How to Determine a Good Sample Size?

There's length, diameter and wall thickness....what width? You also need to know the material, such as PET...

https://www.ventionmedical.com/components-and-technologies/heat-shrink-tubing/technical-information/

Dear Solar Eagle,

Sorry for the confusion. I will be measuring the length and width of the shrink tube after the heat shrink process (As we know, both the length and width will be reduced due to material shrinkage after heat shrink process). These measurements will then be tabulated in a table (depending on how many samples are measured) for the calculation.

Would you happen to know how can I determine a good sample size for my study?

Thanks.
Andrew

Well it seems to me that if it's a tube, you should be measuring the length and diameter....I think the number of samples should be determined by the variability of the product and process....

Sampling is sampling. It is what it is? How to determine a large enough sample size to get a substantiative result? A matter of opinion or experience. Beyond that other factors come into play. I'll leave it at that.

Since polling is not my modus, you can take my advice with a grain of salt.

But there are certain modus operandi that are true, common, sound practices - whether you are polling an audience or even a series of material devices.

Approximately how many shrink tubes do you anticipate having to measure in any given period of time?

There's a parameter to work from...

Take a look at the C=0 zero acceptance number sampling plan by Nicholas Squeglia available on ASQ.org. Pretty standard Quality stuff.

Try mil-spec 105E. All about sampling size determination.

Would this be a corrective & permanent action required by a customer?

if so ask the customer. but quote AQL to limit the quantity a customer will try to require.

if it is an internal optimization I would go with AQL.- however I would have to be doing the inspections by hand not a camera system. this should drastically increase your inspection times. hence increasing the lot sampling.

It's my understanding Mil-std 105E was cancelled some time ago and replaced with Mil-std 1916 and / or ANSI/ASQZ1.4. I've always used either the Z1.4 or aforementioned C=0 plans - your choice.

Question - if you're using a vision system why sample? Since it's automated why not just 100%?

Good question. If the sampling is being done for purposes of quality-control (perhaps), why stop after an indeterminate set of iterations--To save money, or to calibrate the machinery used in the product's manufacture?

say you have to mount a IC having 32x32 pins

the tube won't always shrink homogeneously enough 'cos it's poor structure

-- you need optimize the tube mounting to least expensive (at overall) steps

say ? center to out -- inspect each step (recycle/repair on fail)

optimize processing schemes for the individual work areas

---

versus you use 32x32 x say 1/4' 's ~2ft of tube for nothing gained ... ever

how determine a good sample size? Use calipers

This might help...

http://www.statisticshowto.com/find-sample-size-statistics/

Preferably, you need hypothesis testing to reject or accept a certain criteria such as acceptable length or diameter.

You can compare means or differences between means, could be from a paired observations and different ones.

T distribution let you sample about 30 samples on excess you can use Z distribution. You don't need a lot of samples if its a product from the same machine. But, if you want precision increase sample size to the point that you would have a relevant results.

There is this term in quality assurance and control "confidence level of 95%" and level of significant which just 100%-confidence level.

Go check some references or consult a statistician. If you are an engineer, oh, you can digest what the books says with out a doubt.

You could actually do some hypothesis testing also of different sample sizes and see if 100 samples would have produce the same variance with 25 or 10 samples. If they have the same variances then reduce sample size to minimum.

It would be practical to have a fewer samples in the aspect of cost and time if above condition is met.

Sorry but IMO this is putting the statistics ahead of the requirement.
Who gives a toss about the variability or the length unless it has to perform some function.
Decide what is the minimum acceptable length to perform the required function as your start point, then in order to avoid waste of materials or re work you need to determine the length before shrinking that will give the desired result, and that's where the stats' come in.

If the requirement is that they all "look the same" then the sample size depends on the process, raw materials etc in which case you prob' want to sample 30 from each bacth over say a day, week or whatever period suits the above mentioned factors. Or even constant monitoring to look for drift.
I tried this on a product, asking production to record the test figure from a sample from every batch.... they never did, despite it being written into the test spec' . Yes I am enjoying my retirement!
Del

Also, 2 things. I am bit confused if its the sample product you want to test or your new method.

Test an alternative or previous method employed in your QA/QC and compare with the new one. You could check either which is superior and more accurate. If the new method is accurate to the point of nm then well its good.

All samples in the population in whatever process production machine you got there is honest enough to tell the truth and that's what is needed in production. You could try to tamper product(especially in food industry) test and get fired later on.

Please, you seem adequately skilled to NOT use AQL, but to use SPC and quality assurance processes.

AQL by any standard essentially guarantees that the long term average of your product will have the failure proportion that you set as a threshold.

Step 1: determine the actual capability of your process. (Measure the parts, preferably in the sequence they are made, determine the single sided process capability, standard deviation and so on. 30 sample minimum, but ideally a group of 30 samples for each variable realistically possible such as each production shift, each colour of tube used, each supplier used etc.)

Step 2: Determine the control limits for that process. (There are tables with the factors necessary based on 2 sigma and 3 sigma towards the fail specification limit)

Step 3: Set the control limits for the sensor based on those parameters, where 2 sigma provides an alarm that the process is making parts close to fail, while 3 sigma would indicate a "stop" condition.

If you look up "control Charts" you will find the charts with relevant factors and scale and also some other rules relating to "runs" and other patterns. Usually the control charts run with grouped samples so that you can also monitor variability from the range observed in each sample subset.

with this you can move to a "zero defect" outcome.

good luck