How can I adjust my Cpk if I purposefully am running higher than average?

[Jake with the Calipers]

Long time lurker, first time poster, hoping someone can set me straight on on setting up SPC to reflect our process accurately.

We are currently using Mitutoyo MeasurLink Realtime and I have two measurements on a part that I have been tasked with setting up an SPC program using standard deviation and I have kind of hit a wall and I am not sure if it is limitations in our software or a lack of knowledge on my part, quite possibly both.

For a bit of background on this part we have always ran higher than nominal on these two dimensions to mitigate any potential issues with a mating part (that we do not manufacture). This is a deliberate decision to help avoid a misfire/mismatch.

As I understand it although our spread is tight (good Cp and Pp values) since we target above center of the tolerance zone our Cpk is going to suffer and not accurately reflect our decision of running the parts higher than average. The typical answer would be make the parts closer to the middle of the tolerance zone to improve Cpk but we really want to remain above that center to avoid any chance of a functional failure. Currently my Cpk for these features is floating around 1.0, would love to get above that magical 1.33 barrier ...

Tolerances:

Feature 1: 0.083" +/- .003" ..... we target/average about 0.0852"

Feature 2: 0.410" +/- .003" ..... we target/average about 0.4124"

In our SPC software I have the shifted our target to 0.0852" and 0.4124" for those features while keeping the LTL and UTL as defined by the print and that seems to have helped slightly but I still don't believe it accurately reflects our process. What should my next steps be, if any, to get a Cpk that better reflects our intentions? Do I need to adjust or shift my LTL and UTL? Or do we need to use a different metric for proving our stability?

I would attach some data but I appear unable to do so as a new user.

 

[Golfman25]

Move your nominal dimension. Make it equal to your target. The standard cpk formulas only work if your process is centered.

 

[Bev D]

OR you could just stop calculating capability indices as they nothing more than statistical alchemy. Aka stupid and meaningless and they provide NO insight to your true process performance as you have discovered…

If you have to provide a number to your Customer just make something up. Mathematical tricks take too much time.

 

[gakiss2]

The Quality Tools that you are tasked to use (SPC) is built on the foundational belief that a stable, controlled process centered on the Nominal drawing specification and exhibiting only 'common cause' variation is your ideal outcome. Since that is not your ideal outcome, SPC signals that you have not achieved what you not looking for. Ihe next question is: What behavior are you looking for? You might start with understanding what is the lowest value that the system will tolerate. From there the simple approach would be discuss setting a specification such as 1.081 Minimum. Probably your SPC software can handle one sided tolerance. Of course you would need to get your customer to agree. The big question being whether or not your customer can tolerate parts that are larger than your current tolerance allows. If there is a maximum size that can be tolerated then use that information to set up a better specification that reflects your true need. Get the OK from you customer then happy SPC ing.

 

[Jake with the Calipers]

I appreciate the feedback, truly.

If this wasn't something being requested by the customer I wouldn't be as concerned as we have made hundreds of thousands of this part and have several other ways of maintaining stability during the process, ways we tried to explain to the customer but they are really fixated on SPC data and a Cpk above 1.33 to prove our stability. I kind of feel like this is being mandated from above by someone that doesn't fully understand our process or even SPC in general and I am just a humble quality manager trying to make it make sense.

I agree that trying to get Cpk for features that do not run at nominal is kind of a fool's errand, even with some mathematical trickery it doesn't truly show our capabilities. I just wanted to do some due diligence before reaching back out to our customer. If they still insist on Cpk I will probably just adjust my limits to put our target values in the center, even if that doesn't reflect the tolerances on the print itself they will still be within those defined tolerances in actuality and I should have a more accurate Cpk for what we are actually doing.

 

[Bev D]

Since that is not your ideal outcome, SPC signals that you have not achieved what you not looking for. Ihe next question is: What behavior are you looking for?

Correction: SPC is not signaling that you don’t have what you are looking for. The Capability index is signaling that. But the capability index is NOT part of SPC (Shewhart would rollover in his grave and Deming certainly belittled this statistical malpractice).

@Jake with the Calipers You might try reading Statistical Alchemy for a pretty cogent argument against capability indices. The paper also has a long list of references to further understand this abomination of the quality engineering discipline.

 

[Golfman25]

Correction: SPC is not signaling that you don’t have what you are looking for. The Capability index is signaling that. But the capability index is NOT part of SPC (Shewhart would rollover in his grave and Deming certainly belittled this statistical malpractice).

@Jake with the Calipers You might try reading Statistical Alchemy for a pretty cogent argument against capability indices. The paper also has a long list of references to further understand this abomination of the quality engineering discipline.

None of that really matters when today's customer "demands" a Cpk of x. It's the world we now live in and the only thing you can really do is attack it from the beginning. Today, before we accept any Cpk requirement, we make sure everything is understood -- we have the proper nominal, tolerance, measuring techniques, etc. That has limited a lot of problems.

 

[Golfman25]

I appreciate the feedback, truly.

If this wasn't something being requested by the customer I wouldn't be as concerned as we have made hundreds of thousands of this part and have several other ways of maintaining stability during the process, ways we tried to explain to the customer but they are really fixated on SPC data and a Cpk above 1.33 to prove our stability. I kind of feel like this is being mandated from above by someone that doesn't fully understand our process or even SPC in general and I am just a humble quality manager trying to make it make sense.

I agree that trying to get Cpk for features that do not run at nominal is kind of a fool's errand, even with some mathematical trickery it doesn't truly show our capabilities. I just wanted to do some due diligence before reaching back out to our customer. If they still insist on Cpk I will probably just adjust my limits to put our target values in the center, even if that doesn't reflect the tolerances on the print itself they will still be within those defined tolerances in actuality and I should have a more accurate Cpk for what we are actually doing.

You stated you're running on one side of the dimension because the part fits better. Just run you numbers with a new nominal target, what you currently run at, and then equalize the upper and lower tolerances. See what you come up with. You can then work with your customer from there.

 

[Miner]

@Jake with the Calipers What is your Cp? Is it greater than 1.33? Then what is the largest amount that you could run above the nominal and still obtain a Cpku = 1.33 then what would be your corresponding Cpkl? Maybe you can meet this customer partway by shifting your mean by a smaller amount yet still giving yourself some margin.

 

[Jake with the Calipers]

You stated you're running on one side of the dimension because the part fits better. Just run you numbers with a new nominal target, what you currently run at, and then equalize the upper and lower tolerances. See what you come up with. You can then work with your customer from there.

Golfman, I did exactly that when I came in this morning and my Cpk greatly improved after doing this, closer to Cpk 3.0 which is outstanding, and I believe is a better correlation to what we are trying to achieve. This isn't a hard requirement by our customer at this time, just a request that could streamline things in the future once we "prove" our process stability.

I fully admit I am new to the SPC game and have had little interaction with it so I appreciate any resources and will check out that Statistical Alchemy resource Bev linked.