Could you try to run in transient? In most cases, NaN values are a clue for severe numerical instability which is easier to track down in transient. Heat transport models through galleries are hard to solve, as they involve very strong parameter contrasts in between the gallery and the surrounding rock. You may have to try to reduce conductivity in the gallery in order to gain on stability.
You may want to have a look at this publication: https://link.springer.com/article/10.1007%2Fs10230-008-0063-3