Friday, 13 th July 2001 Of Course It is tip # 13 as well as It is Friday the 13th Why/When is Taguchi Method Appropriate?NOT.

Friday, 13Friday, 13thth July 2001 July 2001

Of CourseOf CourseIt is tip #It is tip #1313

as well asas well as It is Friday the 13thIt is Friday the 13th

Why/When is Taguchi Method Appropriate?NOTNOT

When Taguchi Method is NOTNOT Appropriate

Tip #13Tip #13

Friday, 13Friday, 13thth July 2001 July 2001

• No NNooIIssEE

– When you can notnot think of NNooIIssEE that can be

included during the experiment

• NNooIIssEE can be included during experiments

– but can notnot think of Control Factors that have strong

correlation to the NNooIIssEE

– but can notnot effectively capture the effects of NNooIIssEE


No No NNooIIssEE

1. IDENTIFY THE MAIN FUNCTION, SIDE EFFECTS, AND FAILURE MODE

2. IDENTIFY THE NOISE FACTORS, TESTING CONDITIONS, AND QUALITY CHARACTERISTICS

3. IDENTIFY THE OBJECTIVE FUNCTION TO BE OPTIMIZED

4. IDENTIFY THE CONTROL FACTORS AND THEIR LEVELS

5. SELECT THE ORTHOGONAL ARRAY MATRIX EXPERIMENT

6. CONDUCT THE MATRIX EXPERIMENT 7. ANALYZE THE DATA, PREDICT THE OPTIMUM LEVELS

AND PERFORMANCE 8. PERFORM THE VERIFICATION EXPERIMENT AND PLAN

THE FUTURE ACTION

When Taguchi Method is NOTNOT Appropriate in each of the 8-STEPS8-STEPS

• Main FunctionMain Function

– When you have no clue no clue as to what is the “Ideal Final

Result” (the ‘distance’ between the ‘current’ result and IFR gives the necessary ‘boldness’ to vary Control factor levels widely enough to exploit non-linearities)

• Side Effects Side Effects – When you can notnot think of Side-Effects

• you can notnot think of NNooIIssEE that can cause such side-effects


Step #1 : Main Function / Side EffectsStep #1 : Main Function / Side Effects

• Include NNooIIssE ???E ???


included during the “experiment”“experiment”


included during the “measurements”“measurements”

– When you can notnot think of NNooIIssEE that is analogous

to “aging”“aging” or “slow degradation”“slow degradation”

• during the “experiment” “experiment” or “measurements”“measurements”


Step #2 : IncludingStep #2 : Including N NooIIssEE

• NNooIIssEE can be includedincluded during experiments

– but but can notnot think of Control Factors that have strong

correlation to the NNooIIssEE

– but but can notnot effectively capture the effects of NNooIIssEE• External NNooIIssE E (in explicitly added NoIsE Factors)• Internal NNooIIssE E (in Control Factors)

– butbut do not not wish to increase either the • experimental effort • experimental resources


Step #2 : Capturing effects of Step #2 : Capturing effects of NNooIIssEE

• When you can notnot think of Quality Quality CharacteristicsCharacteristics that closely represents the “energy energy transfer”transfer” mechanism in the main function

• When the Quality CharacteristicsQuality Characteristics can not not be quantitatively measured

• When the Quality CharacteristicsQuality Characteristics is notnot monotonous (and has ‘phase-transitions’ or represents a ‘multiple valued function’)


Step #3 : Quality CharacteristicsStep #3 : Quality Characteristics / Objective Function

• When you can notnot think of “Variations”“Variations” in quality quality

CharacteristicsCharacteristics as being important.

• In other words, you are able to give importance only to the

“mean”“mean” value

• When you are interested only in improving the “mean”“mean”, or or

even worseeven worse, you are interested only only in studying the factor

effects (on “mean”“mean”)

• When you are notnot interested in identifying Control Factors – which help reduce the “Variance”“Variance”

– which help adjust the “mean”“mean”


Step #3 : Quality Characteristics / Objective FunctionObjective Function

• When you can think of only ‘one’one’ (desirable) Quality CharacteristicsQuality Characteristics and can notnot think of another (desirable or undesirable)

• When you can notnot think of two “contradictory”“contradictory” requirements i.e. Quality CharacteristicsQuality Characteristics (While Taguchi Method is capable of ‘improving’‘improving’ both)

• When you are not not able to give priority to “Tomorrow’s Problem” (reducing VarianceVariance)

and end up giving priority to “Today’s Problem” (improving “Mean” “Mean”)


Step #3 : Quality Characteristics / Objective FunctionObjective Function

• When you can think of Quality CharacteristicsQuality Characteristics that

have more to do with meanmean like smaller-the-bettersmaller-the-better

oror Larger-the-BetterLarger-the-Better and can notnot think of any

other Quality CharacteristicsQuality Characteristics that has to do with

variancevariance like Nominal-the-bestNominal-the-best

• When there is no need or scope of finding an

adjustment factoradjustment factor (defined as the control factor

that has negligiblenegligible effect on variancevariance and largelarge

effect on meanmean)


Step #3 : No Need to determine an Adjustment Factor

• When you can notnot think of Control Factors that

are strongly correlated to NNooIIssEE Factors

• When the number of Control Factors is not even

twicetwice the number of NNooIIssEE Factors (This is a ‘thumb’ rule – originating from the assumption that at least one of the two control factors will have a favorable and strong nonlinearity that will

help reduce the effect of NNooIIssEE on the Quality Quality CharacteristicsCharacteristics))


Step #4 : Number of Control Factors and NNooIIssEE Factors

• When Control Factors are chosen correctly (in the sense that these are strongly correlated to NNooIIssEE Factors as well as have strong effect on Quality CharacteristicsQuality Characteristics) but but the levels are not not “wide apart”, with the result that the nonlinearity is not fully exploited (ending up in getting only sensitivity)

On the other hand,On the other hand,• When the Levels of one one of the Control Factors are so so

widely separatedwidely separated that only that control factor dominates (and other control factors show less than 5% effect)

• For example : Temperature in a bio-culture growth has levels of 25ºC, 37º C and 50º C

• This will dominate over all other control factors


Step #4 : Control Factors Levels (t o o w i d et o o w i d e or too too narrownarrow)

• When you can notnot guarantee that all the Control Factors are indeed orthogonalorthogonal to each other and

you have chosen an orthogonal array that does not not allow study of all suspected interactions

• When the number of Control Factors and the

chosen OA is such that there are nono degrees of freedom left for estimating errorleft for estimating error (this forces one to declare control factors with less than 15% effect to be pooled as error)


Step #5 : Select the “inner” Orthogonal Array

• When the OA selected for NNooIIssEE factors (also called the

‘outer’‘outer’ array) is biggerbigger than the main OA (also called

the ‘inner’ array) for Control Factors. (The main idea behind using an ‘outer’ OA is to reduce the number of testing conditions and a ‘bigger’ array defeats defeats this main purpose).

• While the ‘outer’ array primarily gives the desired ‘worst ‘worst

case’case’ conditions, it should not lead to ‘failure’‘failure’ of the

experiment. (‘failure’ could be defined as – not able to

quantitatively measure Quality Characteristics or – causing

damage/breakdown of the process equipment)


Step #5 : Select the “outer” Orthogonal Array

• When the experimental conditions (other than the combinations of control factors that appear in the

“inner” or “outer” OA’s) can notcan not be maintained over the entire Matrix experiment

• When NNooIIssEE can not not be effectively captured on/in the samples or during the measurements

• When all all experiments are not not satisfactorily completed (even “one” lessless would give incorrectincorrect calculation of factor effects and predictions)


Step #6 : Conduct the Matrix experiment based on “inner” and “outer” OA’s

• Zero-Reading for a Larger-the-better type S/N Ratio

or identical readings for Nominal-the-best type S/N

Ratios (both give rise to “division by zero” when

evaluating the above mentioned S/N ratios)– If you get oneone measurement less than the “detection “detection

sensitivity”sensitivity” or multiplemultiple measurements within the “measuring accuracy”“measuring accuracy” of the measuring apparatus

In fact, including NNooIIssEE helps here, the measurements

becomes larger than the least-count or measuring accuracy


Step #6 and #7 : Make the Measurements and calculate the S/N Ratios

• When the confirmation experiments give results that are notnot close to the predicted results (i.e. are not within the prediction error) Some important control factor is not chosen

Some NNooIIssEE factor that has a dominant effect NNooIIssEE is not captured effectively There is no control factor that has strong correlation to NNooIIssEE

Interaction between Factors : There isis interaction between two dominant control factors and it has not been studied or the chosen OA does not allow this interaction to be studied


Step #8 : Conduct the Confirmation / Verification Experiments

More Tips More Tips Links belowLinks below

16. Taguchi Method

1st Priority : VVaarriiaanncce e RReedduuccttiioonn

2nd Priority : Factor EffectsFactor Effects

15. “inner”“inner” L9 array with “outer”“outer” L4 and L9 NoIsE arrays

14. Taguchi Method

““inner”inner” L18 array with “outer”“outer” L4 and L9 NoIsE arrays

13. Taguchi Method

Why/When is Taguchi Method not not Appropriate?

Friday, 3rd Aug 2001

Friday, 27th July 2001



Tips 12, 11, 10 Tips 12, 11, 10


12. Taguchi Method

““inner”inner” L8 array with

“outer”“outer” L4 and L9 NoIsE arrays

11. Taguchi Method

Useful at ALL Life-stagesALL Life-stages of a Process or Product

10. Taguchi MethodPerforms Process “centering”“centering” or “fine tuning”“fine tuning”


Friday, 29th June 2001

Friday, 22nd June 2001

Tips 9, 8, 7 Tips 9, 8, 7


9. Taguchi Method

Identifies the “right” NNooIIssEE factor(s)

for Tolerance DesignTolerance Design

10. Taguchi Method

Finds best settings to optimize TWO quality characteristics Simultaneously

7. Taguchi Method

When to select a ‘Larger’ OA to perform “Factorial Experiments”



Friday, 1st June 2001

Tips 6, 5, 4 Tips 6, 5, 4


6. Taguchi Method Using Orthogonal Arrays for Generating Balanced Combinations of NoIsE Factors

5. Taguchi Method Signal-to-Noise Ratio for Quality Characteristics

approaching IDEAL valueapproaching IDEAL value

4. Taguchi Method improves " quality “ at all the life stages

at the design stage itselfthe design stage itself

Friday, 25th May 2001



Tips 3, 2, 1 Tips 3, 2, 1


3. Taguchi Method Appropriate for Concurrent EngineeringConcurrent Engineering

2. Taguchi Method can study Interaction

between Noise Factors Noise Factors and Control Control FactorsFactors

1. Taguchi’s Signal-to-Noise RatiosSignal-to-Noise Ratios are in

Log formLog form


Friday, 27th April 2001

Friday, 6th April 2001

end

Friday, 13 th July 2001 Of Course It is tip # 13 as well as It is Friday the 13th Why/When is Taguchi Method Appropriate?NOT.

Documents

noise factorsinternal

appropriateno noise

effectsinclude noise

effects of noiseexternal

thwhen taguchi method

effectswhen taguchi

main functionwhen

factor effects