1
1winnyAK
Last time I was here, I wanted to create a sample plan using the AQL Sample plan and now I don't think I am doing it right because my batches are so small and I want 95% of the time the parts are good or 99%. I found a post on here that got me confeused.
Can someone please help me explain to me I'm I going the wrong direction on my sample plan approach?
"CAPABILITY MEASUREMENTS like Cpk and Ppk are typical measures of how well you are actually doing. You calculate these by looking at your measured results, finding the mean and standard deviation, and then doing some calculations comparing the mean and standard deviation to the specs. (And as Bob would rightfully point out, for precision machining you might want to do the calculations slightly differently than you would for other processes.)
A typical rule of thumb is to measure at least 30 pieces. More is better, but 30 gives pretty good results. Because capability calculations are based on variable data (ie numbers that can take on many results over the range of interest), a relatively small number of measurements (like 30) will give clear and effective results. These sorts of capability measurements are well known and well accepted in many industries.
This should be done in conjunction with control charts to see that the processes is behaving consistently.
AQL SAMPLING PLANS, on the other hand, are based on a value you choose. Basically you are say "if the parts meet the chosen AQL (or are better), then the lot will usually be accepted, and if the lots are noticeably worse than the chosen AQL, then the lot will usually be rejected." In your case, an AQL sampling plan at the end (or when they receive the parts) could be effective as insurance on top of your other measurements. For example, if a shipment got mislabeled, a sample of a few parts would quickly spot the error.
Because AQL plans use attribute data (eg pass/fail), they require large samples if you want a high degree of certainty. You might check this thread for info Confidence and Reliability - Reference to a 90/95 confidence and reliability level. For exampl, if you wanted to be 95% certain that at least 99% of the parts are good, you would have to sample 299 parts! Since you don't make this many parts, then this is not effective. But 30 parts for an estimate of Cpk could give you a similar certainty.
Let me recap my opinions on the matter:
Can someone please help me explain to me I'm I going the wrong direction on my sample plan approach?
"CAPABILITY MEASUREMENTS like Cpk and Ppk are typical measures of how well you are actually doing. You calculate these by looking at your measured results, finding the mean and standard deviation, and then doing some calculations comparing the mean and standard deviation to the specs. (And as Bob would rightfully point out, for precision machining you might want to do the calculations slightly differently than you would for other processes.)
A typical rule of thumb is to measure at least 30 pieces. More is better, but 30 gives pretty good results. Because capability calculations are based on variable data (ie numbers that can take on many results over the range of interest), a relatively small number of measurements (like 30) will give clear and effective results. These sorts of capability measurements are well known and well accepted in many industries.
This should be done in conjunction with control charts to see that the processes is behaving consistently.
AQL SAMPLING PLANS, on the other hand, are based on a value you choose. Basically you are say "if the parts meet the chosen AQL (or are better), then the lot will usually be accepted, and if the lots are noticeably worse than the chosen AQL, then the lot will usually be rejected." In your case, an AQL sampling plan at the end (or when they receive the parts) could be effective as insurance on top of your other measurements. For example, if a shipment got mislabeled, a sample of a few parts would quickly spot the error.
Because AQL plans use attribute data (eg pass/fail), they require large samples if you want a high degree of certainty. You might check this thread for info Confidence and Reliability - Reference to a 90/95 confidence and reliability level. For exampl, if you wanted to be 95% certain that at least 99% of the parts are good, you would have to sample 299 parts! Since you don't make this many parts, then this is not effective. But 30 parts for an estimate of Cpk could give you a similar certainty.
Let me recap my opinions on the matter:
- SPC (control charts) shows that you are being consistent
- Capability (Cpk) shows how well you actually have done relative to specs.
- Sampling (AQL) provides a rough check of how well you have done "
. This is a large sample size for lots of 7,000 - 8,000 units that are produced in 8 hrs. I have read a lot of threads about sampling plans, and I would like to know the difference between using ANSI or the OC curve (alpha / beta risk). Is there a difference statistically? Which one is preferred by FDA? Also, do the alpha and beta risk is soemthing agreed by the company or is there a norm? Any info or recommendations will be very appreciated!!!!
