View Full Version : Crack detection

06-24-2008, 12:47 PM
I couldn't think of any other forum to put this, I am trying to create a simple system to check CD cases for cracks. I've considered ultrasonic crack detection, but the systems available for thousands of dollars. There was a theory of heating the surface and the cracks would return different colors in a thermal camera, but thermal cameras are thousands of dollars. I also tried black and white scanning with different brightness and contrast settings, while it shows the cracks sometimes, its too inconsistent. Anyone have any thoughts or ideas on this? Thanks a lot!


06-24-2008, 07:58 PM
Interesting question - and welcome to the forum!

You might try inexpensive ultrasonic transducers (such as those in Trossenrobotics.com store), and I think I'd have a look at Infrared too. Maybe a large IR Illuminator with the camera pulled back far enough to see the entire surface. It should definitely show-up under IR some wavelengths of IR. And that brings us to a UV illuminator too. Longwave is reasonably safe and inexpensive, and should also show surface defects.

For any of these, have a look at RoboRealm (roborealm.com). It's an image processing and interaction system that would allow you to test and prototype these theories very quickly.

I'm assuming in the above (excepting ultrasound) that there's no label in the case, and the case should have at least some reasonable light transmission properties (ie. not a black DVD case out of PVC or styrene...) If that assumption isn't safe, we should rethink this...

Again - welcome to the TRC!

06-24-2008, 09:54 PM
Thanks for the reply, the CDs do have labels, and CDs inside, and most have the black insert you are talking about. Most if not all are sealed in shrink wrap. The shrink wrap might pose the biggest problem. I'll check out that website. Thanks again.


06-25-2008, 12:46 AM
Thank you for your patience in clarifying the challenge - that helps me visualize it tremendously!

Definitely the shrink-wrap will likely eliminate any optical methodology. Contact ultrasound is probably the direction to look. Whether hobby-grade transducers are going to return a readable pattern, and at what power and frequency, is something that will need to be experienced in the lab in order to make the call, imho.

Are you looking for a high-speed QA solution? If that's the case, it's probably going to be a "punt to the real-stuff" kind of answer...

06-25-2008, 09:18 AM
It would need to be faster than a human checking a CD, so I guess pretty fast. I read that website some, and it brings an interesting question to me. Why do we understand when a case is cracked, and how would you tell a computer to figure it out. It's amazing how such a simple task of identifying a crack in a plastic case is actually an extremely complex task.

06-25-2008, 12:44 PM
The shrinkwrap is the killer. Personally *I* am frequently disappointed when a jewelcase falls apart in my hands when I take the shrinkwrap off. I don't think a human has much better chance than a machine.

Without the shrinkwrap, optically recognizing imperfections in the plastic (under controlled lighting) seems pretty trivial to me. If it's not a smooth transparent surface with fairly equal light transmittance/absorbtion across the topology, it's bad. A human would, imho, recognize it in a very similar way.

I'm not sure that even ultrasound is going to work through the shrinkwrap because it changes the apparent topographic map of the surface.

I don't know that it's really how fast a human can recognize it, really. I know in many instances it's how many a human can recognize per hour. The first hour is iffy, the second hour is probably great, the eigth hour you have junk slipping through constantly. Humans have a short attention span and are very inconsistant. A machine that can recognize half as many in the first or second hour may be 20x better at the tenth hour.

In video, we have to change humans out every few hours, and limit the number of screens they have responsibility for, as well as overlapping for a double check. The machines don't have either of those limits. They do tend to have more false positives when properly configured, but that's a much rarer exception-case which is basically a voluntary double-check request.

Does that make sense?

06-25-2008, 01:31 PM
That makes perfect sense, so ultra sound is out of the question then, do you think heat signature is also out? I imagine the cracks would heat differently than the smooth surface, but then again the shrink wrap might mess that up too. Do you know of any cheap thermal cameras?

06-25-2008, 02:01 PM
I don't know of any inexpensive FLIR, really. There are diodes that can be optimized to detect certain heat signatures, but I don't think that'd apply here.

I'm not at all *certain* that ultrasound is out of the question, or even that optics can't be trained under ideal controlled lighting to catch the majority of imperfections. AI/NN (neural networking) can work some miracles for pattern recognition.

I think the key is going to be picking up a few devices; ultrasound, UV, IR, inexpensive camera, and then testing/training them under ideal conditions with real live wild examples of good vs imperfect samples.

I can't be of any help there. I have exactly zero commercially shrinkwrapped jewel cases, be they flawed or not.

Honestly, I'm just making suggestions and assumptions based on PRE - Pure Rectal Extraction - from prior experiences in machine learning. (Specifically designing/training neural networks to optimize video compression for various types of video compressed at n# of bitrates using x# of codecs)

Nothing I'm writing should be taken as gospel here - none of these assumptions can replace putting real live rats in the lab, they're just off-the-cuff suggestions for directions to contemplate.

06-25-2008, 02:51 PM
Could the cracked cases be detected by flexing? You could build a rig that would flex the cases with a known amount of force and measure the deflection. I would imagine that a cracked case would bend more than an intact one.

06-26-2008, 08:31 PM
Better idea metaform...just drop a heavy weight on each case. Then you know its cracked...oh wait...not the point. (just kidding metaform, the flex test might be just the solution. I'm not sure about resolution in roborealm. Might be the issue with the black and white scans. Have you thought about a macro lens? You could set up a scanner that would make several passes at hight contrast black and white. Might not be that quick, especially with the shrink wrap on.


06-26-2008, 08:36 PM
What about lazers? Could you scan from the side and point a lazer at the case? the lazer would refract differently when it passed through a crack, and perhaps you could use that.


06-26-2008, 09:08 PM
Ok...I just did a test. Set up a camera at about 30 degrees to the CD case. Passing a laser over the case caused a refraction when it hit a crack, the one dot became two when the laser passed over the crack. Having a bank of lasers, or making several passes with the camera dimmed to the point where it will only see the laser might work with some tuning.


06-27-2008, 12:46 AM
Ok...so I'm not sure this will work in a real situation, but I was able to find a way to cause a change in video data using a laser pointer. I turned the brightness down on the camera to the point where the only thing visible was the red (almost white) reflection from the laser pointer on the CD case. In bright light you can see a reflection from the laser pointer hitting the crack. This causes the light to spread out briefly. Looking for that spread you can possibly identify a crack in a CD.


If I have time I'll try to write some code using MAX/MSP that will detect the laser pointers reflection. The brightness has to be pretty right on. Too dark, and you can't see the reflection...too bright, and if there is white in the CD label you see that.

Hope this helps.


06-27-2008, 01:35 AM
What happens when it hits a crease in the shrinkwrap?

Betcha it looks just like a crack. :)

06-27-2008, 09:05 AM
I actually don't have a shrink wrapped CD case with a crack in it. That said...isn't that the point of shrink wrap? It's shrunk so it shouldn't have wrinkles...I haven't bought a CD in years so I don't really have a way of telling. One thing I have noticed is that a crack under full brightness causes a flash at the CD cases edge. Shrink wrap may just exacerbate the problem though. I live on an island...finding a shrunk wrapped CD takes getting on a boat...or finding someone who is running a counterfeit operation...


06-27-2008, 11:33 AM
Unfortunately, there are a lot of imperfections in the shrinkwrap. Small depth and sharp corners don't make for a good wrapping job.

06-27-2008, 11:59 AM
darn...foiled again.


06-27-2008, 12:05 PM
"Foiled"? :D

You still might be on to a technique that will work, I think it just needs to be drilled down a bit with real-world samples.