J
jecuzens
Hello,
I have been tasked with performing a nested gage RR for a process wherein a very small volume of fluid is dispensed onto a plastic base. The size of the dispense is assessed by an automated inspection system. The fluid evaporates from the base rapidly enough that it cannot be inspected more than once.
Originally, we created surrogate devices by machining some plastic such that it resembled a dispensed part and analyzed the system in a crossed format. The problem with the crossed study was that the parts did not reflect the entire span of the potential process variation and it would be very difficult to do create parts that satisfied that requirement.
Fast forward to the present, we felt like a nested study should be performed with actual parts. Having never performed such an analysis, I largely followed a test method validation protocol that we use internally which stipulated that we should recruit (minimally) 45 samples consisting of 5 subgroups of 9 samples. "The subgroups should represent the range of samples to be tested". Each "operator" is required to measure 3 samples from each subgroup.
So, in my study design, in lieu of having 3 operators, I have the inspection system perform the analysis at 3 different exposure levels to ensure the system is robust under varying lighting conditions. To cover the process range, my subgroups reflected contrived samples wherein we manipulated the dispense system to create spots of various sizes. Each subgroup had more than 9 samples collected because I felt like there might be some part in-homogeneity that I would have to accommodate upon reviewing the inspection images and data.
After completing all of the sample processing I was left with data wherein I had 3 exposure settings (low, nominal, high) x 5 conditions (small spot out of tolerance, small spot, nominal spot, large spot, large spot out of tolerance). If I run this in JMP and claim that my process variable is size, grouping is exposure, and part / sample ID is the condition, I get a terrible precision to tolerance ratio in a nested model because it seems JMP assumes all of the parts should be homogeneous. If I run this crossed then the PTR is fine.
I guess my question becomes, how does one perform a nested gage RR in JMP wherein I can account for process variation in my parts?
I have been tasked with performing a nested gage RR for a process wherein a very small volume of fluid is dispensed onto a plastic base. The size of the dispense is assessed by an automated inspection system. The fluid evaporates from the base rapidly enough that it cannot be inspected more than once.
Originally, we created surrogate devices by machining some plastic such that it resembled a dispensed part and analyzed the system in a crossed format. The problem with the crossed study was that the parts did not reflect the entire span of the potential process variation and it would be very difficult to do create parts that satisfied that requirement.
Fast forward to the present, we felt like a nested study should be performed with actual parts. Having never performed such an analysis, I largely followed a test method validation protocol that we use internally which stipulated that we should recruit (minimally) 45 samples consisting of 5 subgroups of 9 samples. "The subgroups should represent the range of samples to be tested". Each "operator" is required to measure 3 samples from each subgroup.
So, in my study design, in lieu of having 3 operators, I have the inspection system perform the analysis at 3 different exposure levels to ensure the system is robust under varying lighting conditions. To cover the process range, my subgroups reflected contrived samples wherein we manipulated the dispense system to create spots of various sizes. Each subgroup had more than 9 samples collected because I felt like there might be some part in-homogeneity that I would have to accommodate upon reviewing the inspection images and data.
After completing all of the sample processing I was left with data wherein I had 3 exposure settings (low, nominal, high) x 5 conditions (small spot out of tolerance, small spot, nominal spot, large spot, large spot out of tolerance). If I run this in JMP and claim that my process variable is size, grouping is exposure, and part / sample ID is the condition, I get a terrible precision to tolerance ratio in a nested model because it seems JMP assumes all of the parts should be homogeneous. If I run this crossed then the PTR is fine.
I guess my question becomes, how does one perform a nested gage RR in JMP wherein I can account for process variation in my parts?
