Hi'all,
I've just inherited a new department and in order to try to start to move forward, I'm interested in trying to apply pre-control charting to guide in-process inspection / adjustment decisions.

I think it would suit an application where an operator is running a bank of 5 machines.
I'm told that we tried SPC ("didn't work - batch sizes too small / "Quality" depends on how good the set-up is")There also seems to be a mindset of if it's in tolerance it must be OK (reduction in variation is not viewed as continuous improvement - unfortunately)

Current inspection ssems to be a "finger in the air job" i.e. no data to support when to inspect / when to adjust.

To get to the point, does anybody know of a sampling plan to use a starting point, rather than trying to invent my own.

Cheers,
Don

Don,

There are pre-control plans available but I would caution you that they all start with a caveat that the process first has to be in statistical control before you use the pre-control techniques. If you have not been able to implement SPC successfully it will be difficult to move to pre-control. A Cpk greater than 2.0 is not unusual as a minimum pre-control requirement.

Most plans are based on three zones. The green zone is 50% of tolerance and centered around the nominal dimension, the yellow zone adds another 25%, the balance is the red zone. There is a text, 'Modern Methods for Control and Involvement', ISBN 0-471-87696-X, with a brief discription of the technique or you might try papers by Dorian Shainin (1965) or N. R. Brown (1966). I also highly recommend you look at 'Statistical Quality Control' by Grant and Leavenworth, ISBN 0-07-024162-7. The authors present a number of reason why you should not use pre-control and in fact go so far as to state "The best advice we can give is to recommend almost any technique discussed in this book in preference to precontrol. The pattern of variation is a characteristic of the process and should never be confused with design tolerances."

Hi Rick,
Many thanks for the information, I'll certainly be having a look at the Grant & Leavenworth book.

I was under the impression that pre-control may be ok with Cp between 1 & 1.33 (the limited studies that I've done so far indicate that we have this basic capability)
If Cp of 2 or greater is required, I need to think again :-<

Hi Don,

I know what you are looking at. I also had the beauty of walking into a new department and it is still difficult to teach an old dog new tricks. I do work with a cp of 1-1.33 and it works okay. Soon I will start to introduce a system simular as the 50% / 25% zone. Only, I will use 1, 2 and 3 sigma. 2 Sigma being a yellow zone. If I have not been killed by then I will post the results....

What conditions have to to be in place to define "when" you can use a pre-control chart on an ongoing process?
Jim Jakosh_ Grand Rapids, Mi

Jim,

Can I assume that your currect process is in control?

Ifso, what would be the need or benefit of adding a pre-control chart?

Maybe some more details will help us!:)

PreControl will work well only when you already have a good Process Capability. To qualify a process, you will need five points in a row in the Green zone ( middle 50% of specs) on the precontrol chart. This would happen only if you have a good process capability with the process centered.

A precontrol chart would monitor whether your process maintains the low variation and holds on to the centre as you sample out of it.

I had seen a short article on this and was wondering if anyone could point me in a direction for some good resources on this? Thanks in advance!! :)

If this zip file works okay here is a good article. Also, in Juran's Handbook version 4 there is a good write-up on it.

Thanx Mike for recommending that article. Its titled "The Power of PRE-Control".

Here is the source link:
http://www.symphonytech.com/articles/precontrol.htm

- Atul.

This article is good

http://www.stats.uwaterloo.ca/~shsteine/papers/pre.pdf

I would add something I beleive, precontrol is good "if"

* You have not a direct control of the variable's variation
* The quality of the product is not affected in any way by the variation inside of the specification limits (like a level of a water tank)
* It's too much important to fit into the specs
* You are just starting, SPC control charts beleibe that the process in specs (you shall not tink as a pattern OUT OF SPECS), so your process could be out of specs and in control, ones you centered the variation inside of the specs, try SPC control charts to reduce variation.


Read Vol 32, #4, Oct 2000 Journal of Quality Technology
"Controversies and Contradictions in Statistical Process Control"
by William H. Woodall

:bigwave:

The qualification of the process with 5 points in a row in the green (central 50% of specs) zone would work only if you already have a good Cpk.

Like Shewhart control charts PreControl is good only as a monitoring and maintenance tool.

I just skimmed this article, but I have a few opinions. The writer writes in a very arrogant, abrasive style that I find hard to tolerate for long. He trashes concepts as 100% "nonsense" that I believe, used properly, can do lots of good for some organizations. I think he feels he must resort to sensationalism in the form of attacking some widely-accepted tools and the works of some quite knowledgable and famous people like Juran and Drucker as a way to get attention. Heck, I guess I could get some attention, too, for an article if I titled it "Deming is an Idiot"! (No, I don't believe that -- just an example!) I could show a few examples where something Deming taught, said, or supported did not work so well for a certain company or situation and therefore conclude Deming's works or techniques are -- say -- "worthless" (I didn't want to steal the word "nonsense").

IMO, no one person has all the answers, no one system or tool is the best in every situation, few tools are totally useless in every situation, and anyone who disagrees is... AN IDIOT (in keeping with the spirit of this guy's writing style)!;)

Any other resources or comments appreciated.

There is very little info available on Pre-Control and other techniques developed by Dorian Shainin. There seem to be a lot of controversies about statistical validity of Pre-Control. Read: http://www.pqsystems.com/eline/v200201/analysis.htm

One (and probably only) good resource is the book "World Class Quality" by Keki Bhote.

- Atul

I think the article you cited misses a point.

There is no real way to tell where the process is in control, which means you cannot tell where it is centered and cannot make an accurate measurement of the spread of the process. This lack of information about the process complicates understanding of how the process characteristics interrelate to affect the output of the process.

This is why you should never consider a pre-control chart on a process where stability and capability are not known.

Pre-control is simply a tool to make "on the floor" process checks quicker and easier in cases where you have a reliable process with a high capability. A good case would be where you have a process which has a capability index of over 2. The X-bar & R chart you have been using could be replaced with a pre-control chart. This could help in time saved in taking measurements and recording on the chart and not needing a specially trained employee for SPC. If the process is stable with a 2+ Cpk, all you need to know is if the process is changing.

Although the pre-control is not collecting data it is statistically based on the data collected and analyzed from the control chart. That is why the process needs to be in control and stable. The pre-control chart is then centered on the spec with each side divided in half. Based on a Cpk of 1.0, statistically the data collected from the same process will have 86% of the data points within the two middle areas and 7% in each of the others. From this point on, probability dictates that only 1 in 7 samples will fall in one of those zones. 2 falling within the same zone (1 in 49) signals something has changed in the process.

If the process becomes unstable or for continuous improvement studies, you should revert to the X-bar & R chart or some other tool.

Pre-control is not meant for every process but it gives a nice quick "operator manageable" control when you need it.

Just my opinion.

Dave

D.Scott said:
Although the pre-control is not collecting data it is statistically based on the data collected and analyzed from the control chart. Dave
You can, if you want, collect data with pre-control. Just have the operator record the 2 measured values he uses. If needed, a run-chart can be created from this or other statistical calculations made.

Pre-control is a tool, like any tool it can be properly used or misused, and, like many tools, even when it is not ideal it might be better than nothing if it is the only tool available (pliers can sometimes substitute for a ratchet and socket). Many places that use no statistical tools at all would be well served to use pre-control if they could only pick one tool. But sometimes I think it is too simple for some elitists to like it, so they resort to the old tactics of blanket denounciations and name-calling. What really matters? Results!

Just have the operator record the 2 measured values he uses.

I'm not sure that would be effective Mike. The operator isn't always using 2 samples so some of the entries would have to be individual which would lose the range calculation on an x-bar & r. Using an X & moving range chart for single entries would be influenced by the result of the first sample being out of the center zone.

I see where you are going but I think the pre-control is designed for quick, non-analytical evaluation. If you wanted to collect data, why would you bother with pre-control? Just use the X-bar & R.

Dave

Pre-control does not estimate stability and capability in the way normal SPC does. It has 'thumb-rules' for telling if the precess is stable and in control, for example, out of the two units sampled, if both are in Green zone or if one is in green and one in yellow, the process is said to be in control, both in yellow zone or one in red zone is a clear signal to stop the process and investigate cause of variation.

I agree with Dave that it is a quicker and easier-to-understand-and-react method on the shopfloor. During initial process characterization, 5 consecutive readings in green zone would ensure Cpk >=1.33 (as calculated in conventional SPC). For Cpk value of less than 1, pre-control would stop the process so frequently that it would become a nuisance (!). Also pre-control allows reduced frequency of sampling for a process that remains stable and in control over longer periods (equivalent of Cpk=2?). I think this may be the reason why it is said to be more suitable to 'monitor' the process which is already 'capable'.
??

I have used PreControl charts and found them to be a useful tool. I've worked through the derivation for the rules and they are statistically valid. They are fast to set up, which manufacturing appreciates.

To answer Dave Scotts's question:
If you wanted to collect data, why would you bother with pre-control? Just use the X-bar & R.

To create a control chart you need data that came from an in control process. But wait a minute, that's why we needed a control chart! PreControl helps break that cycle. "Pre" infers that something else follows and PreControl wasn't meant to be permanent.

I feel PreControl's weakness is that it needs good product specs. I used to know a Product Engineer (before he suddenly left) that set spec by taking +- 3 sigma. When the process improved he would update his specs!

A few years ago, it was normal to know a lot about product specs, without much process knowlege. This made PreControl very useful.

Happy Holidays,

Tom

An extract from ASQ publication http://www.asq.org/pub/jqt/past/vol32_issue4/qtec-341.pdf :

It is difficult to make meaningful comparisons between
pre-control and control charts since there are typically no clear statistical objectives or assumptions made for pre-control.Upon careful study,Ledolter and Swersey (1997a) identify specific situations for which pre-control has alue,but conclude in general that the method is not an adequate substitute for statistical control charts.If one follows the view of Deming and others that models should not be used to determine statistical properties,then it becomes impossible to argue effectively against pre-control or any other such method.

The article is excelent, as I mentioned early in this Thread, but to have a full picture of the article you may be interested in this link.

http://www.asq.org/pub/jqt/past/vol32_issue4/

About the extract, It's obvius that William H. Woodall is convinced (as may statisticians are), that there is no other way than SPC, and that if you apply SPC all statistic asumptions are met, so the statistic vality is in order, supporting this on the central limit theory (Donald Wheeler, said that "Even thougth the Central Limit Theorem applies to the subgroup averages, it is not the reason why control charts work....does not apply to subgroup ranges) or with data transformations (that almost always make the data difficult to analize).

The reality is that there is almost no process that behaves in a "normal way = Gausian", slight departure of the Gausian behabiur does not affect really to the statistics, but as the departure get bigger, the statistics may not represent the data.

I tink that Shewhart as his book title: "Economic Control of Quality of Manufactured Product", didn´t look for statistic vality of the SPC, but to a "Economic Control", He himself, altought discussed 5 different criteria for detecting a lack of control, he never used more than one "When a point falls outside of control limit", I asked my self why?, the reason may be
that he realized that there is not "true control limits", but a tool with the objective to control the process , Don Wheeler shows in his book "Advanced topics in Statistical Process Control", that control limits work even when the distribution is non-normal (is you take rule "out of Control limits").

I understand that SPC is a better tool (the best), but also there are cases where Precontrol could be an usefull tool, I realy don't understand why the statisticians are against using "out of specs" as a pattern, I understand that is no statistic pattern but to control the process within a specific range may be nedded, It looks like SPC take in account that the process is into SPECS and there is no worry about geting out of them, but there are cases where and advice that the process is geting out of specs could help to control the process.

As may last post said, there are cases where there is no need of variation reduction (because there is no effect on the quality), but into the specs.

Like the water level of a bottle washing machine, The performance of the machine said that you shoud ensure that the level is at least 30 centimetrers from the top, there is no effect is you have 15 centimeters. You could argue that there is no case to control the water level, but PreControl could be of help.
:bigwave:

I have always thought of SPC Vs Pre-Control as Control (focus on process) Vs Inspection (focus on specs) mentality. How does one demonstrate improvement using pre-control?

As STEFAN STEINER and JOCK MACKAY said in

http://www.asq.org/pub/jqt/past/vol32_issue4/qtec-370.pdf

#“Does control charting work?”
In our experience, the answer to the question is “not
very well and not very often”. Some support for this
position is found in the final paragraph on control
charts of Ishikawa’s (1982) famous guide, where he
wrote “Control charts are easy to construct so are
widely used. But there are surprisingly few really
useful charts”.#

The reality is, in my experience, that most of the time the operator has not the information or knowledge to know the common causes of variation and that the "continuos reduction of variation" does not exist, it's like perfection, nobody can achive it, but, one has to try to get into it, why?, because if at least ones works to do our work better, it worth it, or as I readed more or less "In order to manage, one has to meassure".

If we take SPC and PreControl as one way to manage the operation, both tools work fine, SPC is as I said in the last post, the best tool to understand the process, Precontrol just as Donald Wheeler said "is focused on only maintaining the status quo", so as you said .... "NO IMPROVEMENT", but The question in order is "ARE YOU REALLY USSING SPC FOR IMPROVEMENT?", I know that in theory SPC does, but the people that is working on it does?, don't take it hard, may be that I am a little disapointed (but still convinced on SPC is the best).

:bonk:

Atul Khandekar said:

How does one demonstrate improvement using pre-control?
How about showing a reduction in the number of times the process has to be stopped for adjustment for one? Also, there is no law saying you cannot record the data from your Pre-Control measurements and compare newer results to older results to show improvement if you want. Process yields and/or downstream problems can also be tracked.

We can all argue as to whether Pre-Control is effective, or as effective, as other tools but the best way is to try it and compare in your application. Do what is best and works well for YOUR application and don't worry whether the "experts" with all of their pages of statistical analyses consider your tools to be adequate or not -- do what WORKS.

about pre-control

when Cp>1.0 ,or >1.33 (better), you can try to run pre-control.
remember it's Cp not Cpk, it's very important the average is near spec center.
i think it's a good SPC way !
it's fit for small lots production.
more easy to teach operator.
not need to calculate USL,LSL.

.....

Hi Dragonair - Welcome to the Cove.

Just a quick note - I use the Cpk value because it takes into account the closest point to the specification limit. This gives me a better idea of where a skewed process is running. If the process is skewed too much, you might not want to use pre-control.

Dave

Scott

I know, it's not written anywhere, but..., If the process is skewed too much, you can transform the data or use a target out of the center of the spec limits and calculate the zones in compliance to the target position (I tink this is going to create a polemic issue) so you might use pre-control.

:smokin:

Darius

Thank you Dave, Darius and Mike for your inputs.

I agree that theoretical discussions can be endless. However, finding out what works IMO, involves some experimetation. So before embarking on that, I am tring to achieve the following:

1. To establish a rationale (if not thumb rules) for using Pre-Control or SPC or any other tool in a certain process situation, so that if at all one has to experiment, one can choose the tool that is most likely to work. and / or

2. Asking others what did work for them in actual practice and under what circumstances?

D.Scott said:

Hi Dragonair - Welcome to the Cove.

Just a quick note - I use the Cpk value because it takes into account the closest point to the specification limit. This gives me a better idea of where a skewed process is running. If the process is skewed too much, you might not want to use pre-control.

Dave

sorry, it's a wrong input and expression.

Cpk is near Cp , it's very important condition to use pre-control.

it's not enough only Cpk is very big !

maybe it's a little different with SPC of tradition!


sorry, i am not good at english!

Your English is a whole lot better than my Chinese. You just keep right on posting.

Once again - Welcome

Dave

First of all, this is my first post on the Elsmar Cove forum. I hope this is not too wordy, I am probably breaking an etiquete rule.:)

Secondly, I am new to this tool called "pre-control." I come from the Deming school of quality. So, when you first look at it, it sounds like blasphemy. But, a co-worker engineer and I are currently working on implementing some "pre-control" tools, and I plan to post the results of this effort.

What about this debate?

I think that this is a fascinating debate about which is the better QC tool, "control charts" vs. "pre-control." I must say that Keki Bhote in "World Class Quality," chapters 20 to 22, makes an excellent case against the "traditional control chart" method and for the "pre-control" method. Bhote uses many examples to demonstrate the differences and advantages. He even takes actual examples and shows the data, probability and pitfalls where an operator using a control chart may think that the process is doing just fine and "in-control" when in fact he or she is making defectives with a poor Cpk of 0.51.

I tend to agree with Atul, the proof is in the experiment. So, I will need to perform some parallel process data collections with actual quality data transformed into both tools to examine and compare the two tools, take a look at the line stoppage and corrective action/fix cycles, etc.

Some key things that makes this debate intriguing:

The Deming group would argue that the "pre-control" method is not a clear view of the process variation and it promotes tampering with the system, making rules that the operator must "stop" to make an adjustment--tamper with the system. Deming would say that it invites tampering and does not at all equip the process operator with a tool that lets them continously control and reduce the process variation.

The Shainin-Bhote group would argue that "pre-control" method has enormous power in its simplicity and takes into account the product specification always. And, when the process is already established with a Cpk of 1.33 or better, than the b-risk of accepting bad product--drops to zero.

So which appears to be the better tool?

After reading this thread, and some of the literature behind this debate, I have concluded, for now...

A "traditional control chart" such as the Xbar & Rchart alone does not cut the mustard. If you want to use a traditional control chart, you would need to couple that tool with a running process capability curve-histogram and running Cpk values--a Cpk value that must be maintained daily with a set group of sample data, updating subgroup data regularly. This is not an easy tool to teach to the average machine operator with little to no statistical knowledge. Essentially, you would need to detect, stop and take action to fix that imaginary red-shaded part of the curve that goes over the upper spec. limit. Some software packages have this pictoral tool (curve & histogram) adjacent to the control chart.

The "pre-control" tool appears to be very simple in its "stop" and "go" rules and makes it simple for an operator to understand and follow. Five to six greens in a row assures a minimum Cpk of 1.33, for example.

In the coming weeks I plan to do some actual comparison tests to experiment and prove or disprove my thinking...this will take some time as we need to develop some software or find some that has the "pre-control" tool included. If anyone knows of a software package, let me know.

Wow! Someone willing to do a test! Cool! :applause:

Didn't someone once say a test is worth a bunch of expert opinions? Nah...probably just some nut...:notme:

In the coming weeks I plan to do some actual comparison tests to experiment and prove or disprove my thinking...this will take some time as we need to develop some software or find some that has the "pre-control" tool included. If anyone knows of a software package, let me know.

What kind of process are you trying to evaluate? :cool:

Wow! Someone willing to do a test! Cool! :applause:

Didn't someone once say a test is worth a bunch of expert opinions? Nah...probably just some nut...:notme:

Actually, I think what is more applicable is "Perfect practice makes perfect." You can do a test, but if it is designed incorrectly if will provide elegant statistical results that are wrong. So, a test in and of itself is admirable, but its worth is not guaranteed. :cool:

I think we should make room in our toolboxes for both Pre-Control and Control charts.

Pre-Control works when there is little process knowledge but the process has to run. The only information available is product specifications, but process information is usually collected during the pre-control phase. The "pre" in pre-control implies that "control" is going to come later.

Control charts work well when the results of a process study is available. It is not necessary to know product specifications width. BTW, a post stated that the measurements must be normally distributed for control charts to work. In fact by subgrouping, the Central Limit Theorem kicks in making them normal, besides a little deviation from normal will hurt us much.

Hope this helps,

Tom kjl.j'jll

BTW, a post stated that the measurements must be normally distributed for control charts to work. In fact by subgrouping, the Central Limit Theorem kicks in making them normal, besides a little deviation from normal will hurt us much.


Actually, this does not work for precision machining. Controlling the process as if it is normal leads to over control, not control. Precision machining exhibits the uniform distribution, which holds no use for the mean. A subgroup of 5 points from a uniform distribution (if taken correctly) is a uniform distribution. Lots of people collect data and believe they have a normal distribution, but it is measurement error - which is normal - not the underlying process. It simply has the appearance of a normal distribution. Let me explain further:

How many diameters are in a circle?

An infinite number.

How many diameters does one typically report on a control chart?

One.

What kind of lottery winner would you have to be to think that one diameter out of an infinite number accurately describes the circular feature?

It is statistically an insignificant sample of an infinite population. And, frosting on the cake, the average of an insignificant sample is even less significant. But, that is the X-bar R chart! Garbage for precision machining.

That is why the X hi/lo-R chart resolves those statistical problems. It properly utilizes the uniform distribution in its usage. And, that is why I asked what kind of process is being evaluated. :cool:

I have always thought of SPC Vs Pre-Control as Control (focus on process) Vs Inspection (focus on specs) mentality. How does one demonstrate improvement using pre-control?

Actually, most of the time one doesn't. Since to use pre-control, the process should be very capable and stable, improvement efforts are usually concentrated elsewhere, where the process is not yet very capable or stable.

Geoff Withnell

In fact by subgrouping, the Central Limit Theorem kicks in making them normal, besides a little deviation from normal will hurt us much.

I hate to let these little rumors spread. For many distributions, it seems to be handy. For the uniform distribution, the theorem fails miserably. You do not need triple integrals and partial derivatives to show this to be the case. I have attached a data set of a uniform distribution. When correctly sampled (and that is important, we are evaluating the process distribution, not the sampling error distribution) at 5 pcs, and they are averaged, the resulting distribution is also uniform. Any correctly sampled subset of a uniform distribution is also a uniform distribution.

Only one way to cut that cake. Wasn't that fun? :cool:

I hate to let these little rumors spread..... Any correctly sampled subset of a uniform distribution is also a uniform distribution.

After inspecting the dataset, it appears to be numbers that are trending upwards, then starts over. SPC is not a valid tool in this case because there is no variability hence no need for statistics. If we know one number, we will also know the next.

To demonstrate that the Central Distribution does work for a uniform distribution, try this experiment. Toss the die and record the results. After doing this several times, a frequency chart will show a roughly uniform distribution between 1 and 6.

Next pair the results up and find the averages of the paired observations. Again, create a frequency chart. This chart will have boundaries between 1 and 6, but will have a shape that looks like a normal distribution.

Interesting example,

Tom

After inspecting the dataset, it appears to be numbers that are trending upwards, then starts over.

That is correct!

SPC is not a valid tool in this case because there is no variability hence no need for statistics. If we know one number, we will also know the next.

There is variability, or it would be a straight, flat line. The variability has a rate - just happens to be relatively constant. There is a need for statistics, to ensure that the data remains within control limits, there are not special causes (we know what number to expect next, but it might not occur). Statistical Process Control (although not X-bar and R) is used to ensure the process remains in control, and the signals for adjustment are detected and the adjustments are made to the correct value. You are right in that it shows predictability, and therein lies a great deal of its power.

What you showed was the difference between discrete uniform distribution (your example) and continuous uniform distribution (my example - and the distribution found in precision machining). Nice example! :cool:

Bob,

You had asked what type of processes we are wanting to monitor with the "pre-control" tool. This first phase will be for a machining department, aluminum automotive parts with fairly tight tolerances. Parts are ran on both multi-axis CNC machining centers and automatic line with with gang-head machining line. By the way, these processes have been reviewed/studied fairly recently and meet the 1.33 Cpk requirement, so we are starting with stable machining processes. Actually, many of these part features are 1.67 and better.

Currently, we are looking at some software options. We may see if Atul's company, Symphony Tech, can help develop a software program, adding the "pre-control" chart functions to their existing "SPC Workbench" software, not sure at this point.

Let me know if you know of a software option, we will be also looking into "QC Calc," made by ProLink which we now know does include this tool.

You had asked what type of processes we are wanting to monitor with the "pre-control" tool. This first phase will be for a machining department, aluminum automotive parts with fairly tight tolerances. Parts are ran on both multi-axis CNC machining centers and automatic line with with gang-head machining line. By the way, these processes have been reviewed/studied fairly recently and meet the 1.33 Cpk requirement, so we are starting with stable machining processes. Actually, many of these part features are 1.67 and better.

Currently, we are looking at some software options. We may see if Atul's company, Symphony Tech, can help develop a software program, adding the "pre-control" chart functions to their existing "SPC Workbench" software, not sure at this point.

Let me know if you know of a software option, we will be also looking into "QC Calc," made by ProLink which we now know does include this tool.

Do these machines have auto tool compensation? I am glad you have identified that your processes are held to 1.33 to 1.66, but I am curious how the data was collected and how the capability was calculated. Often I have seen very high capabilities, but they were really sampling error or overcontrol. Even without those errors, chances are your capability is probably much greater if evaluated correctly. If you have what I define as precision machining - that is your primary variation (when measurement error is removed) is tool wear, then precontrol will appear to work. But, at that point, you might as well just set up I-MR with fixed control limits at 75% of the tolerance and run. It is the closest thing in software available to the X hi/lo-R chart.

Unfortunately I am not aware of any software that has X hi/lo-R. Those that I have discussed it with were A) unaware of the uniform distribution in precision machining and B) claimed precision machining was a special case that was not significant enough to generate code for. Do you feel that special? Anyway, using I-MR you will just miss out on the benefits of tracking roundness. I have found it particularly helpful with steel, titanium, etc. Not sure if aluminum will give the same stark results.

Were the capability studies run using data plotted in time sequence prior to preparing the histogram? That is a key analysis for precision machining. Are the multiple heads shared for each part, or do some parts run on some heads, and other parts run on other heads? If a hard only chucks some of the parts, then they should be charted separately, or the head to head variation will confound the results (as on a screw machine).

Would you be willing to share one of your capability studies? :cool:

In fact by subgrouping, the Central Limit Theorem kicks in making them normal, besides a little deviation from normal will hurt us much.

Interesting note:

The central limit theorem (http://en.wikipedia.org/wiki/Central_limit_theorem)(CLT) states that the re-averaged sum of a sufficiently large number of identically distributed independent random variables each with finite mean and variance will be approximately normally distributed (Rice 1995).

The continuous uniform distribution that arises from tool wear is neither random nor independent. So, CLT must not be assumed to apply to every distribution. :cool: