In a previous automotive manufacturing enviroment we established 'guardbands' set at 3 std dev on the gage error. This to ensure we did not pass any product that might be rejected by the customer if tested. The product was crash sensors. We were willing to make a class one error to avoid a class two. In all the millions we made, we have never had one come back or lost in litigation as when retested, they were always within spec. We have had lawyers bringing product back in plastic bags that we could not touch unless under their supervision. They always passed wthin specification. We used horizontal linear accelerators to simulate a crash speed. We could measure in units of .001 MPH or less, but we had multiple test machines. The specification was to 0.1 All were correlated back to a master thruster that was tracable to the NBS. We always used two accelerometers that were compared to each other. If one started to drift the test would abort and the thruster would shut down until re-correlated with new acclerometers.

We were aware we could be calling good product bad, but due to the nature of the product we could not afford to pass bad product. If the reject rate increased statistically, the thruster would be shut down and maintence performed untill the varation was reduced.

Although I am now in a different industry now, I still believe the method we used was valid. All correlations used were stastically based using ANOVAs. The exact method is proprietary

I also used guardbanding when I worked in the semiconductor industry. The test specs would be tightened-up by 2 PE which would provide additional assurance that what we were shipping to our customer would meet their specification requirements. The thought process was that it was better to scrap some borderline product than have a customer return, complaint, SCAR, factory visit........you get the idea.

Rich

Having read and used all of Wheeler's material on EMP, I might be biased, but I didn't see any problem with his column. He is not talking about changing the specifications, he's talking about how much of the specification is useful when the measurement error is shown to be of some certain value. If I measure something that's technically "in spec" at 6.1 units, but I have .5 units of probable measurement error, there is a chance at I am actually out of spec. All Wheeler is talking about is how that can be quantified and adjusted for. He's not talking about changing the specs, he's talking about playing the slice in the measurement system so we can be sure that what we are measuring is in specifications.

I don't think this is an either/or situation. Wheeler answered a narrow question with a viable narrow answer.
He never said we shouldn't improve our processes or our measurement systems - in fact he is one of the longest (living) proponents of doing exactly that. He simply addressed a means of dealing with results that are at the limit(s) in the presence of significant measurement error.

I too advocate and have been very successful at improving process performance and measurement system performance, BUT I've also used guardbanding and still use it to protect the customer until the process or measurement system can be improved. And in several industries that I've been in, the measurement systems may be difficult to improve to the point where guardbanding can be eliminated quickly even when the process itself is greatly improved. Processes go out of control, vendors go out of control, physics happens. 'Marginal' materila will be produced. And while the author is correct that simply changing the limit at which we accept or reject product doesn't improve the PROCESS OUTPUT, it does improve the quality of the SHIPPED product. A little bit of scrapp or rework is far less expensive than failures in the Customer's hands...

I see a trend in Quality Digest articles i.e. criticize Don Wheeler and try to become famous by doing so :)

Just kidding!

Guru
(Gurbachan Chadha)

Wheeler provides access to other publications on his website that provide the theory and rationale for this latest Quality Digest article. Those who have read these corresponding publications should recognize the profound and extremely important points he is trying to make with respect to the improper estimation and application of traditional gauge R&R..

However, I must admit that I have found the path from paper to practice much clearer with his previous articles (in fact we have implemented his teachings from nearly all of his past articles over the years); in contrast, for this article, the path is a bit muddy... and so I tend to agree with Steven that the intent of this latest article could be misunderstood and misapplied.

Perhaps Dr. Wheeler may find this feedback useful and offer a clarification to his fan base...

Don Wheeler has provided another of his excellent, informative and well explained articles. Steve's reply seems almost like personal sniping at Don, with nonsensical comments like "Wheeler’s notion seems to allow for the backward view that changing a specification will magically cause the process to produce product that meets that specification."

I think whether to use or not to use Dr. Wheelers approach for setting the specs "depends on" the type of industry you are in....If you are in the health care, aero space or auto safety industry, the ramifications of a product failure is not worth the the risk...this approach is applicable there. However, we must at the same time must work to improve: 1. the measurement system and 2. the process. We can use this approach until we have some success. In the mean time we understand that we are throwing away lot of conforming product due to misclassifcation errors.

But if you are in the consumer goods industry like adhesive tape we don't have to kill ourselves by tightening the specs, because it is worth to take the risk by minimizing the producers risk otherwise we will scrap or rework lots of good product by misclassifcation (type 1 error).

The answer to the question is (every enginneers and statisticians favorite answer): "It depnds"

Robin Francis