Which one is better to calculate, Cpk or Ppk?

S

Sindwani

1. If Cpk gives only capability(can do) of process and Ppk gives the true performance(actually doing), isn't it correct to use Ppk only ?
2. Also do I need to prove normality of data & control of process (thru c ontrol charts) before calculating Ppk/Cpk?
3. Is Cpk or Ppk calculated over time like based on daily/weekly/monthly data ?

Thanks,
 

Jim Wynne

Leader
Admin
Welcome to the Cove. :bigwave: I moved your post to its own thread for better exposure.

1. If Cpk gives only capability(can do) of process and Ppk gives the true performance(actually doing), isn't it correct to use Ppk only ?
Whether you use Cpk or Ppk doesn't make much difference unless you have customer requirements that dictate one or the other. They're not generally reliable or useful statistics.
2. Also do I need to prove normality of data & control of process (thru c ontrol charts) before calculating Ppk/Cpk?
Cpk assumes a statistically stable process. Not only that, but in order for Cpk to make any sense at all, data must be collected in subgroups, so the idea of calculating Cpk for a 30-piece sample aggregated from production after the fact takes a questionable statistic and makes it totally meaningless. As for Ppk, it's generally considered OK in some quarters to calculate the statistic when stability hasn't been achieved. This has never made any sense to me. Common sense says that you must stabilize the process, if it's possible to do so, before doing anything else.
3. Is Cpk or Ppk calculated over time like based on daily/weekly/monthly data ?
The process may change over time, but if the process is stable and unless there's been a signficant shift in the mean, there isn't much reason to do the calculations again.


Thanks,
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Cpk assumes a statistically stable process.

Both Cpk or Ppk require a bilateral tolerance with the target in the center, the distribution must be normal. When not normal, either it must be transformed or another capability index must be used.

Not all stable processes are normally distributed, so it is important to understand the "expected" distribution.
 
M

Matt33

In order to have a VALID Cpk, the process must be predictable (i.e. in a state of statistical control - free of special causes)
If it is predictable, the mean and variation (sigma) can be used to predict the future based on the past. The past is the control chart (process behavior chart) with control limits.
If the process is predictable, Cpk can be used not only to state what the current capability is, but, unless something changes, what the future capability will be.
 
M

Matt33

One more thought:
If the process is predictable, then the within subgroup variation will be similar to the between subgroup variation.
Therefore, Cpk will be similar to Ppk.
Therefore, it does not matter which index is used since they will be similar to each other.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
One more thought: If the process is predictable, then the within subgroup variation will be similar to the between subgroup variation.

That is, if the process is random, independent and predictable, then the within subgroup variation will be similar to the between subgroup variation.

You really don't know if the variation is predictable (or random and independent) until you determine the sources of the variation. There are always more thanone - which assumes multimodal. Often the measurement or gage error (two different things) can mask the underlying process variation with their variation - which is typically normal, random, and independent. It fools a lot of people. Best bet, do your CNX evaluation of your sources of variation, then determine if all of the variation is either predictable or statistically insignificant. Then, you have answered the question.
 

JuneFoo

Starting to get Involved
May I ask - why we must collect data in sub-group?

If my product produced is piece by piece, can I gather 100 data and just plug in a spread sheet to calculate Cpk or Ppk (I not sure if a lump sum data, Cpk or Ppk is more appropriate)

Thank You.
 

Geoff Cotton

Quite Involved in Discussions
In my world Ppk represents "short" term variation (stability) whereas Cpk represents "long" term variation (capability).

One cannot comment upon the capability of a process unless the process is stable, therefore for SPC purposes the process stability (Ppk) must be determined first.
 

Bev D

Heretical Statistician
Leader
Super Moderator
In my world Ppk represents "short" term variation (stability) whereas Cpk represents "long" term variation (capability).

One cannot comment upon the capability of a process unless the process is stable, therefore for SPC purposes the process stability (Ppk) must be determined first.

Sorry Geoff, but Ppk (even when using it as a short term calculation, typically in preproduction) cannot determine stability. Stability by definition is a time based assessment. So the typical manner of determining stability is with a control chart.

The confusing thing with Ppk for those customers that require it during pre-production is that the pre-production volumes are low and the supplier doesn't run at rate so a control chart is not feasible. These Customers accept Ppk calculations from a small run with the assumption that this gives some assurance of the capability of the process once it gets into full scale production and is stabilized...
 
Top Bottom