As I get to know the PSO requirements I have again been reminded that there is variation in approach to Ppk and Pp. Some places I have worked require us to find the center of a tolerance zone and work Pp and Ppk from it. I did not like this but our SQE insisted that this is how it was to be done (customer may not be right but act like it anyway).
The PSO for DCx says that Pp can not be calculated for unilateral tolerance. This makes sense to me but leaves me with the question "How can we calculate Ppk and be accurate if there is no Pp value?" Take for example an unilateral profile call out of 1.0mm where the process is in spec at .25mm mean and has s of .1. The formula prescribed would look like Ppk=(1.0-0.25)/(3*0.1)=2.5. This seems strange to me since you would be running so close to the lower specification.
I do understand how it would apply to flatness or max radius type call outs but it seems wrong for others such as the one I gave.
What have you done with this stuff and why?
Mark

Any takers on this one?

Do a search for several threads on the subject of cpk and ppk -- I know there are a few. You can calculate Cpk or PPK on a unilateral tolerance.

Do a search for several threads on the subject of cpk and ppk -- I know there are a few. You can calculate Cpk or PPK on a unilateral tolerance.
Thats not the question.

1. Center of tolerance zone: That makes no sense since neither the Pp nor the Ppk is calculated using the "target" value. Also, if the tolerance is unilateral, there IS NO center of the tolerance zone. How does the SQE propose to 'work the Pp and Ppk from the tolerance zone'?

2. You CAN calculate Ppk without a Pp - you just showed it! Profile is unilateral simply because you cannot go less than zero (or perfect fit) - REMEMBER: zero is NOT a "lower specification". Your concern here is that you don't EXCEED 1.0mm. Your process appears acceptable since at 3 std devs greater than the average your highest measurement would be .55mm, a long shot from 1.0mm.

The only problem with these types of measurements is that the distribution is likely not normal, but skewed.

1. Center of tolerance zone: That makes no sense since neither the Pp nor the Ppk is calculated using the "target" value. Also, if the tolerance is unilateral, there IS NO center of the tolerance zone. How does the SQE propose to 'work the Pp and Ppk from the tolerance zone'?

2. You CAN calculate Ppk without a Pp - you just showed it! Profile is unilateral simply because you cannot go less than zero (or perfect fit) - REMEMBER: zero is NOT a "lower specification". Your concern here is that you don't EXCEED 1.0mm. Your process appears acceptable since at 3 std devs greater than the average your highest measurement would be .55mm, a long shot from 1.0mm.

The only problem with these types of measurements is that the distribution is likely not normal, but skewed.
Rob,
With a profile we are dealing with a 3D part which means the surface can in fact be shy or full to where it is supposed to be (nominal). How can I say that the part is in control when we are riding near nominal which in this case means we could fall out of spec by being shy material?
The only options I see is that the print is wrong in how it is defining the surface, the meathod is wrong for determining control, or I have no idea what I am talking about though I think I do. :bonk:
Mark

It sounds like you are talking about an "edge", like the sheared edge of a formed sheet of metal. In that case, it seems to me that the tolerance is indeed bilateral (I might agree that the print is drawn wrong) - i.e. you can be shy up to so much material and you can be full, or in excess, so much material.

How far "shy" can you go before it is rejectable? Whatever that amount is, that is your "other" tolerance boundary. If it is zero, or none, then the Ppk in your example is really 0.833, which is out of tolerance.

Am I missing anything?

Nope that is exactly my point. Things like profile really do have an upper and lower spec limit. Call outs like flatness, concentricity, perpendicularity, parallelism etc.. are really unilateral and as such are the only GD&T that the rule would apply to for not having a Pp. Am I right? If so then back to my original question if I have a Profile call out that is Said on the print to be unilateral meaning the trim edge can be long but not short etc... What do you do and why?
Mark

PLEASE DISREGARD MY PREVIOUS POST (#7)! :o I crossed over the line from thinking "total deviation" to specific measurements. That's why I hate it when customers want capability studies done on GD&T callouts! :frust:

Your TOTAL profile deviation is what you are concerned with - and that has nothing to do with location. The drawing probably shows a profile of a line callout in the feature control frame with a phantom line entirely inside the material zone.

My first comments were correct, in that, if the deviation (or worst/furthest points along the profile line) stay far away from 1.0mm, then you are holding the profile OK. The assumption of the measure is that the location of that most deviant point is still in spec. That is why YOU CANNOT DO CAPABILITY STUDIES ON GD&T FEATURES! {In other threads I've argued against TP}

There is really only two accurate things you can do. 1) Pick a point (or two or three) along the profile and measure its actual location and collect the data for SPC and Ppk studies. Or, 2) Use a GO/NO GO gage and do attribute studies on them (better yet, design an automatic in-process gage station for 100% checking).

I feel your pain. Sorry about the additional confusion I've caused. NOTE: I worked in a metal stamping plant for 10 years, but that was 10 years ago. :mg:

Alright, now I'm confused enough to join in. I'll start with Profile of a Surface (I believe measuring either one is the same except for the 2D vs. 3D thingy). This callout has linear dimensions as basics.

I understand what both Mark and Rob are saying - in a sense. However, when "reporting" a profile you don't report a negative value. Therefore, I'm with Rob and his statement of hating capability studies on GD&T callouts. If my unilateral tolerance zone is 1.0 and I get a linear readings between 0 and 1.0, the largest reading would be my profile. If I get readings of 1.1, 1.2, 1.3 (for discussion's sake), my profile is 1.1, 1.2, 1.3 (resp.). If my readings are less than "zero" (-0.1, -0.2, -0.3) my profile is 1.1, 1.2, 1.3 (resp.). In other words, it would make more sense to me to "study" the actual linear distance callout than it would the profile callout. Am I completely off my rocker??

However, that doesn't address Mark's original question. When I have to submit a capability study on callouts with natural limits I submit the PpU or PpL. I actually used to submit Cpk and Ppk values with zero as the lower spec. limit. I would get calls asking why I submitted with a Ppk of .05 (for the PpL) and would have to explain that I couldn't produce a part with a True Position (or Profile) of less than zero so why were they even looking at that index. I now only supply the PpU (or PpL).

Thats not the question.

This makes sense to me but leaves me with the question "How can we calculate Ppk and be accurate if there is no Pp value?"

Guess I misunderstood the question then. :confused: Forgive me!

ok. forgiven. ;)
Mark

by the way I like sarcasm.
Mark

REMEMBER: zero is NOT a "lower specification".

Is this in fact true?

Why in fact are there unilateral tolerances:

I see 2 types,

1- as said flatness, performance such as leak etc, where the wanted is 0 and you cannot achieve less.
In these cases then the Ppk is obviously the difference of the average from the upper tolerance /3s
2- dimensional call outs that are based on a belief that this is a usefull way of quoting a tolerance, e.g. a pin that must be 1.6 but can be 1.8. Instead of calling out 1.7 +/- 0.1 calling out 1.6 +0.2/-0.0
I can see no difference in practice.
After discussions I have come to believe that the use of KPC's is in fact another way of reducing the tolerance without reducing general tolerances and thus preventing manufacturers from immedeatly saying the tolerances are to tight and raising prices. It sort of sneeks up on them!!

Is this totally of beam?:confused:

Is this in fact true? ...Why in fact are there unilateral tolerances:

I was speaking only in the context of the question concerning profile. Any feature where the adverb "perfectly" can be used, such as perfectly flat, parallel, round, or in the case of profile, perfectly conforming (following the profile as drawn), then you cannot do better than zero. The closer the average gets to zero, the more skewed the distribution, and the less meaningful the capability calculations. And you certainly cannot calculate a Cpk using the "zero" end of the tolerance zone.

I was speaking only in the context of the question concerning profile. Any feature where the adverb "perfectly" can be used, such as perfectly flat, parallel, round, or in the case of profile, perfectly conforming (following the profile as drawn), then you cannot do better than zero. The closer the average gets to zero, the more skewed the distribution, and the less meaningful the capability calculations. And you certainly cannot calculate a Cpk using the "zero" end of the tolerance zone.


I aree perfectly

Mark, I agree that Pp (as it is defined) doesn't make sense for unilateral tolerances but that doesn't mean that the process potential cannot be examined.

For location tolerances like Position, Concentricity if you monitor and control the individual coordinate distributions X & Y you can figure the Potential Ppu by centering the distributions for X & Y and re computing the position deviations then analyze that "more skewed" distribution for its Ppu.

For profile, I have been convinced that profile distributions should be compared to their bilateral boundaries +/- equal, +/- unequal, Zero/+, or Zero/- and reported with the +/- results and predictions of Pp and Ppk based on those +/- results. The problem with reporting the profile in relation to its positive total zone is that it is very difficult to figure the positive zone deviation for distributions compared to unequal bilateral and unilateral described boundaries.

If the retention of process data is rigorous enough you can even figure a potential capability for orientation tolerances by looking for orientation deviation trends and artifically correcting them (if they are adjustable in the process) and then re-figure the orientation deviation with the residual variance.

When reporting a profile, do you put the actual number from the nominal zone, or do you double your actual measured number? It seems to me that the number would have to be doubled, but I could be wrong. :thanx:

I was speaking only in the context of the question concerning profile. Any feature where the adverb "perfectly" can be used, such as perfectly flat, parallel, round, or in the case of profile, perfectly conforming (following the profile as drawn), then you cannot do better than zero. The closer the average gets to zero, the more skewed the distribution, and the less meaningful the capability calculations. And you certainly cannot calculate a Cpk using the "zero" end of the tolerance zone.

Why not? I've calculated Unilateral Cpk's many times before. . . only use one side. . . Remember. . . Zero is a relative point on a plane. . . it is either the desired state. . . or the undesired state

Deuce,

Profile specifications are declared as a total zone and the zone is by default-disposed equal bilateral (+/- one half of the total zone) on either side of the basic profile. If it is explicitly illustrated otherwise with phantom lines and tolerances from the basic profile on the drawing it can be unilateral or unequal bilateral.

If you are going to express your measurement results statistically in relation to the total zone then you have to figure the size of the total zone originating from the basic profile that is needed to contain each measured point (regardless of whether it is plus or minus). You would do that (as you said) by doubling the deviation from basic. The resultant distribution of these "individual total zone deviations" can then be analyzed statistically. They are less normal “skewed” as the mean of the actual +/- deviations approach the target because all of the individual deviations + or – produce a positive total zone.

I recommend against doing that, however, because even though it will work for equal bilateral specifications it won't work for those depicted as unilateral or unequal-bilateral. To figure the individual zone required for each measured deviation of a tolerance depicted unilateral or unequal bilateral the total zone must be expanded as it is defined. That means that a unilateral zone would expand from the basic in the specified direction for each measured point until it captures the deviation. The problem with that is that negative values in a unilateral deviation will not be encountered unless the expansion abruptly changes to a bilateral expansion after the original unilateral zone is exhausted. So the total tolerance zone for a deviation in the opposite direction of the specified unilateral boundary would be twice the deviation outside the zone plus the total zone itself. For a tolerance illustrated unequal bilateral one would have to expand the zone according to the weighted unequal boundaries until they reflect the total zone required to encompass each measured point. I know that this sounds terribly confusing but there is a simpler and more statistically friendly way of analyzing profile distributions.

Profile tolerances are actually +/- limit boundaries for both geometrically simple and complex forms and if you treat the measured points as a group (departure from total form) or as individual locations on a complex form (local assessments) they can be analyzed as +/- from the basic. When you do that it doesn’t matter how the distribution relates to the positive total zone but rather how it relates to bilateral limits or boundaries just like size. You don’t have to double the deviation and the distribution will not change shape (become less normal) as the mean approaches the target!

I recommend that you decide according to the complexity of the form whether you are going to treat the measured points on an individual part as a group or individually and whether you are going to pool all the individual part measurement deviations or look at individual locations separately then analyze the distribution in relation to the displacements from basic as (+/-). If the distribution is naturally skewed as with stampings or moldings etc. and you can’t apply the appropriate distribution functions because there are negative values, just shift the data so it is all positive, analyze it, and shift it back to predict the encroachment on the boundaries. Report your capability results Cp or Pp and Cpk or Ppk in terms of +/- from the basic profile rather than comparing to the “positive total tolerance specified”.

If you peruse the GD&T threads @ cmmtalk.com for profile tolerancing you will see how I recently came to this epiphany.

Paul