| Author |
 Topic  |
|
|
atopf
Austria
3 Posts |
 Posted - 04 Mar 2022 : 06:22:35
|
Hello
We use the SPC charts in the manufacturing environment, which makes the use of the SPC charts directly related to our products and product liability.
Therefore, our customers (OEM) require validation of the statistics software used.
Question 1:
Do you have a certification from an independent body for this or can you show us other evidence that ensures us this software validation.
Furthermore, we have noticed that inaccuracies occur in the Cp values from a sample size N > 25. We attributed this to the fact that for the constant d2 only values in the range of 2 <= n <= 25 are defined - correct?
Question 2:
Is it possible to define the constant d2 also up to n = 125?
Finally, it remains unclear to us how the library handles non-normally distributed data. Usually, a distribution test, e.g. Anderson Darlington, would have to be made and then the data would have to be transformed according to e.g. Box-Cox or Johnson.
Question 3:
Is there a function that can be used to determine the distribution? And if necessary can be used for the transformation?
atopf |
|
|
quinncurtis
1586 Posts |
Posted - 04 Mar 2022 : 17:00:32
|
1. Sorry, but there has never been certification from an independent body. There has also never been a request for that in the 15 years we have sold the software. It is probably the end user program (what you write) that gets tested, not the support libraries (what we write) used by the programmer.
2. You are correct, in that we clamp the d2 value to a maximum of 25 samples, per the textbooks we used to develop the software. There is no way for a programmer to modify that behavior, it something we would have to do in the library source. If that is a critical element of your application, you can contract us to make the necessary changes. It doesn't sound very complicated.
3. The software does not do anything special for non-normal data and does not make a determination of that internally, nor do we have any externally callable methods (Anderson Darling in your example) that will do this. It follows that we do not transform the data to make it more normal. We have never had a request for these features. If that is a critical element of your application, you can contract us to make the necessary changes.
|
 |
|
| |
Topic |
|
|
|
CP-values and distribution
