Please reference the screenshots.

Some notes:
1) Ducting is significantly hotter than the "pipe."
2) Ducting has a higher coefficient of thermal expansion than the "pipe"
3) Ducting has supports under the vertical sections, with friction.

What I would expect to happen is that because the ducting is so much hotter than the pipe, and because it is partly restrained, that all 3 expansion joints in each horizontal run would be compressing. This is because a relatively unrestrained horizontal run would have neither tension or compression. If the ducting were cooler, then you'd expect tension.

So, while I get thermal expansion of the ducting (the angle of the duct moves away from centroid of expansion), the bellows are also expanding - regardless if I specify 0 pressure thrust.

The vertical expansion joints are the only ones that are compressing. Can anyone explain why the horizontal ones are expanding despite restraints on the ends and 0 thrust?

I cannot see all the restraint types, but from the deflected plot the duct Risers/ELLS are significantly rotating and kicking out and possibly stretching the joints more than the thermal compression component.

Just curious, did you model the ducts with structural elements or what did you use? I have used the string modeler in FEPipe to do some duct stuff in the past but the sections were cylinders but with box plate type expansion joints.

Ducting was indeed modeled in the structural model. There are two zones, and each zone consists of 4 discrete groupings of structural elements. There are two structural models (one for each zone), and if you looked at those individual structures, they're all one mess of overlapping geometry.

However, in the piping model, they're connected by additionally discrete sections of piping that makes it into one coherent model.

The 4 large +Y arrows are under the central "pipe." They are merely +Y with friction.

There are 4 smaller +Y arrows, but you can only see two in the ISO view. The other two are supporting the same 90° elbow, just on the opposite side.

There are no other supports, beyond the 4 anchors, which are true anchors.

Note the deflection scale is pretty significant... x20. The twisting is somewhat fictitious. 1x scale is imperceptible due to the sizes we're talking about.

Vertical expansion joints are compressing by 1/3 inch. Horizontal expansion joints are expanding by 1/2 inch.

Do CAESAR expansion joints thermally expand? Each is only 3.5" thick here, and perhaps its their own thermal expansion that's resulting in these values.

Expansion joint elements in CAESAR II do not develop thermal strain.