100% Sort Verification - Statistical Method for Evaluating Suspect Material

G

Gary E MacLean

Good morning all;

I would like to request some assistance.

We bottle fruit into glass jars with metal closures. They rapidly flow down the filling line at the rate of about 15,000 jars per hour. We have an in-line metal detector that scans every jar prior to having the metal cap put on. We verify the detector with known rejects once every one hour.

Typically we can expect about 5 legitimate rejects per month or about 5 jars every 2.5 million jars. On occassion the metal detector vibrates out of setting. We can fix that, that is not the problem. But, when we go to check our known rejects they fail. By the time we check we have produced 15,000 jars with closed lids and vacuum established.

We segregate all materials then try to figure out what to do. Our FMEA, or scheduled process, says we must rerun all product. That is we must open every jar, tossd the lid, dump the fruit into a hopper for refill and either wash or trash the jars.

What I am hoping is that there is a statistical method for evaluating, sampling the suspect material to make a viable risk analysis and resultant decision that may save the rework and material loss costs.

Can anyone help?

Rate = 15,000 / hour
Fail = 5 / 2,500,000
Check = 100%

Thank you

Gary E MacLean
 

Jim Wynne

Leader
Admin
Good morning all;

I would like to request some assistance.

We bottle fruit into glass jars with metal closures. They rapidly flow down the filling line at the rate of about 15,000 jars per hour. We have an in-line metal detector that scans every jar prior to having the metal cap put on. We verify the detector with known rejects once every one hour.

Typically we can expect about 5 legitimate rejects per month or about 5 jars every 2.5 million jars. On occassion the metal detector vibrates out of setting. We can fix that, that is not the problem. But, when we go to check our known rejects they fail. By the time we check we have produced 15,000 jars with closed lids and vacuum established.

We segregate all materials then try to figure out what to do. Our FMEA, or scheduled process, says we must rerun all product. That is we must open every jar, tossd the lid, dump the fruit into a hopper for refill and either wash or trash the jars.

What I am hoping is that there is a statistical method for evaluating, sampling the suspect material to make a viable risk analysis and resultant decision that may save the rework and material loss costs.

Can anyone help?

Rate = 15,000 / hour
Fail = 5 / 2,500,000
Check = 100%

Thank you

Gary E MacLean


Hi Gary,

I'm not quite understanding the process. You say that the metal detector checks the jars before the lid is put on. Does this mean that the purpose of the metal detector is to find metal in the product itself? You also say, "... when we go to check our known rejects they fail." Do you mean that the "known rejects" are false positives (no metal present), or that you verify that there is indeed a failure involved?
 

Bev D

Heretical Statistician
Leader
Super Moderator
Good clarifying questions from Jim as always. However, if you are really looking for metal in the jars and you are targeting for NO metal in the jars (I hope) then my recommendation would be to continue to do as you have done. refill and retest. the failure rate is far too low requires 100% testing anyway for almost any 'lot' size.
 
G

Gary E MacLean

Hi Jim

Thanks for the rapid response

The process is;
Empty jars are loaded onto line
Empty jars are inverted and washed on line
Empty jars are turned upright and passed through a product filler
Product is put into jars
Jars with product then continue to a liquid fill station
Liquid is put into jars
Jars are passed through a wrap around metal detector
Purpose of detector is to find any foriegn / vagrant metal inside jars
Checked jar with product and liquid inside is passed through capper
Jar is evacuated of gasses and capper applied
Vacuum is formed inside jar

Once an hour jars with product, syrup and known test metal inside are passed through detector
Purpose of this activity is to verify the detector is still sensing metal

So, yes, the purpose of the metal detector is to find metal in the product.
The known rejects are product jars with metal present - the detector should be able to see them.

My problem is that on one of these hourly checks our detector failed to see the test metal in the product jars. It was discovered the detector had been inadvertently adjusted to an improper sensitivity. Now we have 15,000 jars of product that were accepted by the metal detector but we do not know if the detector was really working or not.

We typically have 5 to 6 legitimate metal discoveries for every 2.5 million jars. I would like to know if I can somehow statistically justify either not rejecting these jars of product or sampling a statistically valid number of jars to show that metal is really not present. There, of course, is not supposed to be any metal whatsoever in our product.

Thanks for your help.

Gary E MacLean
 
G

Gary E MacLean

To Bev D

Hi Bev

Yes, we are really looking for metal in the jars. And, yes we are trargeting for NO metal in the jars. I am also looking for some sort ofr risk analysis and / or cost justification. To refill and retest requires us to soak all labels off because the jars have to go through the washer again. Then we dump all product to start back over and we also lose all metal caps. Once they have been applied they are not acceptable for new product use.

The existing failure rate is 5 / 2.5 million. I don't even know the PPM for that but it seems as though we could sample a statistically valid number of jars and make our decision from that sampling.

Can this work?
 

Bev D

Heretical Statistician
Leader
Super Moderator
To Bev D

Hi Bev

Yes, we are really looking for metal in the jars. And, yes we are trargeting for NO metal in the jars. I am also looking for some sort ofr risk analysis and / or cost justification. To refill and retest requires us to soak all labels off because the jars have to go through the washer again. Then we dump all product to start back over and we also lose all metal caps. Once they have been applied they are not acceptable for new product use.

The existing failure rate is 5 / 2.5 million. I don't even know the PPM for that but it seems as though we could sample a statistically valid number of jars and make our decision from that sampling.

Can this work?


nope. statistics wont' help you. at a 5 in 2.5 million failure rate your ppm = 2 or 0.000002. using the Poisson distribution and a confidence level of only 99% your sample size is: 2,302,586
 

Bev D

Heretical Statistician
Leader
Super Moderator
my larger view suggestion is that you develop a process for verifying the effectiveness of your tester after any changes with you known bad units prior to making sellabel product. you will pay a lot for this lesson, it would be a shame to not learn something from it.
 

Jim Wynne

Leader
Admin
my larger view suggestion is that you develop a process for verifying the effectiveness of your tester after any changes with you known bad units prior to making sellabel product. you will pay a lot for this lesson, it would be a shame to not learn something from it.

Good advice; you might also want to consider sending in the control jars at a frequency more often than once an hour, which would reduce the amount of sorting necessary if the metal detector fails for any reason.
 
G

Gary E MacLean

Well Bev

I only have 15,000 at risk jars; there were only 15,000 jars that got checked through the questionable metal detector. My 2.5 million number is how many it takes to create 5 bad ones.

Does that change your prognosis at allo?

Gary E MacLean
 
S

somerqc

As Bev is using the 2.5 million number to determine PPM, I doubt the numbers would change.

I think what everyone is telling you is that you are better off to empty everything and refill the 15,000 jars than to risk a recall and the associated costs.

I am assuming you are registered to some kind of quality system. Most systems outline having control of any calibrated equipment. It strikes me that you may not have the necessary controls required to ensure that the machine maintains the necessary calibration status. Just my :2cents: You may want to investigate how the machine "suddenly" does "see" metal anymore.

John
 
Top Bottom