Getting started with statistically based process improvement

[Scott Thor]

Here's an article I wrote to explain the basics about statistically based process improvements. Feel free to give me any comments you may have.

Scott

[Craig H.]

Scott,

Nice work! Thanks for sharing!

[Jim Wynne]

Originally Posted by thorman

Here's an article I wrote to explain the basics about statistically based process improvements. Feel free to give me any comments you may have.

Scott

Thanks for sharing. A few observations:

• In Step 2 you encourage determining the current state of the process through use of control charts, which is sound advice, but you also suggest that "A basic condition that must be satisfied before any improvements can be made is a stable and predictable process." There's a bit of redundancy there, because in this context "stable" and "predictable" are synonymous. Also, the problem might be instability, so in curing that issue, the problem might be solved. In other words, you can make improvements to an unstable process, contrary to your assertion that stability must be achieved as a prerequisite.
• The charts in figures 1 and 2 show signs of instability other than points beyond the control limits; you might want to point out the fact that there is more than one test for statistical control.
• You say, "Any points outside of the limits represent what are known as special causes..." Not necessarily; when using the normal curve as a model (which might not be advisable) we can predict that some points will naturally fall outside the ±3-sigma limits. It might be good to point this out, and advise against tampering.
• You say, "Typically most improvement projects identify that the process is unstable which then begins the task of identifying why that is?" I'm not at all sure that the statement is true. In many cases, analysis will reveal a predictable but incapable process. Also, the sentence should end with a period and not a question mark.
• You say, "If after getting you process under control the upper and/or lower control limits are outside of the specification limits you have a process that is near the brink of chaos." Clearly, the condition you describe is undesirable in most cases, but "near the brink of chaos" seems a bit strong. Although the instances are rare, there are times when a certain level of nonconforming output is acceptable and economically necessary.
• In addressing "Sustaining the Improvement," you say, "Without the proper controls all processes will tend to work back to where they began before the project." This is certainly true, but the object of the project should be installation and monitoring of controls. That is to say, the object should be to identify the process controls which, if maintained, will result in conforming output, and then monitor and measure those, rather than focusing on part/output measurement. The entire idea of process improvement should be making the process predictable by controlling the process variables that have been proven to contribute to conforming output.

[Steve Prevette]

My suggestions would be:

Yes, I can improve a process that is not stable and predictable. I look at the special causes associated with the statistical signals and deal with them (corrective action when in the bad direction, reinforcing action when in the good direction). Actually, I'd much rather have this situation as compared to a stable and predictable process, as these special causes are usually easy to deal with, and fit with the paradigm of most managers that I must "do something" with the specific results.

It may be worth pointing out that changing a stable process is HARD!

On Figure 1, where did the control limits come from? They certainly do not fit the current data. Is it from some older data? If so, it is worth showing the older data (which should have been in control to get the control limits from) and then show the changing condition related to the trends (and yes, there are more than one) on Figure 1. In showing examples, they should be rigorous, good examples that you would want others to follow.

In figure 2, I'd show the example with at least 25 points. Dr. Shewhart stated do not declare a process stable without 25 points. It sets a good example to show your example chart with 25 points. So many folks want to throw out old data prematurely.

On the statement "Many projects make significant improvements only to fall back into the [previous] state". I would disagree with that. What I see more often is that people declare success on a "lucky" result, not a statistically significant trend on the control chart in the improving direction. Generally, once the control chart shows a significant trend, the improvement has stuck. At least, that is my empirical experience, and it is supported by theory.

Also, I'd suggest being careful with the "gut feeling" discussion. All data are flawed, Dr. Deming stated that there is no true value of any measurement. The best state is where you can reconcile your gut feelings with the data. I'd much rather have a doctor operate on me when their gut feeling and the data are in synch. How many times has that little red flag gone off in your head that something is amiss, but you ignore it due to the numbers, and come to find out . . .

Good luck, and Happy New Year

[Steve Prevette]

Originally Posted by Jim Wynne

You say, "Any points outside of the limits represent what are known as special causes..." Not necessarily; when using the normal curve as a model (which might not be advisable) we can predict that some points will naturally fall outside the ±3-sigma limits. It might be good to point this out, and advise against tampering.

I disagree with this disagreement. We use the 3 sigma limits as the operational definition of a signal. Just as we evacuate a building when the fire alarm sounds, and then check back to see if it was a false alarm if we find no indication of a fire, we should still take immediate action on a point outside the control limits. Yes, it may be a false alarm, but we do take a good-faith effort on taking action. I would not consider taking action on a 3 sigma limit outlier to be "tampering".

[Jim Wynne]

Originally Posted by Steve Prevette

I disagree with this disagreement. We use the 3 sigma limits as the operational definition of a signal. Just as we evacuate a building when the fire alarm sounds, and then check back to see if it was a false alarm if we find no indication of a fire, we should still take immediate action on a point outside the control limits. Yes, it may be a false alarm, but we do take a good-faith effort on taking action. I would not consider taking action on a 3 sigma limit outlier to be "tampering".

If you define "taking action" as looking to see what (if anything) happened, then I agree. I didn't mean to suggest that the potential signal should just be ignored. The statement in the article said that "Any points outside of the limits represent what are known as special causes..." And I said, "not necessarily," which is correct.

[Steve Prevette]

Originally Posted by Jim Wynne

If you define "taking action" as looking to see what (if anything) happened, then I agree. I didn't mean to suggest that the potential signal should just be ignored. The statement in the article said that "Any points outside of the limits represent what are known as special causes..." And I said, "not necessarily," which is correct.

Ah, okay. Yes, perhaps something like "theory and experience tells us that we likely will be able to find a special cause for any point outside of the limits".

[Jim Wynne]

Originally Posted by Steve Prevette

Ah, okay. Yes, perhaps something like "theory and experience tells us that we likely will be able to find a special cause for any point outside of the limits".