Can a Upper Spec Limit be also my Nominal Value in a Capability Analysis Study?

BriGuy0314

Registered
Hi All,

I'm relatively new here and I'm glad I found this community so thank you in advance!

I'm in charge of qualifying a new process at my company which is welding splices. I'm also learning about Process Capability and how to perform a study

Context:
Based one of our requirements we are to cut the welded splice and measure a ratio of a cross sections of welded splices and compare the two. As we want out welds to be dimensionally uniform across the welded splice.

The ratio tol. is set to be 92% or higher, with 100% being the max or Upper Spec Limit.

It is my understanding that the best value I can achieve is to have a 1.00 or 100% uniform cross section between Area 1 and Area 2. I assume that would that make my nominal value 1.00??, which in this case would also be my Upper Spec Limit?

My concern is that this makes for a "skewed" distribution towards the right side, which is backed up by my histogram plot with the majority of my R values centering at 98%.

My other concern is that can we even justify using a Process Capability Study? (Using CpK values to monitor whether the process is capable?)
 

Miner

Forum Moderator
Leader
Admin
If I understand you correctly, 92% is a minimum spec, but 100% appears to be a physical boundary (i.e., no value above 100% is physically possible). This is different from a spec limit. The reverse scenario would be flatness, which has a maximum specification. There is no lower spec, but zero is a physical boundary below which it is physically impossible to go. Treat your capability study as a one-sided minimum tolerance. There is no nominal or upper spec limit. See example below. Notice that only a Ppk lower is calculated.

1716824886958.png
 
Last edited:

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
If a skewed distribution is legitimate, I always use distribution analyzer (variation.com) to determine the correct distribution of the process. It will also determine the capability. Assuming normality is a dramatic statistical fallacy!
 

Bev D

Heretical Statistician
Leader
Super Moderator
If a skewed distribution is legitimate, I always use distribution analyzer (variation.com) to determine the correct distribution of the process. It will also determine the capability. Assuming normality is a dramatic statistical fallacy!
:LOL: Well maybe not dramatic but it is a theoretical model and therefore is not somethign seen in real life. Approximations, yes, but tactual no. Nor is it typical or common…
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
:LOL: Well maybe not dramatic but it is a theoretical model and therefore is not somethign seen in real life. Approximations, yes, but tactual no. Nor is it typical or common…
Well, I always figure a model that fits is a better estimator than one that has no correlation at all! After all, that's kind of the point of a model!
 

Bev D

Heretical Statistician
Leader
Super Moderator
Or as Dr Wheeler points out: models do not generate your data….having some understanding of the shape (no need for a precise model) is helpful. And the time series data is even more insightful.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Yes, time series data is the first step before modelling, for sure. And, true, models do not generate data...but data generates models.

The main point is using statistics for the wrong models yields wrong decisions.
 
Top Bottom