress. This tends to limit the magnitude of any local stress risers as nodal stresses are not presented. If they are above yield then you are likely to have real or incipeint yielding. If the volume of elements that are post yield is very small you can often assume they will bleed out. So, if fatigue is not a concern you are good to go. And here we are likely talking high cycle fatigue so over 10k cycles. If ratcheting is suspected, the stresses grow with cycles and don't level off due to work hardening, then an NL analysis should be done with cycles. The other thing is just look at the stresses a node or two away from the singularity. I do NL analysis as it requires a lot less explaining in a report. This doesn't work if you are doing a linear dynamic analysis, you rationalize and or do some supporting analysis to support rationalizing them away. I hope that helps.