A tale of wealth and greed – episode 1

Wow, I do Cinderella and Thumbelina with my kids. Do you recommend that I switch them to these real life fairy tails? 

Thanks, good advice just like the other advice you give on here.

Peter,
 
To your comment that Shainin’s Pre-Control method has been shown to increase variation, I will assume that you have no experience with the method either through actual usage or doing simple simulations to really understand what comments like yours really mean. I will show you several scenarios and then let’s see what Shainin’s method does or does not contribute. Okay?
 
Scenario #1 – We have a process that is in control with a Cp of 1. The spec is 100 +/- 3. The process is centered at 101.5 after doing what the SPC texts suggests of 20 – 30 samples, compute centerline and control limits.
 
SPC (x-bar, r) – The control limits (set following Wheeler’s suggestions) are 101.5 +/- 1.35. Go create 100 five piece samples using rational subgroups and all of the other things that are proper for SPC. What you will see with all 8 of the rules Minitab provides turned on is 0, 1, or 2 out of control points for each sample of 100. I just did it about 20 times.
 
Pre-control – Using Shainin’s rules for dividing the spec into four pieces, I can’t get the process to turn on! Shainin demands five consecutive pieces in the green (evidence that the process is somewhat centered).
 
Shainin wins.
 
Scenario # 2 – We have a process that is in control with a Cp of 1. The spec is 100 +/- 3. The process is centered at 100 after doing what the SPC texts suggests of 20 – 30 samples, compute centerline and control limits. After a tool change, the process shifts to a mean of 99.5.
 
SPC (x-bar, r) – – The control limits (set following Wheeler’s suggestions) are 100 +/- 1.35. Go create 100 five piece samples using rational subgroups and all of the other things that are proper for SPC that also reflects the shift. What you will see is failure for either 2 of 3 beyond 2 or 4 of 5 beyond 1 on average after about 10 samples.
 
Pre-control – Process turns on okay and fails for two yellows on average at about 20 samples.
 
SPC wins
 
Scenario # 3 – We have a process that is in control with a Cp of 2. The spec is 100 +/- 3. The process is centered at 101.5 after doing what the SPC texts suggests of 20 – 30 samples, compute centerline and control limits.
 
SPC (x-bar, r) – The control limits (set following Wheeler’s suggestions) are 101.5 +/- 0.67. Go create 100 five piece samples using rational subgroups and all of the other things that are proper for SPC. What you will see with all 8 of the rules Minitab provides turned on is 0, 1, or 2 out of control points for each sample of 100. I just did it about 20 times.
 
Pre-control – Using Shainin’s rules for dividing the spec into four pieces, I can’t get the process to turn on! Shainin demands five consecutive pieces in the green (evidence that the process is somewhat centered).
 
Shainin wins.
 
Scenario # 4 – We have a process that is in control with a Cp of 2. The spec is 100 +/- 3. The process is centered at 100 after doing what the SPC texts suggests of 20 – 30 samples, compute centerline and control limits. After a tool change, the process shifts to a mean of 99.5.
 
SPC (x-bar, r) – – The control limits (set following Wheeler’s suggestions) are 100 +/- 1.35. Go create 100 five piece samples using rational subgroups and all of the other things that are proper for SPC that also reflects the shift. What you will see is failure for either 2 of 3 beyond 2 or 4 of 5 beyond 1 on average after about 3 or 4 samples.
 
Pre-control – Process turns on okay and generally does not fail.
 
SPC wins
 
The point is that it is not clear cut that SPC is superior for a couple of reasons.
 
1)      SPC is taught to go collect 20 or 30 samples of 4 or 5, compute centerline and control limits…. So what happens when we desire to have the process targeted? SPC, the way it is normally taught does not address the issue. X-bar is where x-bar is and as long as the Cpk is minimally acceptable, it does not get addressed. If Taguchi is right, this is wrong.
2)      SPC is generally implemented using less than the full complement of test that are looking for a shift or drift in the mean.
3)      SPC is generally not reacted to. Out of control points are dismissed by people who think they are doing the right thing for the customer.
 
Can Pre-control increase variation? Yes, when two conditions exist. The first condition is the Cp > 1 and we use the rules with regard to the specifications.
 
Are there other rules that can be used? Yes, we can set up the zones based on knowledge of capability. When we do this, SPC is minimally better if you use all the rules for detecting shifts and drifts and respond appropriately. Note that it is minimally better, so the real question is which would be used properly? Within SPC, EWMA is minimally better than x-bar, r and again the question would be one of the implementation.
 
Bottom line is that if a facility that does not do SPC completely, including proper reaction to all out of control conditions; I can come into that facility and smoke you using Pre-control with the rules modified for capability instead of specs.
 
 BTW – who do you think the story is about?

I agree on B vs C. I also agree on Component and Variable Search.
Tell me how you set up and maintain your control charts.

Great celebrations came with this renewed success. The greatest of all was an award in the memory of a rodeo cowboy who was also a secretary interested in commerce and quality. This company won the award at its inception thanks to insight of a former child actor who started his career in a movie about a once great man trying to be great again. He now acted as the head of Quality at a distant outpost of the once great company. Must have thought he was back in the movie, None the less, the award was won. Great celebrations ensued and all sorts of jokers took credit, included the Red X PhD from the great university in the desert. Seven years in and life is good. The president relinquished his position to the head of his least profitable business (his computer business was losing money for those who want to split hairs), and the enterprise was declared world class. For those of you who do not know the true meaning of world class, it means we are within five years of falling flat on our rear ends. For evidence of this, look at this once great company (again), Xerox, Compaq, and many others. With the knowledge of being world class, learning ceased. Those who had fun for the previous seven years, well let’s just say they did not enjoy the new politics that looked suspiciously like the politics of ten years before. The change agents left in droves. The self promoters left as well.
 
The change agents wandered aimlessly for the next three and one half years, falling for such nonsense as the “Compaq culture” (translated into 1991 lingo means align yourself with the guy from the inventor of the transistor that was also a failed company by now). Some self promoters went to a great, but obscenely inefficient,  European equipment maker following a man who only existed at 50,000’ or more. All succeeded, but none had fun or left anything sustainable..