raw data

OMG. No. This is NOT how you do statistical analysis. NEVER take a pile of data and apply mathematical formulas to it. Each formula has a specific purpose and a specific set of requirements for a valid answer as well as a specific set of assumptions that are being tested.

What do you need to understand about the process?

Is the data from multiple lots? Were the data randomly selected or was there a sampling scheme (subgrouping scheme)? Do you want to know if the process is capable? Is it statistically stable? (Is there any non-homogeneity?) what are the largest families of variation?

Is the data from one lot? Were the data randomly selected? Do you need to know if the lot has an acceptable defect rate?

First you have to know what you want to know understand about the process. This means you also have to understand the process and how it creates its results.
Next you need to collect data in a manner that represents the process variation. Then you plot the data in time series in a run chart unless the data are only from one lot and were randomly collected, then you plot it in a histogram. Add the specification limits to the chart. LOOK at it. Then you think about what that tells you.

Start there and post your results.

 

Well pick a characteristic and plot your data as outlined above

 

I would start with categorizing your data into the following categories because the category will dictate how you can analyze them:

• Continuous data: length, width, thickness, etc.
• Count data: number of fragments, etc. This may include appearance defects if you count the number of scratches per unit.
• Binary (pass/fail) data: This may include appearance defects if you reject an entire unit based on the presence of any scratches, regardless of the number of scratches.

Your initial questions implied continuous data and Bev's response assumed continuous data.

 

To expand on Miner’s point, you can plot binary (defective parts) data and count data in a time series plot and in a histogram if looking at multiple lots (or multiple sampling periods). You cannot apply specification limits to this type of data…

 

I can’t speak for Miner but I would look at a data set you sent here publicly IF you provide process details, the things you want to learn about the process, your initial analysis as outlined above and you tell us what conclusions you have and questions about your analysis.