Thermal paste never needs to be replaced, in and of itself. Only if separation has occurred (flowed out of joint, oil sweating, dry filler compound), or the joint is opened and the paste must be redistributed (or added) to maintain an excess to squish out on reassembly.
Thermal paste needs to be replaced always when taking something apart. And there shouldn't be an excess at all. Only a thin film should be applied. Just enough to fill the gaps due to surface roughness. In 99.9% of the cases, thermal paste is applied wrong. You can get much more consistent quality by using thermal pads (silicone rubber, graphite, etc).
Thank you for repeating part of what I said
Roughness though, how would you measure that? It sounds noisy, statistical. How do you know when you've put on a thick enough film to fill the gap?
Short of intensive surface characterization, the easiest way by far is to put on a modest excess, and let it squidge out of the joint.
Simple game-theory calculus applies: is it better to have too little paste and expect voids and significantly increased resistance? Or too much and eliminate voids but have a modestly increased thermal resistance as a consequence?
It also depends on adequate clamping pressure, to actually move the paste (or enough movement to help it along, twist and slide and push etc.). Or where paste doesn't manage to squeeze out, how well the design tolerates increased thermal resistance there (by using a more conductive paste, or more conservative derating).
On that note, I've seen some horrendously thick pastes; the PC "enthusiast" space especially revels in their hi-K materials, putty more than paste. Especially easy to misapply those, and end up with a worse joint despite the higher K.
The next best science, is knowing where to apply excess, and how much. A typical plan is to use a halftone density approach, with less near the mounting pads (e.g. bolted modules), more in the unsupported center span; or inversely for parts made with a somewhat rounded surface (intended to flatten out under clamping pressure).
Here's an example of both:
https://www.infineon.com/dgdl/Infineon-AN2006_02_Application_of_screen_print_templates-ApplicationNotes-v01_10-EN.pdf?fileId=db3a304412b407950112b40ed3f71297they don't say how repeatable the surface is (or what or why the surface is like that in the first place), unfortunately, but they do show *a* characterization, and corresponding paste pattern. The resulting fill looks quite even indeed.
Tim