Zero Defects & Taguchi Loss Function - ISO 9000 and the Zero Defects Philosophy

[Marc]

Subject: Re: REQ: Zero Defects /Bartol/Meron
Date: Fri, 18 Dec 1998 11:40:06 -0600
From: ISO Standards Discussion <JENNEJOHNN@UWSTOUT.EDU>

From: Emanuel Meron <ms_comp@netvision.net.il>
Subject: Re: REQ: Zero Defects /Bartol/Meron
>
> From: Dave Bartol <dbartol@hayesmfg.com>
>
> List Members:
>
> I have a question regarding ISO 9000 and the Zero Defects Philosophy
> of Crosby . . .

> For example, a part has the spec. of 11.000 +/- 0.005 inches. Anything
> within the specification is zero defect, but does the part at 11.005
> work as well as the part at 11.000? How about the part at 10.995
> inches?

This issue is addressed by Taguchi in his famous loss function which basically does away with the traditional concept of tolerance where all parts within specs are considered identical and "good", and those outside specs are equally "bad".

Using the concept of "loss" the nominal value is the target and every deviation from it entails a cost (loss). The farther away a part is from nominal the higher the loss, the smaller the deviation the lower the loss. This means that you have to strive for minimum variability around the nominal (target value), not just for being "in spec". The beauty of the theory is in that it provides you with a quantitative way to compare the benefits of variability (loss) reduction with the cost of achieving this reduction. There are many books on this subject, just look for Taguchi's loss function.

Emanuel Meron

[Kevin Mader]

Marc,

Do you have the story on Ford's experience with this topic regarding the Taurus transmissions? It might make a good post for those interested in seeing the benefits of variability reduction.

[Marc]

I don't off hand. Are you referring to the old Batavia experiment - 50% US manufactured transmissions and 50% manufactured in Japan where the major issues were shaft surface roughness and concentricity? The one they did the video on that they quickly 'recalled'???

[Don Winton]

Kevin,

This is the Ford story I have. It is from 'Taguchi Techniques for Quality Engineers' by Phillip J. Ross and is attributed to Ford Motor Company, Dearborn, Michigan, 1987.

____________

Another example of the economic impact of excess variation occurred in the automatic transmission business with a major U.S. automobile company. Ford had contracted a Japanese supplier, Mazda, to make a certain portion of their front-wheel-drive automatic transmissions, with the balance of production made at a U.S. plant in Batavia, Ohio. Both sites were making transmissions to the same set of blueprints and the transmissions were being installed only on American cars. Mazda's version, as warranty records showed, had a substantially lower claim rate than the Batavia version. Ford investigated this phenomenon and found that Mazda's transmissions were made much more consistently than their own. On some critical control valve components (valves, valve bores and springs) which make a transmission shift automatically, Mazda was using only 27% of the allowed tolerance range, while Batavia was using 70%. Ford thought its plant was doing well, and by traditional standards it was, but the Mazda plant was superior. Not only were all the parts made to print, as were the U.S. made parts, but they were more nearly like one another.

____________

Regards,
Don

[Kevin Mader]

Marc and Don,

You got the right story. As I recall, the calibration technicians were called in to evaluate and recalibrate the gages used to check concentricity and surface finish. As a result, the technicians said nothing was wrong with the gaging. "How could this be"?, said the inspector. The gage isn't reading anything (or very little). They went through this process for a second time. As it turned out, the quality level achieved by the folks at Mazda was so much better than the others made by Batavia that the inspectors misdiagnosed the situation as a gage problem.

I wasn't aware of the percentage of "tolerance used" by the two groups, but it goes to show that good enough isn't! When I first read on the Taguchi Loss Function and saw a figure illustrating it, I thought to myself, how simple and yet so profound. Nothing complicated about the message or understanding it. Why didn't I think of it? Simple. I was using the wrong paradigm! Funny how your taught about tolerancing and the acceptance of mediocre results without realizing that this is being done. The loss function challenges organizations to raise the bar and think differently.

As I also remember, the waiting list for Taurus models with the Mazda transmissions were quite long. But also recalling that the Taurus had notorious troubles with their automatic transmissions, the list was quite understandable. Thanks for the feedback gentlemen.

[Don Winton]

Kevin,

Your recollection of the gauging story sounds about right.

Your comments about the percentage of tolerance used are right on target. That is what I like about the loss function. It illustrates Deming's message (reduce variation) into a graphical format that make explanations relatively straightforward. It allows me (or anyone else for that matter) to explain to management, in terms they understand ($$$), the what's and whys of variation reduction.

Regards,
Don

[Marc]

I saw the Ford video of the Batavia experiment back around 1985 or 86, as I remember. I couldn't get a copy. I didn't have a VCR back then to copy it and Ford had already 'withdrawn' the video. This 'experiment' has obviously become quite a classic.

But back to the loss function - again we simply have an example of Common Sense if you take a minute to look at it. No more, no less.

[Dawn]

I think this is a good area for a good question.
The plant manager has asked what a good corrective action would be for 70,000 pieces which ran 6,000 nonconforming throughout 3 shifts within a couple of days. The parts were nonconforming throughout the process-not because of one operator error, not because of a sudden shift in the process. The nonconforming parts were not found until final.
The parts have .007 tolerance and they were using all their tolerance.
My opinion was:
If you do not need .007 tolerance why use it-stabilize the process and hold control charts within .004 or less. The problem is too much variation. Does this work for you?