Cp/Cpk & Flatness Tolerances

I would opt for b.

Reason. Your tolerance is 0.0015", therefore that is what you have to play with. The plate can be concave, convex, rippled, twisted, saddled or whatever form as long as it is within tolerance (or unless other stipulations are given!)

I have used SPC many times in the past to keep an eye on what is happening with a machining process, flatness included!

The only real problem to deal with is accumulative errors. For example, If you were to test for flatness by using a surface table as your datum, clocking off the top would also include the possible errors of parallelism to the bottom surface.

If you could give a bit more info, maybe we could be of more use! Hope it gets the thread started for you.

I think the questioner should add what he means by 'flatness'

Is it the flatness caused by bending or warping of the specimen part of the measurement, or is it only the flatness between faces?

I can't see any problem with accepting the data points as statistical test points to enter into a SPC program - BUT your method of measurement and definition of 'flatness' is primarily of concern in this question.

John.

His term flatness is in line with ANSI Y14.5 1984 (?) {it's been a while} drawing and dimension interpretation.

If there is some other interpretation than this I can not speak to it. However as stated I have given what I believe to be good information to determine if the flatness reqs are being met.

As to whether the reqs are realistic; I did not engineer the part so.....

When it comes to a), b), c) , the answer is d)

Your dimensional tolerance holds the dimensional restraint. So nominal is a bit misleading of a term here. The part is not to deviate beyond the dimensional limit anywhere on the part and within that limit not more than .0015 from high to low between two theoretical parallel lines as a flatness control.

So to measure the part: Part should be blocked in as few a points as possible on precision blocks upon a precision plate (surface table). A test indicator should traverse the part cross directionally (on a height gage) with a TIR that does not exceed .0015".

In response to the cumulative error there is validity there. For example; I can not check any feature to an accuracy that is greater than the cumulative allowable error of my inspection tools. So a surface plate accurate (flat) to .001 and an indicator accurate to .00075 is no good.

As for SPC. Sure, SPC is fine. If your control point is every part and thus 100%. I do not believe that an accurate Cp/Cpk value can be determined for this process.

Seriously. Given the nature of your process (laminate brazing) and your tolerance (.0015) I do not feel that anything other than 100% inspection would be suitable. I am willing to discuss that from anyone on the production side, but from a quality side it only seems necessary.

Regards,

cr3

I am anxious to here from Milo and others on this one.

I have a general flatness tolerance on a drawing of 0.0015".

My first question would be .. why?

I'd have thought a 'small metal laminate made of copper plates brazed together' would struggle to meet such a tolerance.

IF the flatness is really a problem I'd investigate other means of manufacture.

Step 1...talk to the designer...

Del

Hi Del,

I agree with you on your observation with regards to the achieveable accuracy of brazing laminate copper plates.

Could it be that the o/p has chosen the process himself and made an error of judgement?

Or could it be that the drawing is in fact a rewritten one and the tolerance is a left over from a previous project. I have seen that a lot funnily enough.

Maybe the o/p could comment on this to see if we are right somehow.

uhhh.....I think we all agree it is not an easily acheivable spec given the process. As clearly noted in post #2

The answer that best fits is b.

But for 50 parts and to apply spc data you must specify the frequency (1, 10, or 50 parts). The suggested standard rule for a sample size is 26 pcs and using this you can Cp & Cpk. it looks like with the 50 parts you just have an average or the mean.

You have received much good advice, so I'll add to it, hopefully.

Remember that flatness is a calculated result of the difference (positive or negative) between the high and low points on two imaginary planes of your part. Therefore, the individual readings cannot be controlled, but the calculated value can. Since the calculated value is an absolute value (regardless of whether is goes below zero or not) then it should be calculated and controlled with the low of zero (because that it the least variation possible) and .0015 (the maximum allowable), which would give you an imaginary mean of .00075. Of course, if the process were skewed toward zero in this case, it would be beneficial for you and I would only calculate the CpK based on the high specification value (.0015). The mean and standard deviation used should be derived from data obtained from actual readings (hopefully 30 or more) and measured using methodology agreed between you and your customer. Report it as you see it and take the excellent advice of your other colleagues to go back to your customer if necessary. Otherwise, you may be pleasantly surprised to find that you are in better control than you thought.

I wish you the best.

Since the calculated value is an absolute value (regardless of whether is goes below zero or not) then it should be calculated and controlled with the low of zero (because that it the least variation possible) and .0015 (the maximum allowable), which would give you an imaginary mean of .00075.

You had me until "an imaginary mean of .00075"! The mean is the mean, nothing imaginary about it, and there is nothing that says it would be .00075, or close to zero, or close to .0015; it is whatever it is!

You were on the right track, as were others, but didn't quite make it. See my post #12 for the full explanation.

TexasC...(post #2) is right.

Also is dependent on how the tolerance is applied. ie..MMC, LMC or RFS as well as any modifier applied to the Datum(s) referenced.

For one sided data such as flatness, perpendicularity, concentricity, etc. I would use CPku for CPk.

CPku = (USL-X) / 3σ

(Sorry, I don't know how to put a bar over the X)

I am agreed. I case of any unilateral tolerence (such as Form Tolerences i.e parallelism,roundness,runout etc.) we can consider Cp only which is the total variation in the process. Whereas CPK index is the Centering of the process,and in current case we are not interested in mean of the Total tolerence band which is 0.0015".

As far as determination of exact value is considered if we are using an electronic means of checking it, it will adjust the flatness errors of the surface of plate ,subjected part clamped upon.

Sorry Charley, your answer may be a practical solution, but it does not quite answer Jalipa's question.

JohnV came very close to hitting the nail on the head with his short answer in post #11, but IMHO did not explain why what he said was true and he made one slight, but devastating mistake.

Basically, when you have a one-sided spec like this you have got to look at absolute values of the data points, because it is the deviation which is being measure, not the direction. There really are no negative data points, it is the variation from zero we are interested in, so it matters not if they are concave or convex. Zero deviation is our ideal, and we should look at the process capability (C_p) as the number of standard deviations between zero and the spec limit (SL). Call it an USL and C_pku if you want to, I would just say C_p(SL). But there is a difference here, versus a normal "normal distribution" (pardon the pun!). In a two-sided distribution that is centered on the nominal we want a high percentage of data points (represented by the number of Standard Deviations, SD, or σ) to fall between the USL and the LSL. We say the C_pk=1.0 when 6σ fall within the limits because at one time someone thought that was pretty "good". Now we look for C_pk=1.33, 1.67, sometimes even 2.0. So we think of +/- 3σ as our standard and look for the worst case, upper or lower to define our capability. To obtain an equivalent surety of "goodness", for lack of a better word, (wait, perhaps "conformance"?) in the one-sided case we must look for 6σ to fall between the SL and our ideal, which remember is zero!

Therefore, C_pku = (USL-X) / 3σ is not quite correct. We really want to know:

Cp (SL) = SL / 6σ for an equivalent "good" process where the value is 1.0 or higher.

Numerically, this may work out to be the same as splitting the tolerance in half and calling that midpoint "nominal" and finding the how well 3σ fits between it and the "USL", but if you would examine the data points you would find that they don't really follow a bell-shaped curve centered on this nominal, but rather, in a "good" process, tend to increase in frequency the closer you get to zero, which might look something like a "half-bell", except there is no other "half"! It could also be that the distribution is very tight, but skewed from zero, in which case our distribution does indeed look more "bell-like" but it could not have any true statistical taper, as in a true "bell curve" because the distribution hits that brick wall at zero! Thus, we are left with the only way to evaluate our process capability and conformance to requirements being the Cp (SL) value calculated as shown above.

Oh you engineers. Tsk Tsk.

Nor have you answered his question - though a nice discourse on statistical processes.

_{I have two problems}

1. _{Is SPC valid on for flatness – I have been told not. What is your opinion}
2. _{Interpretation of flatness spec:-}

_{In my ISO spec, it states that "indicated surface is required between two imaginary parallel planes X apart" (X in my case being 0.0015"). So is it}

Answer to #1 = NO for this application

Answer to #2 = YES per ANSI

Answer to ABC = D see my previous and others

Oh you engineers.

cr3

Hi cr3,

The main problem is, we don't know any details! If we had something to go on, then we could give more specific info! How do we know what this little gizmo does? Does it fit between two other pieces in some kind of electrical part? Is it part of a rail gun? The drawing has the details, you can see the other attributes and how they interact, so maybe SPC is needed in some form, maybe 100% part checks are needed! With so little to go on, the thread will lead to us exploring all the possible pros and cons of a process and a part which in essence, none of us know anything about! I would like to hear something more from the OP

Nevermind! It's friday, it's sunny, and my oranges are ripe for picking!!!!

Actually we know enough. The part needs to be flat to .0015" and is a control feature for a laminated/brazed part. This is where us manufacturing guys can get over on you engineering guys.

Nothing else matters. Given the tolerance stated for the process employed, a 100% evaluation will be needed to determine if SPC can be employed.

100% of say 20 pcs will give enough info (methinks) to determine statistical sampling needs. From there you engineers can get busy with all your matha ma matics. But I say you aren't going find an acceptable percentage to keep customer expectations met of less than 100%. Again tolerances and processes are not matrimonial.

cr3

I want an orange!

I tried to send them but they spat the dummy on me!

You seem to be well clued up on this topic so would it be safe to say that after he took his sample and checked them, found them to be all over the place due to an incapable process, then it is up to his subcontractors to implement some kind of control and certification if they are a registered company who accepted the contract? Would it be wise to say that a company who is serious about their work would be registered?

What you guys are overlooking is that you might find a sample that has 100% within tolerance, but has enough variability that it will never be statistically capable! That is what SPC is all about, finding out how to ELIMINATE costly 100% inspection by replacing it with a less costly sampling plan that is statistically valid to insure a process capability that yields 99.99...whetever % good parts.

Charley, even you ought to know that 100% inspection doesn't guarantee 100% acceptable parts! Why? Because the inspection is done by humans (or by machines programmed and operated by humans) and humans make mistakes. Also, because all measuring equipment has a certain amount of error, albeit very small.

Now if a customer requires 100% inspection in addition to a statistically capable process, than that is a "belts and suspenders" approach that might be completely valid if the result of a single out-of-spec part could mean catastrophic failure, endangering life, limb, or very expensive property.

However, unless the production rate or overall quantity is so small, so as to make it impossible to achieve a statistically valid sampling plan, do a capability study, and implement SPC, it would be a serious mistake to overlook the benefits of SPC to improving the process and overall part quality.

I don't know what you mean by "This is where us manufacturing guys can get over on you engineering guys." I started my career as a Manufacturing Engineer after getting my Engineering degree, getting into Product Design later, coming back full circle now where I am designing products for use in manufacturing! (www.carrlane.com) Usually, I got called in to solve a problem on the production line after the other "Manufacturing guys" (operators, supervisors, technicians, toolmakers, mechanics, etc.) threw up their hands and gave up because they could not figure it out! So, I got to sweat and fuss, and cuss, until I figured it out and implemented a fix, either a work-around to get production going again, or a permanent fix to improve the process or product design. Therefore, I consider myself an Engineer AND a "Manufacturing guy"!

As I said earlier, it's all in the details and of them we have only the bare minimum to base our responses on! I too consider myself and Manufacturing and Engineering type of guy who has done the full circle, so when I read a post like this, it tends to get frustrating not having anything to go on!

Anyone for Juice?

I guess I'm not the only one confused with which spec to use for flatness. But what is the answer???? a, b or c?

-U