8 replies to this topic

[jfox]

Just wondering what people tend to do who use an X-Ray in their processing plant. 

 

We run tomato products as well as Sauerkraut. As we are setting our system up (new to the game) we are noticing many false positives. Usually around the edges or maybe a pixel that is dark. We run a test jar thru with foreign material taped to it and the x-ray picks that up. But my question is how does everyone handle a positive for foreign material etc?

 

Currently we re-run the product thru again and if it is positive again we hold for further evaluation. If it does not come back positive we allow it to pass thru. 

 

Any help or guidance would be great! 

[GMO]

It depends on who you're supplying on whether that's an issue.  Many customers wouldn't be happy as no x-ray or metal detector is 100% fool proof.

 

I suggest you get the supplier back in.  Ultimately there is a product signal from all products whether you're metal detecting or x-raying them.  It looks as though the product may not be set up correctly in the x-ray machine so what is "normal" is being detected as a defect.

 

You don't say what test cards you're using and for what contaminants.  IME, x-rays are useful for detecting metal but are limited in their capacity to detect anything else.  It depends a lot on the density of your product vs. the density of the contaminant you want to detect.  Plastics particularly can be very difficult to detect in many products (despite manufacturers claims). 

[jfox]

We dont have test cards. Currently we run a jar thru the machine with the contaminate attached. We are most concerned with metal but it is also able to detect glass. The company who services our X-ray machine has helped us set up our products and given us guidance. 

 

My question is, until your product has been "learned" fully by the machine and if it is ejecting many containers for things that do not look like anything (maybe the jar slipped and the edge was detected improperly or it detected a part of the product as a possible contaminate. Our product has onion that sometimes shows up as white while glass shows up darker.) what do you do to verify something was found? Do you run it again? 

 

Clearly if there is something that looks suspicious we simply segregate that product and examine later. But if it looks normal do you just run it again or simply "add to good container learn"? I just wanted to see who else maybe deals with many false positives especially during product learn and setup. 

 

Thanks again! 

[GMO]

We dont have test cards. Currently we run a jar thru the machine with the contaminate attached. We are most concerned with metal but it is also able to detect glass.

 

Ok, I don't get this if I'm honest.  To run an x-ray you should have proper, validated, certificated and calibrated test cards which are the equivalent of your metal detector test sticks for metal detection.  I don't see how you can just stick on a potential contaminant if I'm understanding this correctly?  This is the kind of thing I'm talking about.  http://www.detectame...-test-cards.php

 

Also how are you ever going to be able to detect glass if the product is packed in glass?  I don't understand that?  You are asking the machine to ignore glass and look for glass at the same time.  In my experience this may only be possible if the types of glass you are looking for are different or you have some very careful set up basically using the x-ray as a vision system "ignore this part of the picture but look here" if you get what I mean?  But the latter is very tricky.  If you're not able to competently check for glass you may be muddying the water with metal.  You're getting all these false rejects probably because you're trying to reject glass contamination which is difficult to set up as I've explained but could be risking letting out a product with metal in.  Once you're into processing rejects from an x-ray or metal detector it needs VERY careful control.  Fundamentally I would question if your x-ray really can do what your supplier says it can, i.e. detect glass.  In the meantime you're running a line saying it's ok "it's just learning our product" to which I feel a bit panicky! 

 

I don't think you should be running a line while a product is being learned.  The learning process should be part of set up / validation etc.  At the moment you don't know whether your x-ray is effective and you're not trusting it.  Not a sensible nor duly diligent place to be IMO.

 

At best, and not all retailers like this, you should pass any supposed "false rejects" back through your x-ray 3 times and only pack them if they pass all three times but I'd be very wary of that.

[jfox]

Thanks for your input. 

 

I dont know how it detects glass but it does (via the attaching glass to the backside of a jar method). X-rays arent perfect and I get that.

 

As far as the learning goes, it requires over 2200 items be scanned before it considers the product fully learned. This isnt possible for us unless we are running product.

 

That last bit was really my question. How do you verify if you think the x-ray ejected a false positive? Is there a critical limit people set? 

 

The way I understand it an x-ray is just a tool and you have to use your eyes to verify. But if you dont see anything and you think it may just be the edge of the container causing a problem what do you do if it rejects 20? I would fix the region of interest and rescan. I like your idea of 3 times and only passing if it passes all 3.

 

I really do appreciate the input! 

[GMO]

I would say if it really takes 2200 items to learn your product you're not using it properly until you've passed it 2200 times.  Also I really think you have a sensitivity problem if you're getting loads of false rejects which probably means that yes you can detect glass but it's so close to the normal background image it can't detect it reliably, hence your false rejects.

[Ryan M.]

Sounds like you need to get the equipment vendor involved.  Do you have the xray calibrated to each separate product and container type/size?  This is something that needs to be done.

[glen_oxo]

Hi - sorry to come to this late.

 

Glass in glass detection is definitely possible with X-ray as normally the system expects to see the density associated with the side walls of the jar. Any glass inside the jar will add to the overall density and therefore be detectable (subject to it being a sufficient size).

 

From your description I think I have some familiarity with the machine you describe as the 'add to learn' is quite specific and not the way that many systems operate. My suggestion would be that you should only add anything to the learn if you can be 100% sure it is actually a false positive. Otherwise you risking adding genuine foreign body signals to the learn which will then in turn be ignored by the system.

 

Hope this helps.

[Charles.C]

Hi jfox,

 

No x-ray experience but, as per the previous posts, if you have significant numbers of false positives or false negatives (!!), it presumably means you have a set-up problem of "some"  kind. >> Supplier.

 

You can find published/recommended "repeat mini-plans" to (finally) decide +/- for MDs in the literature (similar to that mentioned previously) but these schemes will not be defensible for high rates of error.

 

My experience with frozen glazed blocks of seafood/MDs has some similarity regarding false rejections. The only practical solution was to adjust the sensitivity but this obviously has its own limitations/risks and requires validation.