E6S-069 Good Intentions, Artifacts and Truncated Lies? - In the eye of the Cash-holder -Part 4

Intro:  Welcome to the E6S-Methods podcast with Jacob and Aaron, brought to you by E6S Industries, your source for expert training, coaching, consulting, and leadership in Lean, Six Sigma, and continuous improvement methods. In this episode number 69, “Good Intentions, Artifacts, and Truncated Lies,” we capture capabilities gone wrong in this tale of an imperfect project. This is the 4th and final part of our “In the Eye of the Cash-holder” series.  Here we go. http://bit.ly/E6S-069; http://bit.ly/E6S-iTunes

Objection 1:  In spec is good enough.  Variation doesn't matter.

Counter 1: Any variation from target results in depleted function of your customer.  Although customer specs are wide, customers often have to account for supplier variation in some other way, and end up releasing lower quality goods into the market.  Variation is evil according to taguchi, not just to customers, but to society as a whole.

Objection 3:  This is old stuff.  Everyone knows this.  Why are we bothering still?       

Counter 3: It is true that these topics are almost 100 years old.  Yet still, many industries still have processes without specifications (business processes, etc), and the idea of measuring process variation is still new and novel.

Part 4 “Good Intentions, Artifacts and Truncated Lies?”

I            Good Intentions

a.       1 process; 12 different operators; 12 different distributions

b.      Steady Eddie “Q” wins the prize. 

i.      Best centering

ii.      Best “normal” distribution.  Can’t fake that.

iii.      Best Cpk

iv.      Eddie’s secret:  Lazy and didn’t care.  Set it up and left it alone. Let natural variation occur, and didn’t chase the signal.

c.       All other “good intentioned” operators demonstrated lesser capability

i.      Created Non-normal, bimodal & skewed distributions, in a process that was naturally normally distributed.

ii.      Created statistically unstable processes.

iii.      Some causes:

1.      over adjusting the process based on the last outcome of the last part.

2.      Biasing toward USL, “more is better.” Ceramic coatings


II         Artifacts

a.       Misrepresentation of Process Capability

b.      Same data, different output of Cpk & graphs

i.      One analysis shows a Cpk of 1.22, while another shows Cpk of 22.75. 

ii.      Ppks are the same for both at .99. 

iii.      Graphical output also looks significantly different

iv.      Root Cause: Sorted data

1.      Stumbled upon this as a result of dealing with messy historical data. 

2.      Data was sorted by the output to eliminate nonsensical data  on the low and high ends of the scales

a.       Major outliers, typos

3.      Data was analyzed without sorting back into time order, not realizing the importance.

a.       Made the “Within” short term stdev look really tight.  Part-to-part is very similar when sorted by part.

v.      Key learnings: 

1.      Capability studies are in time run order to use the short-term stdev estimate

2.      If Cpk and Ppk are way out of whack, you have a data analysis artifact

a.       If red curve and black dashed curve are way different.

3.      Can easily be seen if you also perform a run chart, or capability six-pack.  The trend is a clear flag for sorted data.

4.      If anyone reports a ridiculous Cpk (above 3?, 9 sigma process), question the use of the data. 

a.       When in doubt, ask for the Ppk

III      Truncated Lies?

a.       When a distribution is cut off right at the specification, question the validity

i.      Indicator that bad parts are being reclassified as “good” when operators believe they are “close” enough

ii.      In this situation, parts below the LSL could be reworked easily.  Parts above the USL had to be stripped and completely reworked (a lot of work).

iii.      Given the normal distribution, having the tails become interrupted absolutely coincident with a specification limit (that is not a natural boundary) is completely suspicious.

Outro: Thanks for listening to episode 69 of the E6S-Methods Podcast.  Stay tuned for episode number 70, “What to Expect When Your Not Spec-ing.”  Just because there’s no spec, doesn’t mean there’s no expectation.  If you would like to be a guest on the podcast, contact us through our website.  Join our mailing list!  Subscribe to past and future episodes on iTunes or stream us live on-demand with Stitcher Radio.  Find outlines and graphics for all shows and more at www.E6S-Methods.com. “Journey Through Success”