Sep 19 2011


Published by at 2:24 pm under Uncategorized

The Learned Gurus of statistics – Deming, Toyoda, Ishakawa, Duran, etc. all said statistics were valuable tools in manufacturing.  With the advent of computers and electronic spreadsheets; statistics jumped from the mind numbing practice of manual adding machines and complex calculations to electronic data gathering and instant answers with only a few keystrokes.  The Data Dogs came out of their closets and a whole new occupation (read another layer of infrastructure)  was born.

Unfortunately the Gurus of Statistics were brilliant mathematicians but boring writers.  Very few of their disciples completely read their books to understand the lessons.  Soon the quality folks fell in love with the concept of CpKs, PPM defect rates, DOE’s etc. but failed to comprehend the consequences of the answers.  Most were satisfied with the Cliff Notes versions.  While the Gurus extolled the virtues of statistics they also heavily emphasized it was only a tool.  With the proper application it would vastly improve profits but with improper use, it would only increase overhead costs and degrade profits.

The concept of CpK is simple.  You look at a tolerance, and then compare the variation of the process.  This variation is then statistically analyzed to give you ‘all possible outcomes’ from the process.

Let’s look at a simple example:  Rolling a pair of dice.  The ‘tolerance’ of a pair of dice is fixed.  The lowest number you can roll is 2 and the highest is 12.  The most frequent number and the average of all the combinations is 7.  In this example, our specified Mean dimension is 7 with a tolerance of +/- 5. (On careful consideration you’ll notice this tolerance is quite tight.)  Now someone decides they require at least a CpK of 1.0 meaning all possible combinations will fit within the tolerance and be centered on the mean.  This is kind of fun to do on a spreadsheet AND it seems reasonable.  You’ll find the standard deviation of all of the combinations is 2.44949.  BUT when you do the three sigma calculations you find statistically the lower 3-sigma limit is -0.3447 and the upper 3-sigma statistical limit is 14.3447.  Oops!

If you rolled a conventional pair of dice in a casino and came up with a negative number or a number higher than twelve, large gentlemen would probably have an intense discussion with you before you were tossed into the parking lot. This is the fault of believing only statistical analysis and ignoring Common Sense.  Statistics don’t lie, but amateur statisticians frequently misunderstand and draw bad conclusions.

All too often I work with clients where someone in the Temple Of Documentation required a statistical proof of robustness in their mold qualification procedures.  The molders and manufacturers made multiple runs of thousands of parts and endless mold modifications only to come up with statistics similar to rolling dice.  After having spent considerable time and money on this silliness, the Designers, Quality Geeks, or Engineers actually approve this ‘non-robust’ process as ‘not statistically qualified but acceptable’. Isn’t this blaspheming against the Sacred ‘Qualification Procedure’?

Before agreeing to CpK’s and other requirements; let these procedures see the ‘light of day’:

If the designs are already to the SPI tolerancing standards two things should be apparent. (1) The part should function successfully at both ends of the tolerance and all combinations in between.  (2) If the designer wants a tighter tolerance than the SPI standards provide for, you will have to ‘Inspect Quality In’ using a 100% sorting process. What the designer is really telling you with a CpK tighter than 1.0 is he’s afraid of his own specified tolerance scheme and believes his product will fail unless you manage to be almost exactly on the Mean dimension. (3) What value, if any, is it to the customer to send a statistical analysis with every shipment?

If the SPI Fine tolerance says +/- .002 and the customer wants a CpK of 1.33 he’s asking you to hold a tolerance of +/-.001.  Meaning you’ll have ugly yield numbers.  “But”, my clients say “This is what the customer wants, and the customer is always right!”  My response is: ‘If the customer is always right he will gladly pay for it.  BUT if the customer’s purchasing agent demands lower pricing, you’re gonna have nasty problems to deal with.’

It’s kinda fun to listen to the Data Dogs expound on the virtues of precision, see the designers blindly follow this philosophy and then listen to the buyer pull his trump card to  this talk with his ‘Fast, Cheap and High Quality’ speech.  The buyer’s only goal is to fill his pipeline with useable parts of acceptable quality, in the volumes necessary, at a reasonable price.  Everything else is wastefully expensive fluff and feathers.

Do your homework and present your case to the customer’s buyer. Buyers/managers only speak one language – money.  Designers and quality people tend to be ignorant or at least not fluent in this language. At the end of the day it is money that plays and everything else gets ejected from the game.

There’s a lesson here:  Before you quote a job, look at what it will take to ‘qualify’ the parts you’ll be molding.  Get a firm, committed, written answer to these questions from the company’s buyer.

(1) Who is paying for all these qualification runs and the production reports?

(2) Although the customer asked for it, are the demands of the statistical calculations even practical? Or will it be an expensive exercise in multiple tooling corrections where you not only have to ‘hit the bull’s eye’ but have to consistently put the next million shots from your gun exactly into the bullet hole from the first shot?

Talk to the buyer before you talk to the Techno-Geeks. Send him the SPI standards and the costs to meet his requirements. Let him make the choice. If you have the data (even from another customer’s part of the same material) to back it up, you’d be amazed what the buyer can do to streamline your qualification process and improve your profits. Ultimately the customer always pays for the qualification – either up front, or many times over buried in the part price

Use statistics wisely, they are merely a tool to use to improve your own profits.  They never were Holy Writ.  .

It’s your choice.

No responses yet

Trackback URI | Comments RSS

Leave a Reply