Kuka deceft detection

KUKA vision was and is Cognex based... in the past (KRC2) it was using InSight. current crop of controllers (KRC4) is using VisionPro...

do detect something (anything) you need to be able to differentiate between "something is there" and "something is not there". vision system use optics to see things. adjusting or changing optics can make it possible to look at things of different size (scale).

using image to recognize defects can be very effective but this is still depending on many things.

whatever you want to do, need to start with defining the problem. this means listing all requirements with some numbers. you can use lens to see panoramic view of a mountain. that same lens setup would be completely useless to view hair line fractures. and if you replace it with lens that can see hair line fractures, you can forget about seeing panoramic images unless you spend crazy amount of time, storage and processing power to combine zillions of images into one.

if your process cycle time is 5.8 seconds but vision takes 10 seconds to process image... that will not really work.

you can attach anything to anything. there are screws, glues and gorilla tape that will handle pretty much anything. i suppose it is possible to connect things using cables and wires too. but if you want some data exchange to take place, there would need to be compatible interfaces between two systems. the things is there is no clue what systems you have in mind. KUKA and COGNEX are not products. they are companies with countless products in their portfolios. so if you are going to ask for help, you need to be a lot more specific.

a good way to narrow things down is to pick model (robot system, vision system) read their specs and check if they can do the job in required time and have compatible interfaces. for example you can get InSight cameras with fieldbus option...

if your vision system has Profibus option, check if you can also have Profibus on the robot side.

if your vision system is using raw TCP socket messages, your robot need to be able to do the same and match the message format.

word of warning:

if there is a canned system from robot manufacturer, interfacing is super simple (plug and play).

if that is not the case but you are using fieldbus, things are usually very straight forward. yes there may be stumbling blocks but once you have cleared those and got connection, things are normally smooth sailing.

but if you are looking into implementing some protocol yourself to match whatever your vision system may be using, this is usually very involved. you will spend more time on communication than on actual task of recognizing things with your vision system.

Quote from miraziz

Which camera should we use and can we detect scratch,welding defects,and other defects?

That... depends on so many factors unique to your situation and application, I wouldn't even know where to start. It would depend entirely on how these defects stand out against the background. It's also easy to set such a vision system up such that it would "fail" many good parts because of a shadow, or a bit of dust.

For quality inspection, these days often Machine-Learning vision systems are used, but they require a large number of sample images of both Good and Bad parts to train them on what to pass/fail. Cognex sells some of these, like the InSight 2800 series.

Quote from miraziz

Can we connect any cognex camera to kuka or no

Yes, but... it depends on what you mean by "connect".

You can connect any Cognex to a KUKA by using Ethernet/IP, or ProfiNet, or even TCP/IP, depending on what options you purchase with your robot. But at this level, you have to write your own interfaces from scratch.

If you buy KUKA VisionTech, this gives you an "easy" interface between the robot and camera, but limits your detail control, and constrains you to use only certain cameras and certain options.

Quote from miraziz

Moreover,Can someone tell me whole process step by step

Can you tell me how to build an entire automobile, step by step? That question is so vague is cannot be answered.

You seriously need to consult a vision expert. Before even choosing which camera and software to use, you need to do a deep analysis of your parts, the working conditions, how much variability "good" parts will have, what your pass/fail criteria are.

One of the most vital parts of planning any machine vision application is lighting. This depends on ambient light (sunlight through windows kills vision systems' ability to do anything), the colors and brightness of the material being tested (direct light vs oblique lighting, vs shadows), filtering for colors, long vs short exposure times... the list is very long.

The materials for the Certified Vision Professional (Basic) are free to use, and give a good grounding in the various issues you need to be aware of: https://www.automate.org/a3-content/cvp-basic-certification

Taking the test costs money, but simply reviewing the videos and the PowerPoints will help you understand how deep the water is.

Cognex offers good consulting services for picking out the correct product for your application -- I would approach them. But you will need to provide detailed information, and good example images of what the vision system is expected to test. Cognex is generally quite good at this, but they may well tell you that they would need to do an in-depth study, with test samples, before being able to give you a reliable answer.

a good idea is to have ready samples (or pictures) of defects then ask manufacturers of vision system to recommend system that would work and can talk to KUKA robots. one of the things with vision systems is that they are only as good a image of the problem and image may not be able to reveal the defect like cracks or fractures even when surface is homogeneous.

once you start looking at welds etc. things get even more complicated. the real question is what type of defects you are trying to detect.

if this is a functional test, you should pressurize radiator and do a leak test.

if this is cosmetic/appearance only (see blemishes on the surface) or feature inspection (presence/absence of something) then vision should be good match.

if this is tolerances check, 2D profile laser scanner would be suitable product.

it all starts with what you are trying to detect, how fast, what the variations are, what the environment is etc. frankly connectivity would probably be one of the least of concerns - after all there are plenty of gateways that can convert pretty much anything to anything. see HMS or Anybus product line for details. i would be much more interested in getting reliable test result that meets clients expectations.