Hi,

I want to know what is a good lot size?

Suppose if the purchase order is for 50000 parts. We use the C=0 Sampling plans at AQL %1.0 for in-process inspection. So for a lot size of 50000, the number of samples to be inspected is 74. Since 50000 is a large quantity and because we don't have enough confidence in our process, what we do is break that purchase order of 50000 into 10 lots, which is 5000 per lot.

For a lot size of 5000, the number of parts to be measured from the C=0 sampling plans at AQL 1% is 50. So for 50000 and 5000 per lot it would be 10 lots multiplied by 50 i.e 500 parts to inspect instead of the actual 74.

I don't feel this a right way of doing it because we are inspecting 500 instead of just 74. We can't run 50000 parts in a day, the maximum number of parts that we can run is 35000 on a good day and 30000 on an average. This particular part doesn't run on night shift.

What's happening here is since we divide our purchase order into 10 lots, although are Ppk values look good. How much confidence should I have in the process? and I don't know how the ppk value is, if we measure just 74 parts over the 2 shifts to complete the purchase order.

Can someone tell me a solution for this problem?

Sounds like you already know about determining a statistically valid sampling plan. There are posts here which describe how to do that which you can search for.

The thing I would look into is since you don't have confidence in you process and can only produce 30000 units daily, I would develop a sampling method to catch the out of control process before wasting a day's production and creating 50000 of scrap or rework.

Do you currently have any kind of SPC? What is the root cause of your lack of confidence in your process?

Bryan

We use an SPC software but it s not a recognized one. Anyways, its called Magic Windows Professional and they don't even have a website.

So how should this sampling plan that you are talking about?

The process is having problem because of the inconsistent material. I'm doing Gage R&R's to at least make sure that there is no measurement system variation. The tolerance for the process is .75mm, it is really small. Most of our products are like that but .75mm is a tight tolerance and that is the reason why it is so hard to maintain a Ppk of 1.33 for a lot size of 50000. But right now we are doing double inspection I told you 500 units instead of 74 but we inspect 1000 actually. If you can see that is so much more than 74.

I don't want us to continue this way because I feel we are inspecting way too high than we should.

I agree that SPC is the way to go. Basically you would be sampling throughout the day to find problems soon after they occur. This way you could fix the problem before ruining the rest of the day's output. A typical application of SPC might have someone checking 4 samples every 15 minutes.

There are many discussions her about SPC that you can find with the search features. Another good place to look is the NIST Engineering Statistics Handbook: http://www.itl.nist.gov/div898/handbook/pmc/section3/pmc31.htm

Tim F

I agree that SPC is the way to go. Basically you would be sampling throughout the day to find problems soon after they occur. This way you could fix the problem before ruining the rest of the day's output. A typical application of SPC might have someone checking 4 samples every 15 minutes.

True, sampling plans are primarily used to verify that previous controls were effective - as in receiving or final inspection. They are really irrelevant for in-process inspection sampling. SPC is the way to go, and what the process and dimension is has everything to do what the correct sampling technique may be - not the software. Even the 'best' software can be inadequate based on the process.

As far as sampling frequency, again, that depends on what the process and dimension is. For example, in precision machining the sampling frequency is very clear and easy to develop (for control of assignable causes). For normal probability processes, they are based on best guess of risk and cost of not catching special causes weighed against inspection cost.

For incoming and final inspection, the biggest lot size is the best - to reduce inspection cost.

As far as sampling frequency, again, that depends on what the process and dimension is. For example, in precision machining the sampling frequency is very clear and easy to develop (for control of assignable causes). For normal probability processes, they are based on best guess of risk and cost of not catching special causes weighed against inspection cost.


One problem we face with our processes is because of the tight tolerance like .75mm, the data doesn't follow a normal distribution. Is that ok? or is it a must to follow a normal distribution?

So does precision machining follow normal distribution?

Thanks

One problem we face with our processes is because of the tight tolerance like .75mm, the data doesn't follow a normal distribution. Is that ok? or is it a must to follow a normal distribution?

So does precision machining follow normal distribution?

Thanks

I can tell you that properly controlled precision machining does not follow a normal distribution - in fact, if it exhibits a normal distribution, it is typically out of control. To read more about that, see: Statistical process control for precision machining Part 1 (http://elsmar.com/Forums/blog.php?b=79) and Statistical process control for precision machining Part 2 (http://elsmar.com/Forums/blog.php?b=80)

Sampling for in specification parts does not have to be from a normal distribution. Sampling to determine the distribution of incoming parts as a inference of the process distribution is abjectly invalid.

Our process is not capable of precision machining for sure.

Because our Ppk's vary between 1.33-1.8 for this process and as I mentioned this is for a lot size of 5000. If we make 50000 I can't imagine how horrible our Ppk's would look.

We need to make the process more stable in order to increase our lot size and inspect lesser than we're doing right now.

Our process is not capable of precision machining for sure.

Because our Ppk's vary between 1.33-1.8 for this process and as I mentioned this is for a lot size of 5000. If we make 50000 I can't imagine how horrible our Ppk's would look.

We need to make the process more stable in order to increase our lot size and inspect lesser than we're doing right now.

lot size should have nothing to do with capability or stability at least as you have described it.

perhaps if you could post some of your data - even the results from yoru inspectiona sampling - we can provide better suggestions.

One other question: when you are doing sampling inspection how are the samples selected? are they random? if random what mechanism do you use to ensure randomness?
Inspection lots can be of any size so long as the samples are randomly selected. And the distribution doesn't have to be Normal to do sample inspection.

However, the distibution does have to be Normal for the standard capability formulas. IF you have a uniform distribution you woudl actually be understating your Ppk values....