SPC (Statistical Process Control) Overview

artichoke said:

Please read my post. I did not say that machining is "insignificant". I stated that machining is one of more than 400 industry codes.

Click to expand...

Well, if significance was not the point of the 1 out of 400 statistic you mentioned, then your actual point remains clouded. Especially when that statistic has nothing to do with the amount of machining in the overall scope of all manufacturing industries, due to a vast number of captured shops not identified by their claimed SIC code.

artichoke said:

The belief in the normality fallacy widespead. Processes do not need to be normal for the use of Shewhart control charts. I strongly recommend Don Wheeler's excellent book on the topic: "Normality and the Process Behaviour Chart".

Click to expand...

I have read Dr. Wheeler's books, as he has read mine. It is true he has identified that processes do not need to be normal for the use of Shewhart control charts. However, that is no more of a 'law" than the need for the charts to be normal. Not all non-normal charts can successfully use the Shewhart control charts. Fact is, if you look at one of the points for the claim that processes do not need to be normal to use Shewhart charts, you will find the review of Burr's 27 non-normal distributions. One of the distributions not covered by Burrs 27 non-normal distributions is the continuous uniform distribution. Too bad, it clearly shows that it does not play by the traditional Shewhart control chart rules. But, the X hi-lo/R Chart for that distribution does support Shewhart's premises for economic control, as illustrated throughout many postings in this forum, suitable for searching.

By the way, Dr. Wheeler's response to my book was "I have to admit that I have not directly addressed in my books the complex processes you have addressed."

bobdoering said:

Fact is, if you look at one of the points for the claim that processes do not need to be normal to use Shewhart charts, you will find the review of Burr's 27 non-normal distributions. One of the distributions not covered by Burrs 27 non-normal distributions is the continuous uniform distribution. Too bad, ...

Click to expand...

Don Wheeler's book "Normality and The Process Behaviour Chart", examines 1143 distributions (not 27), with many other than Burr. He gives an excellent validation to Shewharts's work.

Don Wheeler also examines uniform distributions on page 117, "Advanced Topics in SPC".

artichoke said:

Don Wheeler also examines uniform distributions on page 117, "Advanced Topics in SPC".

Click to expand...

It is a nice, generic justification for control limits, in general. I agree in the use of control limits, in general. But, it does not specifically support usage of the X-bar R chart (the original conversation) for a continuous uniform distribution. It may work for a discrete uniform distribution - if you find such a process to control, such as rolling a die. But, that does not relate to my point.

To use Dr. Wheeler's logic (page 116, "Advanced Topics in SPC"): the strongest justification that X hi/lo-R charts are ideal for precision machining is it works well in practice, and it provides effective action limits when applied to real world data.

What's more - it makes sense, and it make sense to the operator. It also provides far more useful data to the practitioner than the X-bar R chart. It can tell you when to make an adjustment, when to change a tool, and the tool wear rate. The X-bar R chart can not do that - and at best it generates overcontrol. It is easy to show how that can happen. I am sure if you have had an opportunity to properly implement this technique yourself, its benefits would be clear.

Again, these are issues that have already been discussed throughout the forum. Feel free to further search out the specifics.

bobdoering said:

It also provides far more useful data to the practitioner than the X-bar R chart. It can tell you when to make an adjustment, when to change a tool, and the tool wear rate. The X-bar R chart can not do that - and at best it generates overcontrol.

Click to expand...

The X-bar R chart is not intended to tell you when to change a tool nor to measure tool wear rates. It is however an excellent tool for the majority of process applications, regardless of data distribution.

Don Wheeler also makes an excellent point on page 36 "Normality and The Process Behaviour Chart". Regarding the majority of processes, it takes 3,200 data points to test for lack of fit to a distribution out to +/-2.95 sigma ( or 1,480,000 observations to check for normality out to +/-4.5 sigma) ... and by the time such data is collected, the process will have changed.

artichoke said:

The X-bar R chart is not intended to tell you when to change a tool nor to measure tool wear rates. It is however an excellent tool for the majority of process applications, regardless of data distribution.

Click to expand...

The fact that X-bar R chart is not intended to tell you when to change a tool nor to measure tool wear rates shows its dramatic limitation in value in precision machining. It does have its purpose in other cases, and I am not denying that fact.

artichoke said:

Don Wheeler also makes an excellent point on page 36 "Normality and The Process Behaviour Chart". Regarding the majority of processes, it takes 3,200 data points to test for lack of fit to a distribution out to +/-2.95 sigma ( or 1,480,000 observations to check for normality out to +/-4.5 sigma) ... and by the time such data is collected, the process will have changed.

Click to expand...

Academically, and in many cases, that is true. But, the number of points correctly collected in precision machining required to establish the distribution depends on the tool wear. It might only take 10 points - it could take a week of points. The better the tool wear, the more points it takes. It is the significant sampling error in the X bar-R charting methodology that creates its misinformation in precision machining. As I have mentioned, all of its problems have been clearly illustrated elsewhere in the forum for those that care to study the problem, and do not need to be repeated here.

It is good to read Dr. Wheeler's books, but it is not a good idea to stop there. Every author has a limit to their experiences, and he is no exception. His work is academically rigorous, but the empirical evidence shows the need for another approach in precision machining. One that has been shown to be very effective. It is the fear of the practitioners imposed by those who are compelled to think that the X-bar R will control their process (it will not) by blindly citing such references that is the crux of the problem in the field. You have provided adequate evidence of that. And to your original question, that is the point of the avatar - my goal to put a stop to that madness. Its point is not to say X-bar R is never good, or that the rules for control are always wrong. There is no madness, no worries, no cares for those cases. But the fact that they work for them can not be extrapolated -or "rubber stamped" as is typically the case - to the case of precision machining - or perhaps other "exceptions" you cited earlier.

bobdoering said:

But the fact that they work for them can not be extrapolated -or "rubber stamped" as is typically the case - to the case of precision machining - or perhaps other "exceptions" you cited earlier.

Click to expand...

If you take the time to read Dr Wheeler's books thoroughly, you will find many practical examples and very detailed analyses, with no signs of "rubber stamping" as you suggest.

artichoke said:

If you take the time to read Dr Wheeler's books thoroughly, you will find many practical examples and very detailed analyses, with no signs of "rubber stamping" as you suggest.

Click to expand...

I own his books, and have read them. They were a very good foundation and starting place to progress from. And, we have progressed beyond the trend line charting analysis he has proposed for tool wear. Its limitations have been established (some of which you mentioned), easy to prove, and resolved with newer methodology. The beauty of the newer methodology is that it is so clear, logical and practical that it does not require very detailed analysis. Its elegance is in its simplicity and effectiveness.

The "rubber stamping" does not refer to his work, but the incorrect application of the statistical process control concepts in the field.

Hello everybody,
I have a pre study material, where some charts and questionys are offen for me.
Could somebody help me with MSA, SPC, FMEA questions in the zip file?
It would be very helpful for me to take the TS evaluation!

snowman6705 said:

Hello everybody,
I have a pre study material, where some charts and questionys are offen for me.

Click to expand...

Offen?

Stijloor.

O excuse me for my not real excellent english.
I mean I can not answer the questions in the application excercises.