E6S-059 - Rule out the Ruler Part 5C- Guage R&R Variable MSA

Intro:  Welcome to the E6S-Methods podcast with Jacob and Aaron, your source for expert training, coaching, consulting, and leadership in Lean, Six Sigma, and continuous improvement methods. In this episode number 59 we close out or “Rule out the Ruler” series with part 5C during which we analyze a real-life gauge R&R study and discuss the AIAG rules.  Here we go.

***Rule out the Ruler - Measurement (Decision) System Variation***                                                                               

Part 5C- Interpretation of Crossed MSA data

I            AIAG (Automotive Industry Action Group) rules?

a.       Criticisms: Too stringent for most systems.  Dr. Donald J.Wheeler explains in-depth his criticisms of how the Crossed GR&R unfolds with the AIAG rules.

http://www.qualitydigest.com/inside/twitter-ed/problems-gauge-rr-studies.html

II         Minitab session window output (Book example)

a.       Look at the p-Values for if any factor is statistically significant

i.      If p<.05 (rule of thumb) then that factor plays a significant role in the variation of the measurement study. 

1.      In a “good” gauge the only factor we want to be significant is Part-to-Part.  All other factors will only be significant if you have a repeatability or reproducibility problem.

2.      If the “operator” is significant, it means the gauge can statistically tell the difference in who is doing the measurement by how the output is varying.

b.      % Contribution indicates how much variation in the study is coming from the parts or operators.  

i.      Higher contribution should come from the Part-to-part rather than R&R.

c.       Distinct categories is an indicator or precision. 

i.      In this example, the gauge can only (statistically) split this study variation up into 4 buckets.

1.      Can distinguish and differentiate between different buckets

2.      Cannot differentiate items within the bucket.

d.      % Study variation = % Total Variation

e.       % Tolerance = % P/T

f.       % Process = % P/TV

g.      What to do?  % Study roughly marginal, P/T poor,

i.      Sometimes these rules conflict.  What do they mean? Exercise some good judgement

ii.      Can look at each aspect, to determine where to focus improvements

1.      Repeatability problem – Operator Training, SOP/Method or Gauge Stability issue

2.      Reproducibility problem –

a.       Operator – Training, fixture or  & consistency issue;

b.      Operator*Part interaction – one operator consistently treated a particular part differently than all others.  Stands out.  Unique relationship between the two.

III      Graphical Analysis (Real-life example)

a.       Interpreting GR&R graphical outputs

i.      (top left) Bar chart % contribution

ii.      (top right)  Individual values of all measurements by part with a mean connecting line

1.      can visualize spread, grouping, and variation in residuals

iii.      (middle left)  range chart by operator and part.  Measures the spread between all the measurements of that specific part by that operator

1.      higher value means more variation.  Can pinpoint where possible errors occurred

iv.      (middle right) all values plotted by operator with mean connecting line

1.      may indicate if one operator has a consistent bias than another

v.      (bottom left) Xbar by operator and part. The mean of all parts by each operator

1.      may indicate a bias or an error with a particular part

2.      Generally looking for the same pattern with both operators

vi.      (bottom right) Operator*Part interaction plot

1.      can show an interaction (if lines cross drastically)

2.      can also show if there is a bias, if they are parallel and not overlapping.

b.      Box Plots

i.      Graphs the spread of data categories by part and operator

1.      visualize extent of varation, and where it comes from

2.      visualize if gauge is picking up differences between parts

ii.      what about this surface roughness gauge?

1.      works ok on bright parts with some statistically significant difference between operators (bias)

a.       Reality Check: So what?  What do the results mean.  Do they necessitate action? i.e. just because you find a statistically significant difference, operator-to-operator, doesn’t mean you have a practical difference that requires improvement to the measurement system

2.      does not work on matte parts. Too much variation all around.  Cannot tell if there is a bias, because it is too inconsistent. Possibly higher residuals (variation) at the M2 condition, expected to be  more rough

Outro: Thanks for listening to episode 59 of the E6S-Methods Podcast.  Stay tuned for episode number 60, where we get back to the flow with part 3 of our “Value Stream Essentials” series, Value Stream Mapping for Major Performance Gains. If you would like to be a guest on podcast, contact us through our website, on twitter @e6sindustries and join our discussions on LinkedIn.  Subscribe to past and future episodes on iTunes or stream us live on-demand with Stitcher Radio.  Leave a 5-star review while you’re there. Find outlines and graphics for all shows and more at www.E6S-Methods.com. “Journey Through Success”