What's new
What's new

Source of good quality camera for inspection line

Neil

Hot Rolled
Joined
Jan 1, 2005
Location
Lockport, NY
We need a good industrial camera for inspecting something that can resolve 50 microns or so through a C mount lens. Who are the top 3 names in this kind of thing? It looks like Canon gets involved in this kind of thing (although Nikon appears more involved in microscopy). Any thoughts or recommendations? Are there any reputable forums on this subject? I did a search here an not much came up. Not sure if I need "machine vision" as we are doing visual inspection but magnifying it and displaying it on a large video screen. We would use a macro lens, possibly C mount if the quality is good enough for the sensor used.
 
Edmunds optics.
Get the catalog and tech info.

You say 50 microns but do not say size of the part to be imaged. Is this .002 thou a pass fail limit? if so do you need a lot inside at the image for good/bad judgment?
This is can get complicated, is your display screen up to the task? Frame rates? Need telecentric or somewhat?
I'll ask these things of people and 98% of the time I get a blank stare as if I'm am talking Martian.
Be aware that color cameras and black and white cameras are very different in actual resolution vs pixel count.
Welcome to a rabbit hole. I am not aware of any good forums on this so if you all out there have one please post.
Bob
 
Carbide Bob:
I am an amateur photographer with a degree from Google U. so I know a little.Also I am a mechanical engineer with a bit of machining knowlege. Also, we used to shoot high speed video out at the WSL at China Lake so here are my assumptions, please correct me if I am wrong:
1. The application is looking for contamination in a volume of space.
2. We have defined the needed depth of field and width of view and used a nifty calculator for depth of field and married it to smallest circle of confusion that will enable us to use up our inspection volume. The entire inspection volume is to be checked. See this link: Online Depth of Field Calculator. Our circle of confusion is down around 15 microns (theoretical of course, the real world is full of toleranced lenses, cameras, screens, etc)
3. When shooting high speed video(flashback to the Weapons Survivabiltiy Laboratory at China Lake) the application engineer told me if I gave up color imaging and went to black and white I could get more resolution. Is that generally true?
4. We are NOT recording video, the inspector merely views what they are inspecting.
5. I am being deliberately cagey about the "inspection volume" as I don't want people to know what technology we are developing. Lets just say its smaller than a breadbox but bigger than a TV remote. I don't even want to say what industry this is for, its that sensitive.
 
#3
Camera photosites only count photons, they know nothing about color.
To make color you but a Bayer filter in front of the chip. The input from the set (a R site, a B site, and a G site) give a colored pixel. (Google this)
In fact you have just divided the camera resolution by 3. Software in the camera then interpolates to out the specked resolution number but a lot of these pixels are made up information.
Also alignment of this "color mask" to the chip sensor is never perfect so some loss there.

Almost all cameras except the ultra big buck scientific stuff will have some dead pixels. Again software in the camera guesses at what should be output.
This means that if a feature is only one pixel in size there are places in the FOV where you will never see it or detect it.

On top of this there is a thing called "fill factor" for different sensor types. Unlike film not all of the sensor front is photosensitive.

A single feature, say a floating piece of dust needs to be at least 3x3 pixels in B/W and 7x7 in color to be seen reliably.

High resolution sensors need to pump a lot of data and the interface is a real speed limit. Real time viewing of a 14MP camera on USB3 is a problem. Higher speed interfaces go up in price fast.
Also monitor resolution in such a setup. In between the camera stream and the monitor output some scaling normally goes on. This can make small things disappear or grow.
If you do have the bandwidth and a 8K monitor how small a displayed feature can your human reliably see? How big is the pixel on the monitor becomes a question.

You are going to need to do some experimenting pushing hard to look at worse case.
This is where many fail. They want it to work so on the bench get something great only to have it not work so great 30-60% of the time in the field.
Try to make it fail, not to make it work.

Unsure the application but is stared at long enough there is the garbage man problem. A garbage man does not smell garbage anymore.
A continuous visual inspector does not see the 1 in 1000. If over and over the brain fills in or filters the information. That a whole another story and studies.

I do not know what you are doing or if any of this is worth posting here but my 2 cents worth as I do so like video inspection.
Yes I for sure know some sense and no sense here. A rabbit hole of senors, display, and humans viewing it.
It is a system and weakest link.
Bob
 
Dear Carbide Bob,
Thanks, this is eye opening. I know I have a lot to learn. I didn't know 8k was a thing for monitors OR cameras. Wow, just wow.
 
OK, I found some cameras. I'm looking at
1.Amscope
2.Thor Labs
3.Baseler
4. Pixelink
5. EdmundOptics (Edmund Scientific)
All are 1/3" or 1/2.5" cameras.
Edmund seems to hit the sweet spot with a 1/2.5" CMOS with 5 megapixels for 310 bucks. They sell Allied Vision cameras.
 
Update:
Looking at a 4k camera, direct to monitor via HDMI by Accu-Scope. Using Navitar lenses or Fujninon on demo.
 
If you want to automate the decision process, then Cognex or Keyence would be the first 2 places I would go. We use a considerable number of both in our manufacturing facility.
 
We are not automating at this point, we are just using a large 4k monitor to assist visual inspection by humans. We talked to Keyence and their camera alone is over $5k. We can get cameras of equivalent value for under $2k using Sony Chips.
However, thanks for mentioning Cognex, maybe they could come out here too.
 
Add light

Update:
Looking at a 4k camera, direct to monitor via HDMI by Accu-Scope. Using Navitar lenses or Fujninon on demo.

We make camera systems for some specific industrial inspection processes. Most systems have a resolution of 1080p because that is adequate for a direct display solution. Currently HDMI output 4k cameras are the most cost effective as you don't have a computer. They will work fine but may not last for years as they run hot. Also, distance from camera to display can be an issue with HDMI.

As far as lenses go it all depends on working distance and field of view. Do you ever want to change these parameters? Lighting (strength/direction/type) as related to the reflectivity of the material, geometry and distance are your next considerations. Entocentric lenses would be fine for what you are talking about but if you ever measure you might consider telecentric.

For our soldering station we purchased a 1080p chinese microscope camera a few years back for $500 which has an autofocus feature and put on a Kowa macroscopic lens. The image is good enough, but the on screen display can be a bit annoying. Rather than an autofocus, you may consider a smaller iris size and adding more light for greater depth of field. In general for this application a smaller sensor size would allow for smaller lenses, which can be helpful.
 
Last edited:
I should have also said you can run some of the machine vision cameras via linux .. you'd need something better than a raspberry pi to run 4k
 
I'll second Cognex. At my last job we used their cameras even when we didn't use their software.

Not knowing anything about your application, you might want to consider partial automation. That is, have the computer look for your contamination AND have an operator look for it. As Bob pointed out, people make mistakes looking for things in pictures. There was some research a while back on automating reading mammograms. What they found was that the computer had more false negatives than a radiologist, but a computer and a radiologist were far better than two radiologists. The human and the computer made different mistakes, so working together they caught more tumors.

In my (somewhat limited) experience, lighting is going to be a bigger factor than the camera in detecting contamination. Talk with the folks at Cognex. If your contaminant is constant, you may be able to choose your light color to maximize contrast and detectability.
 
Carbide Bob:
I am an amateur photographer with a degree from Google U. so I know a little.Also I am a mechanical engineer with a bit of machining knowlege. Also, we used to shoot high speed video out at the WSL at China Lake so here are my assumptions, please correct me if I am wrong:
1. The application is looking for contamination in a volume of space.
2. We have defined the needed depth of field and width of view and used a nifty calculator for depth of field and married it to smallest circle of confusion that will enable us to use up our inspection volume. The entire inspection volume is to be checked. See this link: Online Depth of Field Calculator. Our circle of confusion is down around 15 microns (theoretical of course, the real world is full of toleranced lenses, cameras, screens, etc)
3. When shooting high speed video(flashback to the Weapons Survivabiltiy Laboratory at China Lake) the application engineer told me if I gave up color imaging and went to black and white I could get more resolution. Is that generally true?
4. We are NOT recording video, the inspector merely views what they are inspecting.
5. I am being deliberately cagey about the "inspection volume" as I don't want people to know what technology we are developing. Lets just say its smaller than a breadbox but bigger than a TV remote. I don't even want to say what industry this is for, its that sensitive.

LOL. I assume your work is done under a public solicitation, which means that everything in that solicitation is public and unclassified. If you look at your project documentation, it will include a classification guide that tells exactly what is UNC and what is classified. If you do not have a classification guide, then it means EVERYTHING you are doing is unclassified. Also, as you are probably aware, if you are making a classified part, that has to be done in a secure facility, which requires an FCL and everything that comes with that (safes, coded entry, FSO oversight etc etc). Also, obviously anyone doing classified work or entering an FCL has to have a security clearance.

The only kind of work that has secret job descriptions is black budget work for the NSA, NRO, CIA or NGIA, and is highly unlikely you are doing that, because that would require a SCIF and industrial SCIFs are usually only found at very large military contractors like Ratheon, Lockheed Martin, General Dynamics, etc).
 
Hikvision would be the one to use to end your problems with whatever you are manufacturing. They offer a discreet service where they upload videos of your manufacturing process to their government servers and study in detail every part of what you make. Within a month you will see the product you make is now a cheap product cheaply made by the Chinese. Within a year it will not be profitable to run the machine any longer thus eliminating the need to monitor quality with the camera.
 
If the information and talent is you, and some classified place wants you there, you do not need a security clearance.
 








 
Back
Top