Power Tools for Six Sigma - Putting It All Together NOTES Rev2

1

2

Diagnostic testing for the companion animal, livestock and poultry veterinarian, water (coliforms) and dairy (antibiotics) industries

We manufacture and sell chemistry, hematology, xray and other test instruments and consumables

We manufacture and sell rapid assay tests and Elisa plate assaysWe manufacture and sell rapid assay tests and Elisa plate assays

We provide diagnostic laboratory services

We develop and sell practice management software for veterinarians

3

Power tools go beyond the typical t-test and ANOVA (which are really just significance tests for differences in means)They are more complex than the fundamental one factor tests, full factorials and fractional factorials that are typically taught.They may incorporate split plots, blocking, chunk-type factors and other special design elements, but they are study designs in their own right.

4

Powerful tools are those that are: Versatile in their use; applicable to many diverse situations Easy to use Intuitively insightful; do not rely on statistical summaries, p values, etc. Compelling in their results

5

6

7

8

5-Whys on statistical engineering steroids

9

10

Some causal systems will use hybrids of the 3 primary split categories

11

12

13

14

15

16

Whats Wrong: The most common approach and works well with simple Problems and obvious (easily observed) causes.Whats Changed: The weakest question. It requires that the cause was known, measured and recorded. It is also prone to post hoc, ergo propter hoc* errors.Whats Different: Involves determining the differences between the diagnostic pairs. When used with a convergent elimination strategy it is highly effective.Whats Happening: This is the strongest question It includes the other threeWhat s Happening: This is the strongest question. It includes the other three questions as appropriate and when coupled with a convergent elimination strategy is the most effective approach for highly complex problems

17

18

Independence and replication are cousins of each other in statistical analyses.

19

20

21

22

23

24

Critical to Quality = Critical to the Customer. Only the Customer knows this.

CTQs are typically performance characteristics. These can include visual or cosmetic characteristics and reliability based times.

25

26

27

28

29

Definition of a Problem: "An undesired situation whose cause is unknown". What is wrong with What? Object, DefectState The Problem in terms of effects, not causes!

To determine effects

$$$$

To determine effects

$$$$

Define the Business Case: Why is important to solve The Problem?

Importance may be measured by:

To isolate causesTo isolate causes

Cost to The Company Cost to the Customer (may not be able to assign $) Customer Satisfaction Time (delays, time or effort to resolve or repair) Quantity or number of occurrences

30

Safety

31

32

33

34

35

37

38

The fuel lid supplier (who welded the spring to the fuel lid) had very wide specifications for the weld distance. Once the specification on location was tightened and communicated the Problem was eliminatedthe problem of too large an opening distance was also eliminated.

39

40

41

42

43

44

45

46

47

The materials could have lost potency from the time of the initial qualification testing to the final release testing due to an inherent instability of the materials, stress or other mechanism independent of the process steps. The quickest way to split process from time was to determine if the cause lay with the process or not.

48

49

Anewlotoftheantibodycoatingandtheconjugateareusedforthisexperiment.

50

The best results occur whenever the second operation (Blocking) is performed using the hand process.

51

Again new lots of materials are used

52

This experiment like any experiment may not work.But its relatively easy, quick and cheap, so failure will teach us somethingThe only way to determine if an experiment will work is to do the experiment; we cannot answer the question thru intellectual debate.At this point some members of the team saw the blot step used in the hand coating and were sure that it was the causal difference. This is known as a squirrel or shiny objectobject

53

Note that with these new materials the machine step does pass. However there is still a difference between the hand and machine steps. Two lots had passed in the previous year so an occasional passing lot is to be expected.In Problem Solving, pass-fail status doesnt matter; distinctive differences do.

54

The hand blotting vs. machine aspirate didnt seem to matter, but the nature of the machine vs. hand wash did. The machine wash was doing something destructive to the assay.

55

56

The science of the surfactant is to enable the Ab coating to be more receptive to capturing the positive sample and conjugate.One new Ab coating lot and two new conjugate lots were usedThe full standard hand and machine coating processes were used as internal controls

57

Notice the use of both controls and replication across multiple levels of all other factors is used consistently throughout the diagnostic process. This provides protection against other factor changes and ensures our results are conclusive and compelling

58

59

60

61

62

In general the term independence refers to the individual data points in a sample.

63

64

65

66

67

68

69

70

71

72

Measurement error is best determined using a crossed study. One of the exercises in this section will show the pitfalls of trying to determine measurement error using a standard observational multi-vari study.Inferences about the effect of measurement error can sometimes be made from a properly designed nested study. See the manual on Measurement Systems Analysis for MSA study designs and analysis.

73

74

75

76

77