The answer is, it depends. What do your parts look like, what aspects of them do you want to inspect, is there good contrast between the part and it's background. Pictures would help.

Machine vision is all about contrast. If you can get good contrast on the edges you want to inspect and have a very stable, consistent environment then it can work great. If not it may not work.

Yeah. It really depends. There are some things Machine Vision could see and detect well, other things that MV would be almost useless at. And a huge range in between where choosing specific sensors, lighting, spectra, angles, etc, could make it possible.

Your best bet would be to look for vendors that specialize in this application, and ask them how they would handle your specific application. But in general, "classic" MV is only going to find really obvious flaws in a weld, and even then, what's "obvious" to a human eye is often not to MV.

There might be some Machine-Learning based MV options on the market that might suit. Those need to be trained, however -- basically, pass a very large number of parts through the MV system, and tell it which images showed good welds and which ones were bad. The weakness of ML inspection systems is that they can fail unpredictably when you present them a type of flaw that never turned up in their training. ML systems seem to have intuition, but really, they're just complex pattern-matching algorithms.