OK,
I have been struggling with this for quite sometime and am looking for feedback and suggestions. We perfrom in-process inspection and enter the data for a characteristic into our SPC program, Hertzler. The data is entered in subgroups of five. The Cpk and Cp is then calculated based on 25 subgroups. Now, I have the inspectors enter all readings taken, evaluate the results, i.e. look for trends, runs, etc, and then adjust the process accordingly. The problem lies in the fact that by entering data that is to the high or low side of the specification drives the Cpk and Cp numbers down, less than 1.33Cpk. I am suppose to be maintaining a Cpk of 1.33 or better, but due to entering in real world data, my values stink. Now I understand that SPC is a tool to moniotor and adjust the process to ensure that product remains within specification, but I am not meeting capability requirments. What am I doing wrong, or am I doing wrong? As a note, I changed the number of sub-groups that values are caluated against to see what effect it would have, and it was if anything detrimetal to the values. :mad:
Rich

I wonder if your software is set up correctly? Is it also automatically recalculating control limits every X subgroups? If so, there should be a way to lock the limits.

Also Cp and Cpk are the results of good control chart practice, to get good numbers, you have to have use control charts effectively.

If you have data points out of spec, capability will always be poor.

Here is how I try to explain SPC to operators.

The control chart is like a highway. The center line is your target. Try to drive right on it.

The white lines are the control limits. Start to steer back to the center when you hit them.

The ditch is the customer spec. If you take no action at the white lines and wind up in the ditch, you are in a world of hurt (and it is your own fault).

Cp is how wide your truck is compared to the white lines. A big wide truck is a lot harder to drive through a tunnel than a narrow one.

Cpk is how close your truck is to either white line. A narrow truck could be driven closer to the sides of the road than a wide truck.

For best results, stay on the center line and work on your truck to make it narrower.

Making the truck narrower leads to reducing variation and that keeps us all employed.

Perhaps you should de-emphasize Cp and Cpk tracking. Just work to react to the chart trends and Cp and Cpk will become OK.

Another thing often overlooked are the spec limits. Are they realistic? Can they be relaxed? Sometimes specs get set for no good reason.


I'm sure more learned responses will be showing up shortly.

RG Ohidy :

I fully agree with the excellent contribution of Caster (!good example on the road¡). Nothing to add.
Just to emphasize the fact that the process readings have to be analyzed on the floor by the people that is making posible the process. Do not be in a hurry to calculate capability unless you have a controled/stable process.

Qualiman

RG Ohidy
I changed the number of sub-groups that values are caluated against to see what effect it would have, and it was if anything detrimetal to the values.
If you are telling us about the capability index, if the pp or ppk is used no change happen whatsoever (they use total variation estimate, so don't depend of the subgrouping), if Cp or Cpk is used, there can be a small change because of the change on the calculations but, the variation estimate is affected by a correction by the sample size (changing it from subgroup stdev to individual stdev) so it must be the same (aprox) if you didn't make wrong subgrouping.

I don't know your values of your stdev but I can imaginate total variation and within valriation estimates almost equal, if so maybe you could concider the use of IX MR chart, don't use subgroup of 5 just because it's the usual recommendation each process has it's individualities.

The problem lies in the fact that by entering data that is to the high or low side of the specification

I agree with Caster that if you are at the borderline, you will obtain a small value, and I must add if you are at both borderlines (specs) the case is worse (with one side you could have at least a good Cp, but at both specs..., It's odd, is there a reason of that behabiur?

I understand that SPC is a tool to moniotor and adjust the process to ensure that product remains within specification,
Not exactly, just reduce variation, if you see the traditional rules for SPC, they don't add out of specs, it's taken into account that you met the sepcs requirement.

:agree1: Caster, excellent analogie

Another thing often overlooked are the spec limits. Are they realistic? Can they be relaxed? Sometimes specs get set for no good reason.
That's my own tauma. :biglaugh:

I don't personally do a lot with capability indeces. First, primarily due to dealing with non-manufacturing data, second, because they cause too many problems. The capability indeces are (to my knowledge) supposed to reflect long-term performance of the process. Now I don't know if this is in violation of ISO or other requirements, but if I was requested to do capability, I would put in the current baseline average and control limits and use them as the basis. If there are points outside the control limits (or other trend signals), the process capability is undefined.

So I think by rerunning the index continually, you are causing more problems than you are solving.

OK,
I have been struggling with this for quite sometime and am looking for feedback and suggestions. We perfrom in-process inspection and enter the data for a characteristic into our SPC program, Hertzler. The data is entered in subgroups of five. The Cpk and Cp is then calculated based on 25 subgroups. The problem lies in the fact that by entering data that is to the high or low side of the specification drives the Cpk and Cp numbers down, less than 1.33Cpk. I am suppose to be maintaining a Cpk of 1.33 or better, but due to entering in real world data, my values stink.

Can you tell me what kind of variable you are measuring? When you look at a control chart of the last 25 subgroups of size 5, are all the points within the 3 sigma control limits? Again, looking at the last 25 subgroups only: Is the std deviation in each subgroup, and then across all the subgroups (an average of the std devs) what you would expect it to be? In other words, is the process centered with the specification limits? You mentioned to the "high" or "low" side of the spec. Can you center the process? Have you done a gage study on the in-process inspection tool? (I am assuming this is a product characteristic you are measuring, rather than a process characteristic).

Help me understand a bit more, and I might be able to steer you in a slightly new direction (keeping up with the driving analogy of previous posts.)

--Jodi :D

After scanning these posts, I think that there is more than one statistical analysis technique needed here, and that you may have 3 choices based on what I have read. . . .

1) Look at the range chart as well as the X-Bar chart. is the process erratic?

2) Generate a histogram of your data and look at it. Is it Normal (Some believe that there is no such thing as a normal distribution), is it skewed indicating some special cause of variation?

3) choices if you are in control, have a less than 1.33 Cpk, and have optimized your process:

a) Try to get the tolerance changed (Most customers have already added their allowance into the spec you have to work with. If you don't tell SPC the truth, it will not work)

b) 100% sort/verify (YUK!)

c) improve/change the process to meet capability.

SPC is also a tool to use to determine not only is the process capable, but it is an indicator tool for process improvement.

One additional note, make sure that your measurement system is not the element of the process inducing excess variation and noise into the process..

I don't personally do a lot with capability indeces. ..........because they cause too many problems.........
..........capability indeces are (to my knowledge) supposed to reflect long-term performance of the process.....
........So I think by rerunning the index continually, you are causing more problems than you are solving.
An idea we were playing with was to calculate the capability index once a month and plot them on a control chart. We would then only react if the chart told us to. Unfortuately, we never got that far along before SPC was abandonded.

We also had a customer who demanded monthy capability data. We doubted anyone ever looked at it, so we started chaging the data. We got up to Cpk of 200 over a period of several months, then down to zero and finally into negative Cpks (out of spec). Each FAX had a statement to contact us for an explanation if there was a concern.

Finally one of our sales people found the very nice lady who carefully collected all our FAXs and filed then away with thousands of pages of other supplier data that was never examined. We soon after got that request changed to "let us know if there is a problem".

Lots of good thoughts here on the subject. I'll get started and see what the results bear.
With regads to allowable toleranace, its an automotive application and they will not grant any increase in tolerance. Been there, done that.
As for my comment on knowing that SPC is a tool to prevent out of specification parts, I would further add that I understand that its a tool to eliminate variation "to prevent product from going out of specification". My original statement was simplistic.
As for the limits being applied, I used the constants table, but roughly the tolerance is about 70% of the full tolerance.
I'll look over the data against the input provided and let you know what I learn. I am hazarding to guess its the way the numbers are crunched in the software, as the process is fairly stable.
Thanks,
Rich

As for the limits being applied, I used the constants table, but roughly the tolerance is about 70% of the full tolerance.
How is your study going?

From this quote are you using 70% of the print tolerance in your capability calculations?

We got into quite a mess last year with something similar.

We had a red/yellow/green type of tolerance which used 67% of the print spec. Looking back, it is hard to understand what we hoped to accomplish.

When we changed to SPC, we ended up with a mix of full specs and 67% tolerance. The result was some lost face due to some smart people pointing out that "SPC is set up wrong".

I think full print tolerances should be used for capability calcualtions, otherwise the tolerance is unrealistically tight.

Regards

Caster