Cpk, GD&T & Minitab? Please review my methods

[WKHANNA]

I have been struggling with this issue for months.

Can someone please review my methods and tell me if I am doing this correctly?

I have a GD&T for position to within .016" of a supplied powder metal gear that is pressed onto the knurled portion of a shaft that we fabricate in-house. Our process control on the knurl form is excellent & Cpk on knurl diameters at this machine is >2.8 with a tolerance of +/-.002.

I only have 30 measurements since this was the total production run for the customer's order for which I must supply a full PPAP. The customer is a V large US corp who does not care that the sample is far too small to yield meaningful data. They simply insist I supply data that provides a process Cpk of >1.33 for the specified characteristic.

I have entered my data into Minitab using the ‘Normal’ Capability Sixpack.
The entry for the Lower Spec has been left blank.
The entry for the Upper Spec is .016.
The entry for sigma Tolerance is 3.0.

Cpk results to 1.43 for my data.

Aside from the sample being far too small & the fact that this is an assy fit which obviously is a multi modal function, I am close to correct in my use of Minitab?

Regards,
Bill

 

[Barbara B]

Sounds like a probable senseful approach, but I have four questions:

• Why did you change the sigma tolerance to 3 (default: 6)?
• If I understand your measurements correctly, there can only be positive values with 0 being the optimal value. Correct?
• Is there a chance to have a look at your data?
• Is the use of the capability sixpack recommended or did your customer only wants a Cpk value?

Regards,

Barbara

 

[WKHANNA]

Thank you for the reply, Barbra!

Sounds like a probable senseful approach, but I have four questions:

• Why did you change the sigma tolerance to 3 (default: 6)?

I read elseware here that 3 should be used for this type of application.

• If I understand your measurements correctly, there can only be positive values with 0 being the optimal value. Correct?

Correct.

• Is there a chance to have a look at your data?

Yes.

0.0130
0.0030
0.0090
0.0090
0.0050
0.0075
0.0100
0.0035
0.0100
0.0040
0.0050
0.0090
0.0105
0.0100
0.0090
0.0065
0.0110
0.0045
0.0120
0.0045
0.0090
0.0120
0.0115
0.0030
0.0100
0.0070
0.0060

• Is the use of the capability sixpack recommended or did your customer only wants a Cpk value?

This is the typical format the customer expects.
Bill

 

[bobdoering]

The attached is the technique I use for the analysis of the data. I look for a distribution that actually models the data, and the best fit is a Johnson family distribution. Long term another may fit better, but for sure it is not normal. Compare the p values - the normal distribution is a terrible fit comparatively - and that matters when making decisions on capability. If the model is terrible, how much better is the quality of the conclusion?

 

[WKHANNA]

Bob,

You always seem to come to my rescue here!

I agree the model is not very confidence inspiring!

This is a situation where the customer has spec’d the tolerance on two parts that require a pres fit. One part, the powder-metal gear is supplied and fairly stable relative to size variation. However, the mating shaft that we are machining is under-toleranced by the customer and will not provide an assembly fit that will meet the GD&T spec on the customer’s assy dwg.

To make things even more difficult, the shaft has a flat machined along its length for gear positioning (giving it a “D” shaped cross section). The ID of the gear is also "D" shaped. This characteristic always forces the gear center position away from the shaft center.

In other words, the customer’s own drawing will not yield assemblies that meet their own requirement!

We have to machine the shaft to the highest end of their tolerance to within .0020” total on a part they tolerance to .006 total.

But trying to get them to understand this and change the drawing is like trying to get consensus on the national budget……it just ain’t gonna happen!

So long as they get their 1.33 Cpk on a piece of paper, all is right with the world.

 

[bobdoering]

This is a situation where the customer has spec’d the tolerance on two parts that require a pres fit. One part, the powder-metal gear is supplied and fairly stable relative to size variation. However, the mating shaft that we are machining is under-toleranced by the customer and will not provide an assembly fit that will meet the GD&T spec on the customer’s assy dwg.

To make things even more difficult, the shaft has a flat machined along its length for gear positioning (giving it a “D” shaped cross section). The ID of the gear is also "D" shaped. This characteristic always forces the gear center position away from the shaft center.

In other words, the customer’s own drawing will not yield assemblies that meet their own requirement!

We have to machine the shaft to the highest end of their tolerance to within .0020” total on a part they tolerance to .006 total.

Wow.

And a single "capability index" is supposed to describe all of these conditions? Stunning.

 

[Jim Wynne]

Wow.

And a single "capability index" is supposed to describe all of these conditions? Stunning.

Yet another case of "Don't give us what we asked for--give us what we need."

 

[WKHANNA]

Yet another case of "Don't give us what we asked for--give us what we need."

Actually, I have to give them both for the same price!

 

[Barbara B]

Bill,

I read elseware here that 3 should be used for this type of application.

The default value of 6 will be taken in the calculation of Cp, but for the Cpk 6 will be divided by 2 automatically to get the formulas correctly (see Minitab help files "see also" > "methods and formulas" and/or the attached screenshot example for Cpl).

But from a statistical point of view this kind of analysis doesn't make any sense at all, as the requirements aren't met (like normal distributed measurements).

If the analysis is done with the default value of 6 and the assumption of a normal distribution, the result for Cpk is 0.73 (capability within, see attached picture). That will probably lead to more senseless questions and demands from your customer

You can transform the measurements with a Johnson transformations (use "Transform" in Stat > Quality Tools > Capability Sixpack > Normal in Minitab) and magically the resulting Ppk (overall "capability") is calculated as 1.54 which is above the recommended 1.33.

This is mathematically unreliable and therefore bunk, but maybe it will satisfy your customer. But all in all I agree with bobdoering and Jim Wynne that it would be better to take other approaches for the evaluation of this kind of process.

Regards,

Barbara

 

[WKHANNA]

Thank you again, Barbara.

All this has been extremely helpful to me.

You guys are the best!