Acceptable quantity levels/Sigma levels for reducing the frequency of quality tests.

Hello, I am currently a graduate engineer for a manufacturing company, we produce injection mould caps and blow mould bottles along with steel drums. We are currently looking to reduce the frequency of the injection mould physical quality checks we carry out from every 2 hours to every 4 hours.

I have information on the quality control parameters (Cycle time, filling time, shot volume, etc) for each part produced over the last year. The process is very stable with low variation. I was wondering if anyone knew of any standard sample sizes and sigma levels/AQL levels/Standard deviation/Variability which I can compare with the quality control data I have gathered to act us justification for the reduction in frequency of our quality checks (instead of just saying that we feel the variability is small and the sample size is adequate to say it is a low risk change).

Please let me know if you have any advice/information that can help me out.

Thank you very much!

welcome PaulC333!

First understand that sample size calculations and AQLs have nothing to do with frequency of taking samples.
The frequency of sampling is based on a couple things: What kind of sampling are you performing? how much bad product you are willing to make before you detect it? And when do the natural change points occur?

There are 2 basic types of sampling: Lot acceptance and in-process SPC or Change control. It sounds like you are doing in-process/change control or SPC type sampling. These do not have statistically calculated sample sizes. The sample size is typically quite small (SPC is and should be very small; 3-5 units is usual sufficient and for plastics a single sample is often sufficient). The purpose of in-process inspection is to sample when a natural change occurs (new material, new setup, process adjustments, startup, shutdown, etc.) If something goes wrong at these change points it only takes a few units to detect it. If the process is stable and you are confident that it will remain so then you can reduce the frequency of when you take samples.

Lot acceptance testing does use AQL/RQL sample size calculations and random sampling. The AQL/RQL choices shouldn’t Be based on historical, current or future defect rates but on the defect rate you do not want to ship to your Customer and so that should not change (contrary to some popular beliefs).

Does this help?

Thank you for the quick reply Bev D,

This does help me better understand the different quality sampling.

Our current sampling performed consists of a physical weight check of the injection mould cap, internal and external thread diameters, go/no go gauge, checking the cap fits a sample bottle and checking the regrind percentage.

The purpose of the testing is to make sure we do not send out any out of specification or defective parts to our customers. We currently do this every 2 hours but given that the test very rarely fails and the low variation between the injection mould quality control parameters (Cycle time, filling time, shot volume, etc) means we feel it would be beneficial to change this to every 4 hours to save time.

I was hoping to use questions like "how much bad product you are willing to make before you detect it" as you said but I have been asked to come up with a justification for reducing the frequency of our sampling using the variation of the quality control parameters (Cycle time, filling time, shot volume, etc) I have gathered.

I was wondering if anyone knew of any statistical thresholds (for example sigma level) and an adequate sample size which I can use to compare with the quality control data I have gathered to act us justification for the reduction in frequency of our sampling.

Hope this explanation clears up my enquiry, although it is quite long itself.

Thanks,
Paul

PaulC333 said:

Thank you for the quick reply Bev D,

This does help me better understand the different quality sampling.

I was hoping to use questions like "how much bad product you are willing to make before you detect it" as you said but I have been asked to come up with a justification for reducing the frequency of our sampling using the variation of the quality control parameters (Cycle time, filling time, shot volume, etc) I have gathered.

I was wondering if anyone knew of any statistical thresholds (for example sigma level) and an adequate sample size which I can use to compare with the quality control data I have gathered to act us justification for the reduction in frequency of our sampling.
Paul

Click to expand...

Sorry there are no statistical justifications for changing the frequency of sampling, which is what you are trying to do. Not to be rude, but did you actually read and understand what I posted? What is it about my response that your management cannot understand? If you are persisting to pursue an answer that doesn’t exist because your management is persisting then the answer is to explain the truth to them. Or just make something up. Because frankly if they can’t understand my response then they really don’t even understand statistics, so making something up won’t hurt.

To help clarify and simplify the answer: since there is no statistical reason for determining the frequency of sampling, there can be no statistical reason for reducing the frequency of sampling.

Thank you for the clear reply, I understand what you are saying. I just posed the question again to be absolutely sure that it is not possible, I should have just asked you to confirm that it was not possible but I was not 100% confident in my understanding of your first reply. I will make them aware of your answer, thanks again.

It really boils down to how much material are you willing to accept as "at risk"? When you sample every 2 hours, you have 2 hours of product "at risk". That will be 2 hours of product that you will either have to hold before further processing or track down and quarantine after additional value has been added. If you increase your time to 4 hours, you have doubled your population "at risk". That is why Bev said that you should coordinate your sampling with changes that are known to cause shifts in the process. That helps you minimize the population 'at risk".

Paul,

Welcome to the cove.

I was in injection moulding once and I was responsible for setting the type and frequency of the production quality checks. I went through years of data on production quality checks and customer complaints and I found the following conclusions:

• Product was inspected hourly for common visual injection moulding defects, 2-4 dimensional checks, part weight. Sample size was 1 per cavity
• The overwhelming majority of production and customer issues were common visual injection moulding defects; short shots, flash, burning, splay, etc. Very few were size related
• Those that were size related were all from one customer, about 3-5 tools from around 200 live tools at any given time

Based on this we cut down process inspection for nearly all products over a 2 year period. Aside from a few problem jobs we eliminated dimensional inspection to first off, last off and start of shift only. I would have cut it down to just first/last off but I feared they would lose measurement competency if they only measured parts every few days. This, tied with the process improvements we made, reduced customer complaints and reduced shop floor issues, problems were identified sooner as the processes were better and we focused quality checks to what our data showed was likely to change.

I hope this helps.