Help on identifying the correct statistical tool - Defects vs. Total products

[encee98]

Greetings!

The company I am assisting used a column chart to show the number of defects vs the total number of products produced per month. They used one color for the defectives and another color to indicate the total production quantity. There were three months in the past year where they encountered defectives, to wit:
April: 1 out of 2,013
June: 3 out of 2,700
September: 1 out of 2,100
during the surveillance audit, the auditor commented that the chart they were using was not suitable for the data. The auditor said they have to use a tool that can show the defective units vs the production quantity. Aside from increasing the number of defects so that it will be reflected in the chart, is there any other tool that we could have used?

Best regards!

 

[Govind]

encee,
My first observation is you are interchangeably using the terms "defect" and "defectives".
By definition, they are different.
Presence of one or more "defects" in a product will make the product "defective".
Example a product could have a scratch, a dent, and a stain mark. If these scratch, dent and stain are beyond the acceptable limit, we call these defects. There are 3 defects. Because of these defects, we call the product defective.
There are Statistical Process Control Charts are for defects and defectives.
Another point: Production Quantity, Total Quantity. this is not clear. This can be subjective. Many organization use "Lot Quantity" for this purpose. Again with equal sample size and Unequal sample size, defects and defectives, you can choose between p,np,c,u charts.
https://www.moresteam.com/toolbox/t402.cfm
It is good to know that your auditor did not just look for a graphical representation and nod the head. He has asked you to do some meaningful analysis. Thanks to ISO2K.
Regards,
Govind.

 

[Bill Pflanz]

Whether its defects or defectives, if the other nine months were zero than it really doesn't make any difference what chart is used. Unless I am not understanding your data, there are so few defects that no chart will do more than confirm there are no real problems.

I guess if the defect is in something critical, for example a medical device, that results in a death then you would want to prevent the defect. In this case detection to prevent its use is important but of no value to graph.

Bill Pflanz

 

[cyberjyothi]

Hi,
pl.give the charts with data(using Excel or what ever u use for drawn the graphs) ,what u r shown to auditors.Based on that i may suggest the relevent chart for u r data.

Thanks & regards,
Jyothiswar

 

[encee98]

Thank you for the replies.

Govind, I meant defectives. Thank you for taking note. The company is producing medical grade oxygen and is compliant not certified to ISO 9002:1994. In the Philippines, companies producing medical grade oxygen are required to apply for a product quality certification license. One of the requirements for the license is compliance to the requirements of ISO 9002:1994.

Bill, the defect was on loose seat valve. The critical quality requirements for medical grade oxygen production is purity (should be not below 99%) and presence of trace impurities. During production, if the process sample results show that the purity is still below 99%, they do not collect the gas. And they do hourly sampling and testing.The company had not encountered any problems in this area.

Actually, my point was, since the company is still using the 1994 series, and they have had no problems with product quality, they could declare that they do not need to employ statistical techniques to establish, control, and verify process capability; but the auditors told them that they have to. The auditor wanted them to adopt a control chart, but what would the use of a lower control limit be when they could always stop the process before they plot the data in the chart?

The company is working towards making the transition to 9k2k.

Jyothiswar, I'll try to get a copy of the actual data.

 

[Bill Pflanz]

Bill, the defect was on loose seat valve. The critical quality requirements for medical grade oxygen production is purity (should be not below 99%) and presence of trace impurities. During production, if the process sample results show that the purity is still below 99%, they do not collect the gas. And they do hourly sampling and testing.The company had not encountered any problems in this area.

I came from the chemical industry so I can relate to your problem with the auditor. The standards do not require that statistical techniques be used for final quality but "to establish, control, and verify process capability".

Since you cannot allow out of spec product the control chart of final product may show variation but would never show out of spec or poor quality. I could see some benefit though in control charting your process to determine process shifts that could lead to off spec or higher costs. If you can use the gas for non-medical use at a lower specification and wait for the higher quality when available at no additional cost then even that may not mean much.

Maybe you can use some statistical techniques within your process and that will satisfy the auditor. I do not see any value in control charting finished product specs based on what information that you have provided.

Bill Pflanz

 

[Steve Prevette]

One thing you may want to consider at these low rates is to look at the number of items between defectives (sort of a MBTF) instead of counts or counts per unit. Dr. Wheeler advances this idea in "Understanding Variation", and I also have a writeup based upon Dr. Wheeler's work at https://www.hanford.gov/safety/vpp/lowrate.htm

 

[J Oliphant]

related experience

Since you cannot allow out of spec product the control chart of final product may show variation but would never show out of spec or poor quality. I could see some benefit though in control charting your process to determine process shifts that could lead to off spec or higher costs.
Maybe you can use some statistical techniques within your process and that will satisfy the auditor. I do not see any value in control charting finished product specs based on what information that you have provided.

Bill Pflanz

I have a related experience. we have a purity test where we routinely measure the presence of certain chemicals to make ensure they are below certain levels (regulatory requirement). the regulatory requirement says if it can be proved that the contaminatminants cannot exist in the product stream, and we can prove it, we can be exempt from the testing. in at least 8 years, and 673 determinations we have never even detected the contaminants -and our process would assurely vaporize them/remove them from the product stream. my boss wants statistical analysis, but it is rather hard.

we have 673 0's! what is our pk, our ppk- he asks? Very high, I say with a smile. well, we once went down the road, of calculating the statistics, if the next sample we had a detectable result. that would roughly mean that a presumed normal curve- with a frequency of 1/673 being barely detectable. one could guess on the the median (I supose-after all it is between detectability and zero) and set up a spec limit to get the ppk, or the theoritical probability of exceeding the spec. it was roughly 1:10 million or many thousands of years of constant production.

ok, THAT was crazy math. there were all kind of crazy prob. untrue assumptions. But it ACTUALLY bothered my boss that they was the probability of exceeding spec. (when we haven't in recorded history ever gotten an detectable result- 1/10 of the regulatory limit).

it proves, that with statistics ANYTHING is possible. and that if the event is rare enough, statistical capability just won't prove that the impossible CAN't happen. and IS IMHO, a dilemna, just when statistics would be most useful - it isn't.
It seems, you can only calculate the probabilies, when things routinely happen, not when they never do...

 

[encee98]

Yipes! If I were Alice in Wonderland, I'd say curioser and curioser. Anyhow, we were given reprieve because the Auditor said they were just preparing us for the 9k2k transition (hence the extra focus on statistical technique) and that we should do better by that time.

So back to the drawing board to draw a better chart...

for the inputs...

 

[KarenD]

Govind,

Thanks for the .morestream link! We had a customer SQE on site today and we found out together that our powerwash system is being monitored, but that the operator is not using his control charts accurately. This link is a great place for me to refer him to so that he can refresh the basics of SPC.

Cheers,
Karen