I am trying to determine an easy to understand method for analysing the repeatability and reproducibility of a Measurement system for a destructive test. The test requires the breaking off of weld nuts and recording of the torque value. Can anyone help?
Send to friendSign In to get started!
Comments
firebrew 9/24/2001
Actually, you can indeed estimate sigma and calculate conttrol limits by lumping all the data points into one theoretical distribution. If you do however, you are no longer "doing" SPC.
I'm not sure what you are doing, but it ain't SPC.
The AIAG describes a methodology of doing this to estimate what they call a "Process Performance" index.
It allows you to describe the results of a process evalutation without making any assumptions about process stability or statistical control. The results of such an analysis are accurate for that one frame, but because statistical control is not verified, it is not valid as a predictor of future performance.
In nearly all cases, when you estimate sigma from the individual data points, and then calcualte 3 sigma limits, you will enclose all the data points within the limits and you can easily assume stability when it may not be there.
It's exactly the same as doing a "Statistics 101" frequency distribution analysis like you learned in your first semester class in college.It characterizes a pile of data using statistical words, but cannot predict future performance.
J Bruman
guest 9/12/2001
The objective of a process behavior chart is to separate the routine variation from the exceptional variation. The only reliable way to filter out the routine variation is to use the within-subgroup variation to compute the limits. Using global, descriptive measures of dispersion will mix both types of variation together and make it impossible to separate the two. Thus, all the correct ways of computing limits for process behavior charts use local, within-subgroup measures of dispersion rather than the global, descriptive measures.