A friend design a machine manufactured in welded steel plates, this machine was modeled in Solidworks 2013, creating all welded part and assembly. Subsequently, he developed studies stress and fatigue, using the Simulator Solidworks.
I've checked at the design. When I run the command "interferes detection" the result is 40 small interference in the assembly.
I have question, Is valid this study FEA if the plates (part) were overlapped in the model?
Well, it's hard to say if it's valid or not just by your description, but it is possible to have it running although you have interference in your assembly.
One solution is, of course, to have no contact at all, but this is probably not the case here.
Shrink fit is another solution.
Or if you change the global contact condition to no penetration it will also run (or if you add no penetration contacts manually).
if the interferences are very small, the sfotware will just assume that they are bonded. but remember, you're letting the software decide what happens in that region. the material, it choses, the "loss of stiffness" might be a factor, how those nodes work together might not act the same way you expect them to act.
this is one of the reasons we always recommend removing them. worst case, just clear them up with a cavity function.
check out this vid from chris ma on my team who talks about the tools that can help you with this: SOLIDWORKS Tips & Tricks - Combine, Cavity & Indent: What's the Difference? - YouTube
and here are the general recommendations that we give people for flow simulation which also apply to solidworks simulation: SolidWorks Quick Tip - 4 Things to Do Before Every Flow Simulation Analysis
Normally I have been unable to even mesh a solid with overlapping parts so my first answer would be that the FEA is invalid. Reflecting on the overlap issue, since nodes must transfer displacements/loads to connected nodes I can't imagine how this would work. The only possibility would be if SW could ignore the overlapped region and somehow insert a boundary between the two items in order to even mesh. I would still certainly mistrust any high readings in those affected areas.