Hi Everyone, thank you so much for your responses. Please find my response below in red.
60sec * 4000 parts per week * 1/3600 hour/sec = 67 hours of error free operation.
budget of $15000 is just laughable. it may at best cover cell guarding. but you still need to get robot ($30k+) and 3D vision ($70k+) and that is just begining and only some of the hardware. then comes labour (design, assembly, programming, testing etc). so robotic approach is just not going to cut it. unless your budget is at least two orders of magnitute larger.
15,000 is just the materials cost. That's why a bin picking robotic arm is our preferred approach since it would reduce the complexity of the other hardware.
Is there some good place I can do a price + feature comparison of the available options for robotic arms?
so to stay within budget you will need to forget about industrial automation and think radically different. check old episodes of junkyard wars and see what you can scavenge for free or for few bucks.
Okay, how would you suggest the correctly identified parts since there would be around 200 such categories (faulty and working for each of the 100 categories)
for example using openCV (free), second hand PC (can be cheap if not free) and USB camera ($50) instead of industrial 2D vision (single camera $10k).
Yes that is what we plan to do. Intel Realsense / Kinect 3D stereo cams cost around $200. But would they be accurate enough?
We can even hack our own stereo vision module if there's no good alternative in the market.
Then there's the problem of 4000 different kinds of fasteners -- most vision systems cannot retain memory for more than a dozen or two shapes at any given time, at best. So expecting to present a completely random mix of 4000 different fasteners to the vision system and expect it to identify and pick out each one is... really pushing the technology envelope. Not to mention that the vision system would have to be taught, manually, each and every fastener, and then taught the pass/fail criteria for each fastener.
We plan to have our own vision & processing module.
Current idea to first bin pick an object, move it to an inspection station, classify it correctly and then store it in the correct category.
My suggestion would be use some kind of shaker feeder to get one from the box, use a conveyor belt to move it to checking area, process it by camera and make some simple pneumatic sorting station with 2-3 lanes that you could make the sort easy. That should be well within budget. For final solution you need a very good scaning station that can recognize hundreds of different types and if they're in good condition, and you can either make a sorter yourself or just use a robot to drop it in a box, but I would avoid bin picking and just use some feeding mechanism to get to the station
The modules that you mentioned namely - shaker feeder, conveyor belt, pneumatic station, feeding mechanism are available for purchase separately?
Or would we need to build them using even more basic parts?