Speed is the key thing!
Typically you need specialist fast equipment to be sure of capturing all the detail.
Many fast moving processes such as paper or film manufacture have traditionally used what are nown as linescan cameras, they effectively only have a single row of pixels which are scanned repetitively and very, very quickly. Fast processors then build useful 'views' from this data.
CCTV and IP cameras can be used for other (less fast) process monitoring e.g. why are pallets occasionally toppling over. I read of a good example caught on CCTV in a bakery, very occasionally a batch of biscuits would catch-up as they exited the oven and cause a big mess and loss of production. It was very infrequent and hard to diagnose. Eventually, after installing constantly-recorded CCTV they spotted that when a kink in the drive chain was coincident with a small defect on the sprocket and only then when it was also coincident with the leading edge of the batch of biscuits - they got the problem.
One other thing worth mentioning - lighting!
Correct lighting makes a heck of a difference.
Typically, for fast processes, you need loads of light, for fast shutter speeds - I have applied 1,000Watt TH lamps directly over inspected processes.
Also, light form behind paper or film allows you to see through it - ideal for watermarks or similar.
Light perpendicular to the surface allows you to see surface defects, spots, dirt, etc.
Low-incident light e.g. across the surface allows you to see wrinkles and creases.
Similar features and benefits will apply to other processes.
If you need the additional information such as where the detected fault lies, you will need corresponding additional input data e.g. an encoder on a suitable process roll will provide you with distance reference information (speed X circumference = m/sec). Lateral position info e.g. across the web would normally be limited to knowing the cameras lateral position i.e. according to the detail of resolution that you require, any single camera will only resolve down to a calculable small (pixels per mm) size. Therefore, you're likely to need multiple cameras across the web, and will fairly easily from your overall control system be able to learn that say camera three spotted a defect after 300m. Linescan cameras would probably be able to resolve down to a fairly exact pixel position across the web.
Carrying that information forward to the next stage in the process and knowing what you can then do with it is another trick. One of the simplest solutions is to have the detection system drive a mark-making system, which might actually make a very bold ink mark at the web edge. At a later processing stage that bold ink mark can be easily spotted and the marked product removed or rejected.
All fairly specialist stuff, not really something that the products in our webshop are intended for, but I hope it helps a little ....
