Let's fix Six Sigma!

[bobdoering]

There are no companies that use SS that I know of where there is real, verifiable evidence of the grandiose claims of money saved.

That is likely to be true for all quality system implementations. When QS9000 came out, the claim was the overhead would be covered by the vast improvement in the quality of the output. That may be true IF YOUR PROCESS STUNK. If you already had a decent process, the incremental improvement may not cover the overhead of implementation.

Quality is free if the broom you buy is cheaper than the mess you clean up...or something like that...

 

[Bev D]

It's not just "charlatans" causing the trouble, although they're a significant part of the problem. It's all of the "practitioners" who feel confident in their use of the tools but have only superficial knowledge of them. There are no companies that use SS that I know of where there is real, verifiable evidence of the grandiose claims of money saved. There might be improvements, but often at the expense of inappropriate--and often unnecessary--use of statistical tools by people who don't understand what they're doing. Let's face it--DMAIC is just PDCA in an ill-fitting rented tuxedo.

well certainly there are bad practitioners in every company, but that 's true for every good thing out there. If you're ever in Maine give me a shout and I'll show you where its' worked and the money and benefits are real.

 

[Mike S.]

Based on my personal experience with SS and what I have read about it, if I owned a company, I would not spend my money on a "Six Sigma Program".

I wonder how many folks would answer differently?

 

[artichoke]

Funny. Unfortunately not true. Bill Smith never wrote any such thing. (you have read his work, right?)

Yes I have but I'm sure very few others have done so. If they did Six Sigma may not have got off the ground.

"Another way to improve yield is to increase the design specification width"
IEEE Spectrum 1993, Bill Smith. In other words, set the specification as you wish.

Mr Smith pulled 1,5 sigma is pulled out of thin air in the same paper: "This is indirect process control, and in such processes, batch-to-batch variation can be as much as *1.5 sigma off target." That, is for uncontrolled batch processes. Mikel Harry took this and with his nonsensical "proof" based on stacks of disks, applied it to all processes.

Also from the same paper:
'So what must management do? The answer has become Total Quality Management (TQM). Interest in TQM began to gain momentum in the West when NBC aired a white paper documentary in 1980. This broadcast featured an American named W. Edwards Deming. His now famous “fourteen points for management,” when followed, appear to move organizations toward prosperity.'

It is a great pity that industry didn't follow his advice !

It was the snake oil sales men who bastardized his - and other's - concepts - but they did make a lot of money:

Mikel Harry and partner led the pack of thieves with his various attempts to prop up the nonsense, from it being a "shift" to a "correction" to "not needed" to a "dynamic mean offset".

They misrepresented Deming and SPC for their own gain.

Any good examples ?

 

[bobdoering]

"Another way to improve yield is to increase the design specification width"
IEEE Spectrum 1993, Bill Smith. In other words, set the specification as you wish.

If he truly meant design specification width, and it was determined from historical evidence that the original specifcations were too tight, then he is correct.

Often original design specifications are made based on paper calculations - best estimate starting places. There may have even have been some tests performed to verify the specifications - but typically small samples versus historical population. Long term history can be a lot more solid evidence of where the specifications should be. If they were designed too tight, and the long term evidence shows that they can be opened up, then that is not equivalent to setting them as you wish. It will improve yield. Is very well be lean, as you do not put more effort into the system than is required. It may also be economic process control.

May be a good thing. It depends.

 

[bobdoering]

Conversely, it can be shown that constantly reducing variation can become a bad thing. It depends.

 

[ralphsulser]

Like tooling costs when an underbody metal stamping part has an unmated trim edge in the wind and is toleranced the same as hole locations. We used to fight that a lot until we got some design engineers educated.

 

[Miner]

"Another way to improve yield is to increase the design specification width"
IEEE Spectrum 1993, Bill Smith. In other words, set the specification as you wish.

Mr Smith pulled 1,5 sigma is pulled out of thin air in the same paper: "This is indirect process control, and in such processes, batch-to-batch variation can be as much as *1.5 sigma off target." That, is for uncontrolled batch processes. Mikel Harry took this and with his nonsensical "proof" based on stacks of disks, applied it to all processes.

In your posts and the articles that you linked, you harp constantly about the 3.4PPM. This is NOT the core of Six Sigma. You can throw this concept out entirely and Six Sigma is still an effective tool.

Yes, it is a tool and should only be used where appropriate. You don't use an elephant gun to shoot a mouse. Six Sigma should only be used on large projects where conventional methods have not worked.

The real heart of Six Sigma is the DMAIC process. This provides a structure around which standard statistical tools and lean methods may be used. The structure ensures that the statistical tools are used in the correct sequence. For example, using Voice of the Customer (VOC) to determine Critical To Quality (CTQ) characteristics, then MSA to ensure that you correctly measure the CTQ followed by capability studies to baseline the process. This is followed by the use of other standard statistical tools such as hypothesis tests, ANOVA, DOE, etc. to determine the process variables that control the CTQs, so improvement efforts target the correct variables.

Regarding the so-called elitism of the Belt structure, very few people have the inclination or desire to learn the standard statistical tools required for the tough problems. This takes a large investment of time and energy on the part of the person so-trained. The belt structure simply identifies the level of training received and applied.

Regarding the widening of tolerances, bobdoering gave a very well reasoned explanation that was right on target.

I would agree that there are many consultants out there that are using Six Sigma as a means to make money, and try to sell it as a One Size Fits All, or the Cure for Whatever Ails Ya. Six Sigma is neither, but is is an extremely valuable tool when used appropriately in conjunction with all of the other traditional quality tools.

As with all tools, use the correct one for the job. If you need to turn a screw, use a screwdriver, not a hammer.

 

[artichoke]

Whether or not Mr Smith is primarily to blame for Six Sigma's specification basis, is a matter of conjecture. What is clear is that the specification (and counting defects) gives no indication of process quality. As Don Wheeler states on this point "the sooner one wakes up to this fact of life the sooner one can begin to compete" (page 202, "Advanced Topics in SPC")

In your posts and the articles that you linked, you harp constantly about the 3.4PPM.: mad : This is NOT the core of Six Sigma.
...
The real heart of Six Sigma is the DMAIC process.

Yes, I do agree that 3.4 dpmo is utter madness.

As people began to wake up to the nonsense of Six Sigma's 3.4 dpmo metric, DMAIC started to be held up as its "magic". They forget (or are too young to remember) that 20 years ago, there were dozens of such multistep (4, 5, 6, 7 or more) methods, including of course PDCA. Whether one is "better" than another is also a matter of conjecture. If there was any magic, it happened 5 centuries ago, with the development of "The Scientific Method", to which all these methodologies own their origins.

 

[bobdoering]

What is clear is that the specification (and counting defects) gives no indication of process quality. As Don Wheeler states on this point "the sooner one wakes up to this fact of life the sooner one can begin to compete" (page 202, "Advanced Topics in SPC")

I agree that counting defects is not a good indication of quaity - it is an indication of non-quality. The idea is to control the process to eliminate defects.

As far a Dr. Wheeler's quote, he prefaces that with:

"World class quality has been defined by "on target with minimum variance" for the past thirty years."

Target meaning the value of the characteristic that has the least "loss" associated to it.

He prefaces that with:

"'On target' will require that one knows how to set the process aim in such a way to get the process average to be as close to the target as possible. 'Minimum variance' will require that a process be operated in such a way that it will display a reasonable degree of statistical control." [My emphasis]

What it leads to is if the loss function of the process is greater than the design specification, then you will have to develop a process specification that is tighter than the design specification to maintain optimum output. Guess what? One of the participants of the loss function had better be being "out of design specification," or you may not impress your customer with your statistical wiles.

The loss function and its is an attempt to tie down the process control, to eliminate the issue of process control does not mean "in specification" because there is no relationship between specification and control limit evaluation. It has been tied down before with capability, and loss function will tie it down more.

Process improvement does not have to come from reducing the range about the target. Again, the variance of the loss function should be considered a "total variance," and any reduction of that variance could be a much more economically effective method for improving the loss function - such as the reduction of tool wear rate used as the key factor for continuous improvement in precision machining as it reduces the process cost of the operator adjusting the process more frequently. Conversely, compression of control limits increases the costs associated with operator adjusting, and it could be much greater than any other cost benefit, depending on the value of the part.

I believe it boils down to you should not use specification as a "basis", nor should you only use "target" as a basis. You should consider both.