1 / 69

Design and Analysis of Experiments

Design and Analysis of Experiments. Dr. Tai- Yue Wang Department of Industrial and Information Management National Cheng Kung University Tainan, TAIWAN, ROC. Two-Level Fractional Factorial Designs. Dr. Tai- Yue Wang Department of Industrial and Information Management

seamus
Télécharger la présentation

Design and Analysis of Experiments

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Design and Analysis of Experiments Dr. Tai-Yue Wang Department of Industrial and Information Management National Cheng Kung University Tainan, TAIWAN, ROC

  2. Two-Level Fractional Factorial Designs Dr. Tai-Yue Wang Department of Industrial and Information Management National Cheng Kung University Tainan, TAIWAN, ROC

  3. Outline • Introduction • The One-Half Fraction of the 2k factorial Design • The One-Quarter Fraction of the 2k factorial Design • The General 2k-p Fractional Factorial Design • Alias Structures in Fractional Factorials and Other Designs • Resolution III Designs • Resolution IV and V Designs • Supersaturated Designs

  4. Alias Structures I Fractional Factorials and other Designs • Assuming that we use the following regression equation to fit the experimental results: where y is an n x 1 vector of the response X1is an n x p1 matrix β1is a p1x 1 vector • Thus the estimated of β1 via LSE is

  5. Alias Structures I Fractional Factorials and other Designs • Suppose that the true model is where X2is an n x p2 matrix with additional variables β2is a p2x 1 vector

  6. Alias Structures Fractional Factorials and other Designs • Thus the expected parameters • The matrix is called alias matrix • The elements of A operating on β2 identify the alias relationships for the parameters in the vector β1

  7. Alias Structures Fractional Factorials and other Designs • Example: 23-1 design with I=ABC

  8. Alias Structures Fractional Factorials and other Designs • Regression model • So, for the four runs

  9. Alias Structures Fractional Factorials and other Designs • Suppose the true model is • and

  10. Alias Structures Fractional Factorials and other Designs • Now try to find A

  11. Alias Structures Fractional Factorials and other Designs • And

  12. Alias Structures Fractional Factorials and other Designs • Comparison [A] A+BC [B] B+AC [C] C+AB

  13. Resolution III Designs -- Constructing • Resolution III designs are useful for screening (5 – 7 variables in 8 runs, 9 - 15 variables in 16 runs, for example) • A saturated design has k = N – 1 variables • Examples of saturated design:

  14. Resolution III Designs -- Constructing • Case of

  15. Resolution III Designs -- Constructing • Can be used to generate factors fewer than 7 • For example,

  16. Resolution III Designs – Fold over • By combining fractional factorial designs that certain signs are switched , one can systematically isolate effects of the potential interest • This type of sequential experiments is called a fold over of the original design

  17. Resolution III Designs – Fold over • For the case of • Reversing the sign in factor D  - + + - - + + -

  18. Resolution III Designs – Fold over • Reversed effects • [A]’  A-BD+CE+FG • [B]’  B-AD+CF+EG • [C]’  C+AE+BF+DG • [D]’  D-AB-CG-EF • [-D]’  -D+AB+CG+EF • [E]’  E+AC+BG-DF • [F]’  F+BC+AG-DE • [G]’  G-CD+BE+AF • Original effects

  19. Resolution III Designs – Fold over • Assuming the three-factor and higher interactions are insignificant, one can combine the two fractions • For effect of the factor D ½ [D]+1/2[D]’ D • For effects ½ [D]-1/2[D]’ AB+C+EF

  20. Resolution III Designs – Fold over • In general, if we add to a fractional design of resolution III or higher a further fraction with signs of a single factor reversed, the combined design will provide the estimates of the man effect of that factor and its two-factor interactions • This is a single-factor fold over

  21. Resolution III Designs – Fold over • If we add to a fractional design of resolution III a second fraction with signs of all the factors are reversed, the combined design break the alias link between all main effects and their two-factor interaction. • This is a full fold over

  22. Resolution III Designs – example (7—1/9) • Eye focus, Response= time • 7 factors • Screening experiment

  23. Resolution III Designs – example (7—2/9) • STAT>DOE>Create Factorial Design • 2 level fractional (default) • Number of factor  7 • Choose 1/8 fractional

  24. Resolution III Designs – example (7—3/9) • STAT>DOE>Analyze Factorial Design • Only A, B, D are significant

  25. Resolution III Designs – example (7—4/9) • Examining the alias structure • We are not sure if A or BD, B or AD, D or AB are significant!!!!! Alias Structure (up to order 3) I + A*B*D + A*C*E + A*F*G + B*C*F + B*E*G + C*D*G + D*E*F A + B*D + C*E + F*G + B*C*G + B*E*F + C*D*F + D*E*G B + A*D + C*F + E*G + A*C*G + A*E*F + C*D*E + D*F*G C + A*E + B*F + D*G + A*B*G + A*D*F + B*D*E + E*F*G D + A*B + C*G + E*F + A*C*F + A*E*G + B*C*E + B*F*G E + A*C + B*G + D*F + A*B*F + A*D*G + B*C*D + C*F*G F + A*G + B*C + D*E + A*B*E + A*C*D + B*D*G + C*E*G G + A*F + B*E + C*D + A*B*C + A*D*E + B*D*F + C*E*F

  26. Resolution III Designs – example (7—5/9) • Note that ABD is one of the word in defining relation, do not project into a full 23 factorial in ABD • It does project into two replicates of a 23-1 design. • 23-1 is a resolution III design, too • Try fold over

  27. Resolution III Designs – example (7—6/9) • 2nd fraction: • STAT>DOE>Modify Design • Specify  fold all factor OK

  28. Resolution III Designs – example (7—7/9)

  29. Resolution III Designs – example (7—8/9) • Collecting data • STAT>DOE>Analyze Factorial Design

  30. Resolution III Designs – example (7—9/9) • Though B, D, BD, and AF are significant, B and D are distinguishable • BD is aliased with CE and FG • AF is aliased with CD and BE. A + B*C*G + B*E*F + C*D*F + D*E*G B + A*C*G + A*E*F + C*D*E + D*F*G C + A*B*G + A*D*F + B*D*E + E*F*G D + A*C*F + A*E*G + B*C*E + B*F*G E + A*B*F + A*D*G + B*C*D + C*F*G F + A*B*E + A*C*D + B*D*G + C*E*G G + A*B*C + A*D*E + B*D*F + C*E*F A*B + C*G + E*F A*C + B*G + D*F A*D + C*F + E*G A*E + B*F + D*G A*F + B*E + C*D A*G + B*C + D*E B*D + C*E + F*G

  31. Resolution III Designs – Fold over • To find the defining relation for a combined design, one can assume that the first fraction has L words and the fold over fraction has U words. • Thus the combined design will have L+U-1 words used as a generators.

  32. Resolution III Designs – Fold over • For example, • Generators for the first fraction: I=ABD, I=ACE, I=BCF, I=ABCG • Generators for the second fraction: I=-ABD, I=-ACE, I=-BCF, I=ABCG • We have switched the signs on the generators with an odd number of letters

  33. Resolution III Designs – Fold over • The complete defining relations for the combined design are: I=ABCG=BCDE=ACDF=ADEG=BDFG =ABEF=CEFG

  34. Resolution III Designs – Fold over • Usually the second fraction are different from the first fraction in day, time, shift, material, methods. • This leads to the blocking situation.

  35. Resolution III Designs – Plackett-Burman Designs • For the case of k=N-1 variables in N runs, where N is a multiple of 4, one can use fold over if N is a power of 2. • However, N=12, 20, 24, 28 and 36, The Placket-Burman is of interest. • Because these design cannot be represented as cubes, called non-geometric designs. • Two ways to generate these designs, check example 8.

  36. Resolution III Designs – Plackett-Burman Designs • Upper half: for N=12, 20, 24, and 36 • Lower half: for N=28

  37. Resolution III Designs – Plackett-Burman Designs • Example for Upper half: N=12 and k=11 Turn into the first column

  38. Resolution III Designs – Plackett-Burman Designs Shift down one row! Add “-” sign

  39. Resolution III Designs – Plackett-Burman Designs • Example for Lower Half: N=28 and k=27 Y Z X

  40. Resolution III Designs – Plackett-Burman Designs • N=28 and k=27 • Arrange the design into X Y Z Z X Y Y Z X - - - - - - - - Add “-” sign to the 28th row

  41. Resolution III Designs – Plackett-Burman Designs • Alias structure • Messy and complicated • Main effects are partially aliased with every two-factor interaction not involving itself • Non-regular design • For the case of N=12 • Projected into three replicates of a full 22 design in any two of the original 11 factors • Projected into a full 23 factorial plus a 23-2III fractional factorial

  42. Resolution III Designs – Plackett-Burman Designs • The resolution II Placket-Burman design has Projectivity 3. • It will collapse into a full factorial in any subset of the three factors.

  43. Resolution III Designs – example (8—1/7) • 12 factors • If 212-8 fractional is used, all 12 main effects are aliased with four two-factor interactions. • Additional experiments could be required • Use 20 run Placket-Burman design • Two kinds of designs, one is to follow the text and Minitab • The other is to follow Example 8 in the text.

  44. Resolution III Designs – example (8—2/7) Add “+” sign Reverse “+” and “-” sign in text

  45. Resolution III Designs – example (8—3/7) Corrected Table 8.25

  46. Resolution III Designs – example (8—4/7) Alternate P-B design for N=20

  47. Resolution III Designs – example (8—5/7) • No effect is significant according to traditional analysis.

  48. Resolution III Designs – example (8—6/7) • Use stepwise regression Stepwise Regression: y versus X1, X2, ... Alpha-to-Enter: 0.1 Alpha-to-Remove: 0.15 Response is y on 19 predictors, with N = 20 Step 1 2 3 4 5 6 Constant 200.0 200.0 200.0 200.0 200.0 200.0 X2 11.8 11.8 11.8 10.0 10.0 9.9 T-Value 2.51 2.78 3.65 4.02 7.30 7.99 P-Value 0.022 0.013 0.002 0.001 0.000 0.000 X4 9.6 12.0 12.0 12.0 12.1 T-Value 2.27 3.64 4.82 8.76 9.78 P-Value 0.037 0.002 0.000 0.000 0.000 x1x2 -12.0 -12.0 -12.0 -12.5 T-Value -3.64 -4.82 -8.76 -9.91 P-Value 0.002 0.000 0.000 0.000 x1x4 9.0 9.0 9.5 T-Value 3.62 6.57 7.54 P-Value 0.003 0.000 0.000 X1 8.0 8.0 T-Value 5.96 6.60 P-Value 0.000 0.000 X5 2.6 T-Value 2.04 P-Value 0.062 S 21.0 18.9 14.4 10.9 6.00 5.42 R-Sq 25.95 43.12 68.89 83.38 95.30 96.44 R-Sq(adj) 21.83 36.43 63.05 78.94 93.63 94.80

  49. Resolution III Designs – example (8—7/7) • Fitted model:

  50. Resolution IV and V Designs -- Resolution IV Designs • A 2k-p fractional is of resolution IV if the main effects are clear of two-factor interactions and some two-factor interactions are aliased with each other. • Any 2k-pIV design must contain at least 2k runs. Resolution IV designs that contain 2k runs are called minimal designs. • Resolution IV designs maybe obtained from resolution III designs by the process of fold over.

More Related