Hi,I have the task to come up with a sample size for a new

[Manana]

Hi,

I have the task to come up with a sample size for a new product, X. We manufacture 5 pieces a day of this product X and therefore has an yearly volume of 1300 pieces. Btw, we can manufacture only one product X at a time and hence its not done in batches. The reason behind this task is to determine variation in the SIZE of the finished product. The various parameters that we'd measure for each of these samples would be the room temperature, the final size, the time, date & the pressure.

Any sort of help would be appreciated.

Thanks!

 

[Bacle2]

Hi,
Wouldn't the total then be 1825, or, if you take weekends off, 1565 ? Also, I am

not clear on what you're trying to measure: do you want a confidence interval for

the true size of the product? And It seems you want to then study the effects of

the factors you mentioned on the size ? Maybe some factor analysis?

 

[chiro]

Hey Manana and welcome to the forums.

I need to stress something and I think it's really important to point out: although you have pointed out some important variables (like temperature and so on) which is a lot better than what some people will start off with (i.e. they just get data and then run a t-test or an ANOVA without thinking about what the hell they are doing), the thing that is missing that is extremely critical is knowledge of the process.

You have quantified some elements of the process but have not explained the process in any significant detail.

You can not expect to be able to get both a decent model for variation or any final advice or recommendations without a very detailed discussion of the process in a high enough depth, and this is something that you need to talk about in depth before getting advice that not only you are more comfortable with using, but more importantly that the advisor is comfortable with disseminating.

In this situation, I would not want to give any kind of advice for something like this without knowing these kinds of things.

 

[Manana]

Hi,

Sorry if I was not clear.
This is a PFA lining process. We work on 3 machines to line a disc but using only one mould. So at any given time we'd be able to work only on one machine although 3 are available. We are looking to make two new moulds because the current mould gives way to material wastage. I'd like to find out the final dimension (diameter & thickness) of the disc when it comes out lined taking into consideration various factors like the room temperature, the machine, the person, pressure, etc.,
Therefore I need help in determining the sample size (how many sample discs should I measure before I determine the size of the mould) to measure the final dimension (diameter & thickness) which would then help in making a new mould thus reducing huge wastage.

Summary: How to determine the sample size?

Please help!
Thanks!

 

[Stephen Tashi]

You've made the problem clearer, butI don't understand the geometric picture yet. The product is a disk, but this is a lining process. I don't see how the two are related. Can you give a link to a picture of this type of product and its mold?

You need to clarify what factors you intend to control.

For example, suppose you make a new mold and your data predicts it will work best at a given temperature and pressure, do you intend only to operate it at that temperature and pressure? Presumably, if your data shows the mold works best with Person A operating Machine M then you won't be able have Person A and machine M be the only ones that use the mold.

If you don't intend to control variables, there can still be some benefit to recording them. However, the most important variables are those that can be controlled.

Even if you can control all the environmental variables, I don't understand how this will help you design a new mold. Isn't the design based on how the properties of the mold affect the final product? - it's dimensions? - its specific heat? etc.

A simple sort of "what sample size?" problem go like this. We want to estimate the mean value of a variable and we want a 95% probability that the estimated value we get from the data will be within plus or minus some given number of the actual value. How many independent random samples should we take to estimate the mean value? This type of problem is not hard to solve.

When you have a question like "What sample size do I need to estimate the mean value of D as a function of variables, X,Y,Z?" you are asking a much more complicated question. If you try to force this problem into the pattern of the previous paragraph, you'll get an answer that says you need a huge number of samples. For example, you would need a certain number of samples for the condition: Operator: Person A Temperature:200 Machine: M Pressure 97 and then you would need a certain number of samples for slightly different conditions such as Operator Person A Temperature 205 Machine M Pressure 93.

A practical approach is to assume the function D = f(X,Y,Z) is from a family of functions that are defined by a few parameters. Then the problem is to estimate a few parameters instead of estimating the value of an unknown type of function at the many possible combinations of variable values.

You haven't revealed what variable or variables make one mold (or final product) better than another. You might be dealing with several aspects instead of one number D.

There is an engineering discipline that analyzes problems that resemble yours. It is called "The Design Of Experiments". That field of study is much narrower that its grand name suggests. Like most engineering versions of statistical theories, I find the jargon it uses repulsive. However, it is one thing we should look at. Perhaps you can find a local consultant in "The Design Of Experiments".