Spec limits for process screening platform

[#1]

Hello All,

I have a question regarding spec limits in process screening applications. For example, I have a metric for thickness of a part. The target is 0.7 with lower limit of 0.6 and upper limit of 0.8. Anything within these limits inclusively will pass. However, as shown from the plot below values of 0.8 are counted as fails. At the same time 0.6 are included as passes. 

Is there a way to control if the limits are inclusive or exclusive? Is it a setting within the preference tab or can it be done within the process screening platform on a case by case basis?

I am currently using JMP 16 but might be moving to 18 soon so any suggestions for each version would be appreciative

[#2]

Are you sure the values are exactly 0.8? In this simple example no values are counted as fails (I'm using JMP18 though, so maybe this has changed)

[#3]

Hey @AdditiveRange10 

Try checking your data for having a crazy small decimal.  I've had data sets come in with weird import problems that lead to a small decimal that threw off my capability analysis.  You can try Round(dt:col,2) to get rid of it.

[#4]

No definitely the values have been rounded before so there are no small decimals.

I ran the analysis using the process screening platform in JMP 16 and it is as I stated above and the values which are 0.80000 are counted as fails. In JMP 18 however using the process capability platform I get this. Selecting all of the points which are inclusively inside 0.6-0.8 I get 93/107 = 87% which is the passing rate as shown in the analysis

Still the question remains that visually the histogram looks misleading with the [0.8-0.9) bin seeming outside of the USL. Do you have any suggestions on how to change that. I know the hand tool might be useful yet it does not seem a good option for scripting an analysis for many metrics

[#5]

With hand tool you can change Bin Span and you should be able to script that but it might require some figuring out.

[#6]

Hey @AdditiveRange10 

Maybe between these two you can get there

May need to use the xpath location to assign it to the correct chart.

[#7]

Not possible to make set the specs inclusive/exclusive.  Looks like your measurements are discrete so I would set a LSL slightly lower (ex. 0.599) and a USL slightly higher (ex. 0.801).  I am not sure about JMP 16, but in JMP 18 I see that measurements = spec are considered in spec, but comparing floating point values is sometimes problematic due to imprecise representation.  Always better to do > or < comparisons rather than == for floats.

[#8]

Hmm I just tried something else. I used the process capability platform and plotted 10 datapoints with the same spec limits LSL = 0.6, Target = 0.7 and USL = 0.8. Based on the data there are 3 points that are 0.9 or 1. If I type in the spec limits manually I get a nonconformance rate of 30% i.e. 3/10.

However, if I use the import spec limits with a data table then I get a nonconformance rate of 60% i.e. 6/10. Could this be due to the type of LSL column properties or? It is a continuous numeric column with format of "Best" with width of 12. I tried varying the width and recalculating but not difference.

Again the whole point is that I use values of LSL and USL from a table since I am doing an analysis for many variables and cannot type them in individually

[#9]

Yes, this confirms it is a precision issue with comparing two floating point values.  Best to avoid this with the method I describe above.

[#10]

So you are suggesting to not use the process capability or process screening planforms? The spec table is generated using a Python script so I have some freedom to play with the formatting there? Do you suggest anything in particular?