I need some help understanding a point in the AIAG SPC manual. The manual states the following: "High-speed, automated processes are often found to exhibit autocorrelation on some characteristics. This is often because there is an underlying predictable special cause variation which is large when compared to the common cause variation."

Question: Why is there an assumption that it is special cause, not common cause? Experience?

:confused:

I need some help understanding a point in the AIAG SPC manual. The manual states the following: "High-speed, automated processes are often found to exhibit autocorrelation on some characteristics. This is often because there is an underlying predictable special cause variation which is large when compared to the common cause variation."

Question: Why is there an assumption that it is special cause, not common cause? Experience?

:confused:

Everybody has their own definitions. I don't buy the concept of "predictable special cause variation". If the cause affects every part (e.g. tool wear), I define it as common. Another litmus I use: Can the cause be eliminated? Can you eliminate tool wear? Uh, no. Therefore, some common causes will create autocorrelation. I don't really see that as heresy - it just may be important evidence of an underlying dependent non-normal distribution. If the cause does not affect every part (e.g., broken tool), I then define it as a special cause.

Not all folks agree with me. That's OK. They have a right to be wrong. :tg:

Thanks. I agree with you, but was not sure I was just convincing myself.

This is why the term "assignable cause" is more appropriate than "special" cause. Assignable doesn't inherently imply sporadic or that it doesnt' 'belong'

This is why the term "assignable cause" is more appropriate than "special" cause. Assignable doesn't inherently imply sporadic or that it doesnt' 'belong'

The argument could be made that "common" cause variation is also "assignable" in that it's possible to know where it's coming from. In that sense, "special" is perhaps more to the point.

The argument could be made that "common" cause variation is also "assignable" in that it's possible to know where it's coming from. In that sense, "special" is perhaps more to the point.

true. the terms are operational definitions not laws of physics. I have no problem with either definition.

I tend to use 'assignable' when I know the cause, 'common' when I don't know the cause(s) but they are stable such that the variation is predictable and random not systemic and I use 'special' when there is an unexpected out of control condition that truly represents a change from the stable level...
in the end tho, I find the terms somewhat misleading any way and try to say this variation needs to be improved or it is fine the way it is...it tends to be less confusing for my organization...

Yeah. I have had some people who mixed up special and common and had all kinds of trouble making sense of the literature, sometimes with comic results. I was mostly just wondering the basis on which authors made the statement that is WAS special cause.

I'll give the AIAG some credit, they have tried to boil down the various concepts into what they believe to be a usable basic primer. But, they do get caught up in some common errors, such as normalcentricity. Their glossary identifies the following definitions:

Common Cause: A source of variation that affects all of the individual values of the process output being studied; this is the source of inherent process variation. (Whoa...I actually agree with them! Or do they agree with me? Hmmm.... Tool wear is a good example.)

Special Cause: A source of variation that affects only some of the output of the process; it is often intermittent and unpredictable. (Tool breakage is a good example of that) It is signaled by one or more points beyond the control limits or non-random pattern of points within the control limits (Well, only if you are expecting random points from your process...normalcentric thinking. The signal might be random data when you are expecting cycling data!).

I need some help understanding a point in the AIAG SPC manual. The manual states the following: "High-speed, automated processes are often found to exhibit autocorrelation on some characteristics. This is often because there is an underlying predictable special cause variation which is large when compared to the common cause variation."

Question: Why is there an assumption that it is special cause, not common cause? Experience?

:confused:
A better description of what autocorrelation is where the value of a measurement depends on the value of the measurement preceding it.

Examples could include:


Precision machining where short term variation is extremely small and long term variation is primarily tool wear. A measurement taken is strongly influenced by the preceding measurement until a longer period of time elapses.
Extrusion processes typically exhibit small short term variation. Long term variation appears in the form of material changes. Again, a measurement taken is strongly influenced by the preceding measurement until a longer period of time elapses.

The intent is that common cause variation is random - no patterns - within predictable stable limits.

Special cause - or assignable cause - variation has either a pattern (such as tool wear or extrusion) that is 'systemic' or always there the sporadic change - a shift or very extreme value that goes beyond the limits of stable variation...

It cannot have been easy for them to try to boil down all the SPC material out there into one manual and in general they have done a good job. But you are right, they have made some errors. Have you read the definition for unimodal for example: "a distribution is said to be unimodal if it has only one mode." :rolleyes:

It cannot have been easy for them to try to boil down all the SPC material out there into one manual and in general they have done a good job. But you are right, they have made some errors. Have you read the definition for unimodal for example: "a distribution is said to be unimodal if it has only one mode." :rolleyes:

Hmmmmmm....deep. :smokin:

It cannot have been easy for them to try to boil down all the SPC material out there into one manual and in general they have done a good job. But you are right, they have made some errors. Have you read the definition for unimodal for example: "a distribution is said to be unimodal if it has only one mode." :rolleyes:

What's your definition?

A large part of the problem is trying to force all different types of variation into just two categories ("common cause" vs "special cause"). In reality there are many sorts of variation. For some purposes, this simplistic dichotomy works, but for others it falls short (as we see here). Even throwing in "assignable" doesn't really cover it.

More precise language to more precisely describe the situations would be in order. So I humbly suggest the following might be more precise (and often overkill) for describing variation:



Is the variation random in time and/or location? Then it is "uncorrelated" rather than "correlated".
Does the variation apply to all parts? Then it is "common cause" rather than "special cause".
Do you know what caused the variation? Then it is "assignable cause" rather than "unassignable cause".

So tool wear would be "correlated common assignable cause". A multi-cavity mold with one bad cavity would be "correlated special assignable cause". When my computer randomly freezes and needs to be rebooted, that is "uncorrelated special unassignable cause". An ideal process that is in control would have only "uncorrelated, unassignable, common causes".


Tim

An ideal process that is in control would have only "uncorrelated, unassignable, common causes".

I guess now we have to define what is an ideal process. I prefer:

A process in control is in the ideal state 100% conforming and predictable:
-must remain stable over time
-must operate in a stable and consistent manner
-must be set at the proper level
-the natural process spread must not exceed the product’s specified tolerance

I would be perfectly satisfied that a "correlated common assignable cause" process, such as grinding, would also meet those requirements, if controlled correctly.

Requiring an idea process to be "uncorrelated, unassignable, common causes" sure sounds sort of normalcentric to me.

I guess now we have to define what is an ideal process. I prefer:

A process in control is in the ideal state 100% conforming and predictable:
-must remain stable over time
-must operate in a stable and consistent manner
-must be set at the proper level
-the natural process spread must not exceed the product’s specified tolerance

I would be perfectly satisfied that a "correlated common assignable cause" process, such as grinding, would also meet those requirements, if controlled correctly.

Requiring an idea process to be "uncorrelated, unassignable, common causes" sure sounds sort of normalcentric to me.

Good points. Perhaps I should have said a " typical idealized example of an in-control process" instead of "ideal". There are many examples of perfectly good (ie profitable) processes that do not follow this idealization.