You need to create a physics shape, export as .obj and convert that physics shape to .phy using the phygen or obj2phy.
What is the difference between a model and a physics shape? nothing appart for the job it is meant to do. A model alone wont have any physics interraction so it can be highly detailed, the physics shape on the other hand needs to be extremely simple. When you convert a model in phygen or obj2phy it will automatically patch it together to the simplest shape possible. A continious torus is a perfect example of how this is no good.
Same if you want to make an inward corner for a pipe, you will need to make it in two or three, perhaps even four cylinders depending on the shape of the bend and how accurate you want your physics to be.
I will provide some illustrations.... and excuse my drawing skills.
First lets see what might happen if you make a corner from one single mesh.
The red line is the resulting physics shape.
The converter will make the most effective physics shape possible, even if that isnt what you want at all.
To make the corner correct, you need to devide that corner up somehow, like this for instance.
Here the green line is the resulting physics shape, as you can see it now have the corner made correctly. That is because you actually give the converter additional instructions this way.
At least, this is what I needed to do when making physics shapes in the past, for Source (garry's mod) for instance.
Hope this helps.