Nice article. Working with a couple of colleagues responsible for incoming quality of subassemblies. This says what I've been trying to say much more cogently and interestingly.

We work for what's essentially been a huge job shop. Production articles have rarely been repeatedly produced, so we've been relying on engineering control (meeting spec) to assure we get what we want. Yes, yes, it would have been easy and helpful to use SPC and similar, but what we've done has mostly worked.

We've recently begun working with larger-volume manufacturers, working toward regular purchases involving significant numbers of the same parts, subassemblies, finished assemblies and even services. We've taken our usual approach of certifying an original design, then inspecting the bejeezus out of what comes to us and comparing that to specs. Many of the managers, engineers and others I've been working with are recognizing that we could get even more value if we make sure the manufacturing and operation processes going forward is the same as what got certified in the first place - that the process is predictable, in statistical control.

I'm from heavy industry. I know this will work. It's unfamiliar to my present employer and our suppliers, though. Time will tell whether it catches on in any meaningful way, but so far, the simplicity and head-smackingly visually obvious methods of SPC are piquing interest.

Keep writing. btw - really appreciate the relevant links to other articles.

Good article except for your opening statement: "In everyday language, “in control” and “under control” are synonymous with 'in specification.' Requirements have been met."

Control charts tell us if the process is stable and predictable, not if requirements have been met. I've seen a lot of processes that  are stable but not within specifications.

Once you have a stable process (determined by control charts), you can then use capability analysis to determine if the process is meeting specifications. Cp and Cpk can tell you if you're meeting specifications or not, as shown in this article.

Stable isn't always capable.

Hello Scott,

Very, very detailed article; shall need time to go through fully.

I only wish to express my surprise at two points of your article (as far as I have gone).

Sentence 3 i.e “Out of control,” on the other hand, is synonymous with “out of specification.”

AND

Whether figure 2 –plot of individual values of a parameter ‘should’ and  ‘can’ be called a “CONTROL CHART”.

With due regards; I venture to disagree with both.

Kind regards

Virendra gupta

PS I shall read the article fully and try to give more comments.

Nice article Scott!  You have articulated what is needed in production operations of every type.

I too ran up against the thought that "in control" meant within specifications.  That is when I started using predictable and unpredictable to describe the behavior of a process.  It has less emotional baggage than in control and out of control.

The problems in production are like fog or smog, and process behavior charts (control charts) can spotlight what needs attention and where we need to work.  Moreover, without the framework of an approach such as control charts it is difficult to get the most out of any process.  The problems of production are greater than those of research, where we spend a lot of time on developing techniques for learning.  Moreover, the problems of production change with time, and this is why we need a time-oriented technique such as the control chart.

Thanks for this article.

I found this article especially useful. The Florida Department of Transportation is working with our crushed stone industry to implement statistical process charts. The industry uses a common commercial software with the capabilities, but SPC is not a familiar tool.

We are teaming to develop a system that is logical and works for both FDOT and Industry. A common case we have come across and are trying to assess is what happens when products are within spec, but the Natural Process Limits are outside the specifications. That is Cpk  < 1.  Scott's article makes it clear that this process is not stable or in control, but it is a difficult concept to grasp when we have to codify this situation in a regulatory document. We are considering allowing producers to take note of this situation and take corrective action. If so, FDOT would not take action against the producer. In our current system, we calculate the (difference between the average and the nearest spec), and divide the result by the standard deviation of the last runing 30 test results -absolute. This results in a Z score, which is required to be grater than 1.645 to ensure 95% compliance. In the situation described, the Z value would show that the data is out of complianc and the producer would be prohibited form shipping to FDOT projects. This method assumes all data are normally distributed. SPC does not and more closely matches our scenarios.

The article sheds light on a potentail solution to a real problem, and offers a system that might be practical and mutually benefical.