Cpk for attribute data? Performing SPC on capacitors - Capacitance measurement

[M Caruso]

A customer has given us new requirements for performing SPC on capacitors we will be making for them. The requirements ask for us to collect data and calculate our Cpk. No big deal I think, until I read the requirements more closely. They want data collected and Cpk calculated on a capacitance measurement; I can do that. BUT, then then want data and Cpk calculated for whether the capacitors meets the voltage requirement, pass/fail. Can Cpk be calculated for this?

[Rob Nix]

Since the Cpk calculation requires a "standard deviation", which in turn requires variable data, the answer is NO, you cannot calculate Cpk for attritbute data.

The problem is, and I see it often, is that customers do a lot of "cut-and-paste" when establishing requirements (specs) for new products. We sometimes receive specs for a new testing machine (e.g., torque or pressure decay) that specify a Cpk < 1.67. All it does is test. GR&R may apply; but Cpk does not. So, what do we do?

Bring it to the customer's attention! Questions regarding requirements should be brought up during contract review. But even if the problem is discovered afterward, explain the situation to the customer - they may very well agree with you that it was an "oops". At the very least, they should explain to you what their true intention is.

I hope this helps a little.

[Al Rosen]

In Reply to Parent Post by M Caruso

A customer has given us new requirements for performing SPC on capacitors we will be making for them. The requirements ask for us to collect data and calculate our Cpk. No big deal I think, until I read the requirements more closely. They want data collected and Cpk calculated on a capacitance measurement; I can do that. BUT, then then want data and Cpk calculated for whether the capacitors meets the voltage requirement, pass/fail. Can Cpk be calculated for this?

Could it be that they want it for the leakage current at the rated voltage?

[Tim Folkerts]

I agree with Rob - the requirement as quoted makes no sense and needs to be discussed with the customer.

To go one step further, perhaps you could suggest an improvement. For example, perhaps the cap should work up through 250 V. The pass/fail test might be to simply connect the cap to 250 V and see that it works - that would not work for a Cpk calculation. You could test many caps this way and find a percent defective and then state something like "this defect rate is equivalent to Cpk = x (assuming normally distibuted data)."

Alternatively, you might destructively test the caps by ramping up the voltage slowly until they do fail. That would give you continuous variable that could be used to calculate a Cpk value. It would be a more difficult test and be destructive, but it would produce a Cpk number (if that is truly important to the customer).

Pass/Fail voltage testing:
+ simple
- large sample size required
- gives only a "pseudo Cpk" value

Ramped voltage testing
+ smaller sample size
+ produces a true Cpk
- destructive
- more complicated test.

Just my $0.02.


Tim F

[roland_lu]

Pass/Fail voltage testing:
+ simple
- large sample size required
- gives only a "pseudo Cpk" value

Ramped voltage testing
+ smaller sample size
+ produces a true Cpk
- destructive
- more complicated test.

Just my $0.02.


Tim F[/QUOTE]

Great suggestion Indeed!

I have had another case. One of the parts we make is automotive fuel tank support straps. The drawing calls for the spot weld strength's mean minus 3 times deviation > certain amount of Newton. The part is a safety item, and the drawing identifies the spot weld strength as a critical characteristic and we need to monitor it through SPC.

The problem is we could never get weld strength readings and its deviation, for the strap material broke before the weld did when we destruct the samples. What we got is material's tensile stength and its variation. In PPAP package we pointed it out, our customer still asked us to report the "characteristic's" Cpk. For us, the weld strength is just a pass/fail case (actually never failed). Eventually we submitted the strap's strength Cpk instead of weld stength, customer did not challenge it. So until today we keeps monitoring through 'SPC' the strap tensile strength.

Any comments?

[M Caruso]

Thank you all for your replies. I just needed to make sure that some new statistical method wasn't introduced since I finished school only 10 years ago. I was looking ahead during my schooling and decided to minor in Statistics to complement my Engineering degree. This actually was a big factor in getting my first job right out of college. Anyway.....

From reviewing the specification that was submitted with the requirements, all that is required is that SPC be maintained on major characteristics. I believe someone identified these major characteristics and boilerplated the requirements without reviewing. All requirements repeatedly call for a cert with Cpk and data.
I am going to be contacting the customer and suggest that we keep a p chart on the working voltage characteristic.
We already do 100% testing for each characteristic the customer is asking for us to track. We do it in a pass/fail manner, but some things such as capacitance can be measured and recorded, however the voltage test is strictly a pass/fail test. To change the nature of the voltage test would impact the cost to the customer significantly. Intentionally destroying units is usually done as a proof of concept or during qualification testing but not on a regular basis.
As I read the customers specification, I question why the customer did not consult us first anyway. Their specification has a paragraph that states, "Suppliers shall concur with the use of SPC as described in the individual SPC detailed requirement." Anyway, I will get this straightened out so that all will be happy here and at our customer.

[Bill Ryan - 2007]

In Reply to Parent Post by roland_lu

Great suggestion Indeed!

I have had another case. One of the parts we make is automotive fuel tank support straps. The drawing calls for the spot weld strength's mean minus 3 times deviation > certain amount of Newton. The part is a safety item, and the drawing identifies the spot weld strength as a critical characteristic and we need to monitor it through SPC.

The problem is we could never get weld strength readings and its deviation, for the strap material broke before the weld did when we destruct the samples. What we got is material's tensile stength and its variation. In PPAP package we pointed it out, our customer still asked us to report the "characteristic's" Cpk. For us, the weld strength is just a pass/fail case (actually never failed). Eventually we submitted the strap's strength Cpk instead of weld stength, customer did not challenge it. So until today we keeps monitoring through 'SPC' the strap tensile strength.

Any comments?

I wish I could give you an idea but I'm just not well-versed enough in weld strength testing.

FWIW-
My comment is I think you're on a slippery slope. If a weld should fail in the field (the strap stays in one piece) and your customer comes back at you, you don't have a leg to stand on. This is a "CC" and customers get amnesia when suppliers don't have approved documentation. I know I would be very uneasy about this.

[roland_lu]

In Reply to Parent Post by Bill Ryan

I wish I could give you an idea but I'm just not well-versed enough in weld strength testing.

FWIW-
My comment is I think you're on a slippery slope. If a weld should fail in the field (the strap stays in one piece) and your customer comes back at you, you don't have a leg to stand on. This is a "CC" and customers get amnesia when suppliers don't have approved documentation. I know I would be very uneasy about this.

Thank you very much for your comments.

The thing is the test method is specified by the customer, and the weld strength is better than the strap strength with no exception so far for over 1200 samples, and the strap strength is close to 2 times of the required weld strength, everybody except me feels comfortable. We can establish that the weld strength average is good, but no way we could know the its actual number and its variation with the specific test method. I am sure the weld will not fail, but I am not sure if we did it right.

I hope that somebody could give me some ideas.