Which control chart for incoming (receiving) inspection - Sample size of 10

[Turk 182]

I know this is a "newbie" question, but which control chart would be best suited for a sample size of 10, whereas the samples are being measured for conformance to spec prior to approving.

 

[bobdoering]

Re: Which control chart for incoming inspection

It really depends on the characteristic, and how the characteristic is manufactured (its original distribution). My best guess without any further information is that it should be the same chart the supplier uses (if they use one, and if they do it correctly).

 

[Turk 182]

Re: Which control chart for incoming inspection

Let me be more specific: I am measuring a physical characteristic of an electronic component (10 samples) against our drawing for conformance to spec. I want to know probability that this may fall out of spec.

 

[Bev D]

Re: Which control chart for incoming inspection

Let me be more specific: I am measuring a physical characteristic of an electronic component (10 samples) against our drawing for conformance to spec. I want to know probability that this may fall out of spec.

The objective you state cannot be accomplished with a control chart. you are looking for acceptance sampling tools. Is the characteristic continuous data (aka variables) or categorical (aka attributes or pass/fail) data?

 

[Turk 182]

Re: Which control chart for incoming inspection

...Dare I say both - for sake of learning?

 

[AndyN]

Re: Which control chart for incoming inspection

I'm not sure you can do what you would like, by running a control chart at Receiving. Somewhat along the lines that Bob says, would your supplier give you a copy of their process control charts, instead?

I struggled with the same kind of issues at Receiving and, frankly, without knowing the supplier's process capability, it's a bit of a gamble setting any sample, plotting the results and trying to detect any oot parts. I'm not convinced of the value. If it's a propriety component design (is it your spec or the suppliers?) you could ask them for their results for a specific parameter which is critical to you. I certainly don't see anywhay to detect any out of tolerance items this way, but you might see a long term drift from a large amount of sampling.

 

[bobdoering]

Re: Which control chart for incoming inspection

Getting data from the supplier, understanding and trusting it (supplier development) it a much more powerful tool. I probably can not give you a good electronics example, but I can give you a machining example. For precision machining, we know the process generates a non-normal distribution. By following the live data, and ensuring that the adjustments were made appropriately, you know the product you receive is OK (except for special causes that may not have been dealt with properly, such as tool breakage). Now, when you receive product, it will be a homogenized group from this distribution. Your sample may appear to be normal, bi-modal, or any other distribution. You will not be analyzing the product with the same distribution of the process. You might try to use the wrong statistics, which will make the product distribution and capability look worse than it is. Also, you will find those here in the cove that are adamant that using control charts to determine probability is bad practice. If you want confidence, it better to get it at the source.

That's my vote.

 

[Turk 182]

Re: Which control chart for incoming inspection

Beautiful.... Thanks so much.

 

[Steve Prevette]

Re: Which control chart for incoming inspection

...Dare I say both - for sake of learning?

To answer the original question.

If you are tracking what percent of items sampled have one or more defects, you are likely looking at a p-chart. 10 items per datum on a p-chart is a little small, and I would lean more towards 20 to 30, assuming failure rates are in the 0% to 10% range.

If you are tracking how many defects per item sampled (with the possibility of more than one defect per item) you are looking at u-chart.

I would agree with the assertion that it is much better to monitor the measurement itself rather than the go - no go result. But if you do have several dozen measurements per item, there can be some advantage to just rolling to a p-chart or u-chart. Plus we can compare parts with different specifications.

 

[Bev D]

In addition to what Andy and Bob have suggested (have your suppliers submit their charts - very powerful on many levels) and being cautious about distributional shapes, I would add a few resources for your investigation:

look up the "Lot Plot" method. It was developed by Dorian Shainin while at Hamilton Standard. I believe it is covered in Juran's Quality COntrol Handbook. It's not perfect, but it will get you started in thinking about acceptance sampling using continuous data. Then get a copy of Mil Std 414 (I forget what the ANSI number is but a quick google will tell you.) This will give you further ideas on this topic.

When you have categorical data, you can select your sample size in two basic ways: using ANSIZ14 (the old mil std approach) or you can go directly to the distributional formulas. I've attached a spreadsheet that allows you to do this if you know what defect level you want to detect and the % of time you want to detect it.

As always acceptance sampling requires RANDOM samples in order to achieve a representative sample.

The type of continuous data distribution is irrelevent if using categorical sampling - you only need to specify the defect rate you wont' allow. But it has a significant effect on acceptance sampling wehn using continuous data...all published continuous data sampling plans that I know of rely on the Normal - or near Normal - distribution of data...

You can use the suppliers control charts to tell you when to be extra cautious - if they have an out of control condition you might want to increase your sample sizes - OR ask what they did to correct the condition and ensure that no defective product escaped tehir plant. Remember our goal is eliminate incoming inspection not to get better at it!