Cpk and Ppk for different Countries/Standards

U

Uchiha

Hello all,

I have a question about process capability indexes:
I believe that the definitions of Cpk and Ppk are different depending on the countries (USA, Germany, Japan, France...). Is it possible to have a clear and simple explanation about these differences? (Especially if it is only the simple long vs. short term consideration)
Also by saying "countries" I believe this especially include national standards... What about international standards then? I had the chance to take a look at the ISO 22514-2 (new for the old ISO 21747) and I got even more confused as it does not seem that it deals with the long vs. short term issue but has a completely different view...
Can anyone clarify a bit more?

Thank you very much and I do apologize if this issue was already discussed before
:thanx:
 
W

Wilderness Woody

I don't have the standard available... If you don't get any bites here at the Cove, you could reach out to the appropriate ISO Technical Committee TC69 SC4

TC69 SC4 contact : [email protected]
 

Bev D

Heretical Statistician
Leader
Super Moderator
I don't have access to ISO 22514-2 and I'm sure many don't either. could you describe your new confusion from this standard?

as for the short vs long term we have covered this here in many threads. i don't think this is a 'country' thing so much as it is an industry or Customer specific thing.

Originally there was only one way. Cpk was the name of the index and it was long term capability using the total standard deviation. There was no short term capability. (L.P. Sullivan "Reducing Variability - a new approach to quality" Quality Progress July 1984. Victor Kane, "Process Capability Indices" Journal of Quality Technology January 1986)

later, different authors and companies added the 'short term' capability index. which formula was used was dependent on how much data and what type was available.

Then companies just had to 'customize' it for themselves; because why would you want clarity? <sarcasm>

most software has Cpk using the within subgroup SD and Ppk using the total SD. generically, some customers will use Cpk as it was first described - long term using the total SD

Some have pointed out that if you don't subgroup (eg during early production or even pre-production) you can estimate short term capability by using oen lot of your raw materials, one piece of each equipment and 1 operator and create the parts in a relatively short time, you would need to use the total standard deviation - because you don't have subgropus -then you have calcualted short term capability using Ppk formula...

the key is to determine whether you need to calculate long term or short term then use the formula that is appropriate for the data you have and can get.
 
Last edited:
U

Uchiha

I don't have access to ISO 22514-2 and I'm sure many don't either. could you describe your new confusion from this standard?

I got the impression that the definition given there was that Ppk is the usual long term capability index, and it can be calculated for any process... And it becomes Cpk (same formulas, only the name changes) when the process is under statistical control...

By the way, that same standard talks about some cases where the median is to be used instead of the average in the capability index formula... Anyone has ever used the median and not the average in the capability indexes' formulas? And when is the median used and not the average? And what is the difference in using one and not the other?
 

Bev D

Heretical Statistician
Leader
Super Moderator
if that is what the standard says (can you quote the relavent section?) then the standard is incorrect.

No capability index is to be calculated if the process is not in statistical control. period.

that said there are 2 seeming exceptions that aren't really exceptions:

1. as expalined in my first response, in the early stages of production or protyping a short term capability can be calculated using one source of everything over a very short period of time (there typically isn't enough data to determine statistical control in these cases) and since there is no subgrouping the total standard deviation is used. some manuals (Ford is notorious for this) refer to this as Ppk.
2. a process may be stable but not homogenous. If one uses the traditional Shewhart charts the process will appear to be out of control. If one uses rational subgrouping and the appropriate chart, the process will be in control. In these cases Cpk (using within subgroup SD) will be rather small and Ppk (usign total SD) will be larger.

I believe there are indices that use the median but I must confess that I never use capability indices becuase they are essentially useless and meaningless and therefore a waste of time...
 
Last edited:
U

Uchiha

if that is what the standard says (can you quote the relavent section?) then the standard is incorrect.

The standard says something like this:
The process performance index is Pp = (U-L)/Delta

[...]

If a process is under statistical control, a capability index can be assigned. The formula is the same as the Pp.

!!!

I believe there are indices that use the median but I must confess that I never use capability indices becuase they are essentially useless and meaningless and therefore a waste of time...

Can you explain more why? Also what do you use then?
 

Bev D

Heretical Statistician
Leader
Super Moderator
The standard says something like this:
The process performance index is Pp = (U-L)/Delta

[...]


well we would need an actual quote to assess what the standard says and not just your paraphrase of it...in any event the formula you provided is not the accepted formula for Pp in the vast majority of the literature and other governing body / major industry company manuals or leading statistical software packages. in these documents, delta is replaced by SD, where SD is the total standard deviation of the individual values (and not the average within subgroup standard deviation. I don't know if you remembered incorrectly or if the standard stated it that way in error or as a proposed alternative to using a standard deviation.

I am not sure of this statement:

If a process is under statistical control, a capability index can be assigned. The formula is the same as the Pp.

Although the quoted has issues, I am not sure why this would cause confusion. the statemetn is 'correct' in that if the process is in statistical control, one can assign (calculate) the capability index.


Can you explain more why? Also what do you use then?

Well, it's just insensical to describe variation as a single number as if it were accurate and true. These indices tell you very little about the 'capability' of yoru process. it is really just a lazy manager's need for a basic yes/no answer. The index isn't actionable. once it's bad you go back to plotting your data and understanding the original structure of the data collection to figure ou twhat to do next. If it's good but the structure was bad you are fooled into complacency as the index can't tell you the structure. but a time series plot of the data can. so why not just use that?

other things that just make this worse:
The common indices assume a Normal distribution which often doesn't exist. most people don't understand and/or comply with the requisite analysis structures: statistical control first, samples are representative, small samples are horribly in accurate and imprecise. oen need look no further than thsi forum for evidence of the abuse and mis-use of these indices.

What do I use? a multi-vari chart of the data. multi-vari shows me within positional (if applicable), piece to piece, time time in a lot, lot to lot, operator to operator, raw material lot to lot, equipment to equipment, time to time (days weeks, months, seasons). Plot the individual data poitns against the spec limits. I dont' have to do any math. just collect the data, plot the data, look at the data and THINK about the data.
 
Top Bottom