Long ago, when I was an electrical engineering student, I had a campus job that led to me helping some Mech E students whose grad work involved a novel way to stir molten metal. For safety and convenience, they were using an alloy that had most of the mechanical properties of their target metal, but which would melt in boiling water. I never saw their actual test rig until the day of the first test. I came in to help them review the first batch of data, only to find them staring blankly at the puddle of recently solidified metal around the rig.
See, the advantage of the special alloy’s low melting point was that they didn’t need the resources of a foundry. You could melt the stuff with a hotplate. So that’s what they did–their crucible was an aluminum plate bolted to a normal hotplate…with a cylinder of refractory ceramic epoxied to the metal. With a 60C temperature change, the difference in thermal expansion caused the plate to tear the bottom off the ceramic cylinder.
The lesson: Rigidly joining unlike materials at a critical interface will bite you in the ass. Find another way* if you can, test the shit out of it if you can’t, and expect it to fail anyway.
*If anyone happens to be curious, the other way for the project I helped with was silicon rubber. Remove the rigidity from the join, and the problem goes away. Probably not the best idea for the sub, though.