Determining QA Sample Size and method
Hello,
I am curious what methods other manufacturers of dietary supplements or like products are using for reference in determining a sample size for QA QC checks per hour. We currently are using a calculation similar to the Raosoft Calculator to determine how many units we are checking. For example, we figure margin of error , confidence level, population size and response distribution.
We are mainly trying to figure this for checking seals on product at the moment. Any feedback is helpful.
Thank you,
I always use ISO 2859-1:1999(E) for determination of sampling size such as this.
How To Read the AQL Tables for Quality Inspections (Random Sampling) - Bing video
This is pretty useful as well if you want to easily understand the tables.
I also work with dietary supplements, sorry I left that out.
Hello,
I am curious what methods other manufacturers of dietary supplements or like products are using for reference in determining a sample size for QA QC checks per hour. We currently are using a calculation similar to the Raosoft Calculator to determine how many units we are checking. For example, we figure margin of error , confidence level, population size and response distribution.
We are mainly trying to figure this for checking seals on product at the moment. Any feedback is helpful.
Thank you,
Hi Sarah,
It may depend on what you are specifically measuring/trying to achieve. For example SPC ?
I am trying to determine what a valid sample size would be to ensure that product is sealed, for example. Seals on our jars are the main problem. We do hourly QA checks where the QA Tech will go into the room and pull x amount of jars to ensure they are properly sealed by the induction sealer. Because we are doing hourly checks, we are trying to determine how many jars an hour we should check and at what point to reject the output for that hour. We are currently using something similiar to Raosoft calculator but these methods and AQL all seem to go by "total" lot size for checking. So we are plugging in the total lot size, getting how many checks the calculator says we are to do, and splitting those up across how many hours of production it will take to complete. For example , if we were running 10,000 units and the sample size calculated to 350, and I knew that would take production about 4 hours, I would calculate 350 divided by 4 to determine how many jars I am checking an hour.
I am looking for something more robust, something that makes more sense. I do not feel that this method is working. Any suggestions on what I am currently doing and new suggestions would be greatly appreciated. Does anyone else do hourly QA samples/checks and how do you figure the amount to be checked and at what point you are rejecting the hourly output.
When I was in cannabis we had much the same problem
A) are you sure that the setting on the induction sealer are correct for the size/type of lid
B) production should also being checking some, e.g. for every 100 they seal, check 5 then your QA check is above and beyond
That AQL number SHOULD NOT be divided, it's telling you that's the sample size/check not per lot so based on what you said, your QA should be checking 350 units/hour
what is your current failure rate? and why do you think it isn't working? if you're catching defects, it's working
What settings (attach the link if you can) on the AQL calculator are you using? It's not just sample size that matters either, it's your pass/fail sensitivity
Can you give me an example of how to properly figure checks for QA/Production? For example, If I am making a lot of 10,000 units, they are running approximately 1500 an hour. What would checks looks like and what method are you using to get these numbers? At what number of 'Bad" seals would we stop and move that production that was done for that hour since the last QA check over to the side for rework?
https://tetrainspect...ampling process.
This link allows you to input your specifics, and has pre determined pass/fail based on the level of inspection
if your getting alot of failures, something isn't right with your process, old caps, induction sealer not correct for cap type, dwell time not long enough
And you cannot check the seal immediately, they need time to cool and set
I would have production check there own work, 5 caps for every 100 they make
Until you do that step, you won't know how tight your inspections really need to be
Thank you, I went to the link and plugged in a random number. For 10,500 units the sample size is 315 with accept at 14 units and reject at 15 plus. If I am having production check roughly 5 every 100 , for example, and 30 minutes into the process they found 6, what was your process? Do you continue? I am trying to figure out this AQL method and process. Say halfway through the lot at 5,000 units we finally hit a total of 14 unaccepted units but we are readjusting our process throughout the way fixing issues as they occur with the inductor, technically the lot is not rejected because we are finding the issues and checking every hour. Our capper can be finicky as well as our inductor. I am trying to justify a way that we are doing our checks on paper as well. If I use this calculator and it is saying 315 checks, how should I disperse these really? I apologize for all of the questions.
Hi Sarah,
The AQL procedure specifies the requirements such that the quality of yr defined lot is compliant (or not) to the desired AQL level. Period.
Continuing in the context of previous posts - assuming you can accept an AQL of, say, 2.5%, you could define a 1day production of 10000 as yr LOT and, for example, set Inspection Level to S-4 which gives a sampling pattern (using Post 9 link) as shown directly below which avoids colossal sample sizes (S1-S3 give even smaller numbers). Alternatively you could presumably define a LOT as the quantity produced in 1 hour, it's up to you. But IMO you should understand the meaning/trade-offs if using the S-level settings.
defect matrix.PNG 170.21KB 1 downloads
The above approach iooks to be (vaguely) supported by these (not so simple) links for sampling continuous flow processes -
https://www.isixsigm...ous-process-92/
(note the above link is slightly flexible, albeit speculative, concerning the restriction in Post 7)(also suggests SPC [see later in this Post])
https://qualityinspe...ng-plans-china/
https://elsmar.com/e...-process.56808/
More generally, you maybe need to decide what kind of (acceptance) sampling you wish to implement, eg (a) isolated lot sampling, (b) series lot samplings or (c) SPC.(Statistical Process Control). Crudely, (a) determines compliance to specification for a given completed lot, (b) is similar to (a) but based on a series of completed production lots, (c) monitors compliance of an on-going production lot. A fundamental initial decision is how you define your LOT.
([mainly]a, [slightly]b) are discussed/illustrated in 1st pdf below but, ideally yr situation is maybe best handled via SPC which involves initial verification that process is statistically "under control" followed by setting up simple graphical acceptable/unacceptable defect rates then doing (minimal) sampling at routine intervals to ensure system stays under control. SPC is a very well-documented technique intended to be simple to operationally implement but typically requires some basic understanding of statistics.
GENERAL_GUIDELINES_ON_SAMPLING.pdf 1.52MB 45 downloads
(b,c) are overviewed/compared in 2nd pdf . This document is primarily, mathematically, focused on (b) but the 3 introductory pages have a nicely readable review..
Acceptance sampling continuous processes.pdf 309.32KB 35 downloads
PS - Procedures such as speculated in Post 8 do also exist (eg CUSUM) but tend to get (even) more complicated.
What do you think about Six Sigma ? Do you think that is plausible for the type of manufacturing we are doing or SPC? I do not not know much about Six Sigma or what the industry standard for quality checks is based off of for most companies out there.
What do you think about Six Sigma ? Do you think that is plausible for the type of manufacturing we are doing or SPC? I do not not know much about Six Sigma or what the industry standard for quality checks is based off of for most companies out there.
Hi Sarah,
AQL/SPC are both massively used in Industries where processes are "well" statistically controllable. Related software is probably readily available. Neither worked for me for micro. data because my scatter was too much. There are potentially ways around this (eg logs) but such involved (significantly) more effort for a non-statistician. SPC has been utilised for plate counts in the Milk Industry
I predict will be similar situation for yr defects. SPC requires an initial estimation from actual data to evaluate if process is under statistical control. The procedure seems to all be available in Excel now or via software, eg -
https://www.morestea...control-spc.cfm
.
SPC is relatively easy to implement if process control is "good".
Clients would, obviously, favor zero-imperfection items or administrations; the best satisfactory quality level. Nonetheless, venders and clients normally attempt to show up and draw satisfactory quality lines dependent on factors regularly connected with business, monetary, and wellbeing concerns. https://www.atihongk...in-simple-terms
Clients would, obviously, favor zero-imperfection items or administrations; the best satisfactory quality level. Nonetheless, venders and clients normally attempt to show up and draw satisfactory quality lines dependent on factors regularly connected with business, monetary, and wellbeing concerns. https://www.atihongk...in-simple-terms
Thanks for useful link but please avoid duplicate posts.