How can I make an algorithm to detect faults from electroplating?

Hello , I am actually doing a project on fault detection so i get the brass probe deposited with copper after applying current, so far I made the segmentation algorithm of the working area, removing background and noise .What i am looking for now is doing an algorith to get the faults from a batch of images ,similar like the following depicted.

So I can detect that burnt like area in the farleft of the plate as the mate surface in the right, so far been trying with Image processing toolbox by doin some segmentation mask .

. Here I have a sample how every probe should be with few small defects (maybe lack of light of shooting angle makes it look darker as it is)

So ,how can I start it off ? I mean I could substract the faulty probe with the perfect one so i get the faults or i could just make a deep learning algorithm? But to do so , what should be the inputs of the algorithm? besides I could require lots of data? I made not much probes cause each take some time by myself. Or just applying segmentation inside the faulty probe with textures and color segmentation after applying ROI edging?

Thanks for the advice .

Attached I have more examples of probes