Process Qualification Sample size
- Thread starter Candi1024
- Start date
S
Scott A. Bednar
Hi Candi,
Here is what I use when performing process qualifications. Sorry for the copy/paste. Hope is is readable. Shoot me an email if you need some clarification.
Best,
Scott
Sample size determination for process qualification and validations.
Confidence level and reliability % will be chosen based on risk analysis.
The following confidence and reliability levels should be used corresponding to the severity ranking from the risk document.
*Not many on the forum will agree with the C/R table below; too low or too high. The table that I am sharing is from a specific project at a former employer (medical device).
Severity (Confidence/Reliability)
CA 95/95
CR 90/95
M 90/90
N 80/90
Sample size for attribute data
Required Sample Size
Confidence Level
Reliability 80% 90% 95% 99% 99.9%
80% 8 11 13 21 31
85% 10 14 18 28 42
90% 16 22 28 44 66
95% 31 45 58 90 135
99% 160 229 298 458 687
Juran, J.M., and Godfrey, A.B., Juran?s Quality Handbook, McGraw-Hill, Inc.
Sample size for variable data when nothing is known about the process or its distribution
Required Sample Size
Confidence Level
Reliability 50% 70% 90% 95% 99% 99.5%
80% 9 12 18 22 31 34
85% 11 16 25 30 42 47
90% 17 24 38 46 64 72
95% 34 49 77 93 130 146
99% 168 244 388 473 662 740
Table X of Appendix II, Juran, J.M., and Godfrey, A.B., Juran?s Quality Handbook, McGraw-Hill, Inc.
Sample size for variable data when process average, specification limits, and process standard deviation are all known.
Use the one or two sided factors depending on whether the specification is one or two sided. The ?K? values in the table are determined from the following:
Minimum of (average ? LSL)/s, (USL ? average)/s
Where ?s? is the standard deviation and LSL and USL are the lower and upper specification limits.
When the ?K? value falls between 2 values of ?n?, round to the larger sample size.
Here is what I use when performing process qualifications. Sorry for the copy/paste. Hope is is readable. Shoot me an email if you need some clarification.
Best,
Scott
Sample size determination for process qualification and validations.
Confidence level and reliability % will be chosen based on risk analysis.
The following confidence and reliability levels should be used corresponding to the severity ranking from the risk document.
*Not many on the forum will agree with the C/R table below; too low or too high. The table that I am sharing is from a specific project at a former employer (medical device).
Severity (Confidence/Reliability)
CA 95/95
CR 90/95
M 90/90
N 80/90
Sample size for attribute data
Required Sample Size
Confidence Level
Reliability 80% 90% 95% 99% 99.9%
80% 8 11 13 21 31
85% 10 14 18 28 42
90% 16 22 28 44 66
95% 31 45 58 90 135
99% 160 229 298 458 687
Juran, J.M., and Godfrey, A.B., Juran?s Quality Handbook, McGraw-Hill, Inc.
Sample size for variable data when nothing is known about the process or its distribution
Required Sample Size
Confidence Level
Reliability 50% 70% 90% 95% 99% 99.5%
80% 9 12 18 22 31 34
85% 11 16 25 30 42 47
90% 17 24 38 46 64 72
95% 34 49 77 93 130 146
99% 168 244 388 473 662 740
Table X of Appendix II, Juran, J.M., and Godfrey, A.B., Juran?s Quality Handbook, McGraw-Hill, Inc.
Sample size for variable data when process average, specification limits, and process standard deviation are all known.
Use the one or two sided factors depending on whether the specification is one or two sided. The ?K? values in the table are determined from the following:
Minimum of (average ? LSL)/s, (USL ? average)/s
Where ?s? is the standard deviation and LSL and USL are the lower and upper specification limits.
When the ?K? value falls between 2 values of ?n?, round to the larger sample size.
Thanks Scott, that's good stuff right there. I'm going to put that in my back pocket for later. This is in reference to medical devices.
Part of my issue is that the lot size is small, say 25 per run, 1 run per shift (It's a long process).
Do you think AQL sampling would be a reasonable approach to take? AQL typically is used for ongoing or incoming inspection sampling, so I'm unsure if its a reasonable approach.
Part of my issue is that the lot size is small, say 25 per run, 1 run per shift (It's a long process).
Do you think AQL sampling would be a reasonable approach to take? AQL typically is used for ongoing or incoming inspection sampling, so I'm unsure if its a reasonable approach.
Do you have historical process data? Mean and Standard Deviation
-No
What does your pFMEA and dFMEA say as far as risk level?
-low
I am assuming that your product is costly and the testing is destructive?
-Yes, it is destructive testing.
Variable or attribute?
-I could make it either a variable or an attribute.
S
Scott A. Bednar
Given your responses and not knowing much about the product, I would sample 16 from each production run over 3 days/runs. It may be expensive but that is the cost of doing business in some situations. You could take a look at the "sample plan analyzer" offered by Wayne Taylor (Taylor Enterprises). You could adjust your numbers from a risk perspective and possibly knock off a few additional samples. Contact me via email if you would like to discuss details about the process that you are not comfortable posting. Most of my process validation experience is related to plastics, injection molding, extrusion, post processing, cleaning, CNC (metals/plastics) and coatings (lubricants/drugs).
An alternative is to think of this as a process rather than a comparison of the new material to the old ( or to some specification) In PQ I tend to use control charts. this way I am not thinking about each lot as an isolated entity but part of a continuous process and my sample sizes can be substantially smaller.
If you have historical data we could help show you how sample sizes can get smaller and how you can use control charts.
If you have historical data we could help show you how sample sizes can get smaller and how you can use control charts.
Similar threads
