Robust Parameter Design Using Statgraphics Centurion Dr. Neil W. Polhemus Copy right 2011 by StatPoint Technologies, Inc. 1
Robust Parameter Designs Experimental designs containing 2 types of factors: controllable factors that the experimenter can manipulate both during the experiment and during production. noise factors that can be manipulated during the experiment but are normally uncontrollable. Goal: To find robust operating conditions, i.e., levels of the controllable factors where the values of the response variables are both desirable and relatively insensitive to variation in the noise factors. 2
Two Approaches 1. Crossed arrays (Genichi Taguchi) 2 designs are created, 1 for the controllable factors and 1 for the noise factors. Taguchi called these designs inner and outer arrays. An inner array is created for the controllable factors. At each location of the inner array, the outer array is performed. 2. Combined arrays (Doug Montgomery s response surface method) A single design is created for both the controllable and noise factors. Interactions between controllable factors and noise factors reveal location of robust operating conditions. 3
Example #1 Crossed Arrays Myers, Montgomery, and Anderson-Cook (2009) provide an example aimed at minimizing the rate of soldering defects. Response: Y = defects per million joints Controllable factors: X1 = temperature X2 = speed X3 = flux density X4 = preheat temperature X5 = wave height Noise factors: Z1 = variation in temperature Z2 = variation in conveyor speed Z3 = type of assembly 4
Step 1: Specify response 5
Step 2: Specify factors 6
Step 3: Select design 7
Step 3: Select design (cont.) First select the design for the controllable factors. 8
Step 3: Select design (cont.) Column 1 2 3 4 5 6 7 Run 1 1 1 1 1 1 1 1 2 1 1 1 2 2 2 2 3 1 2 2 1 1 2 2 4 1 2 2 2 2 1 1 5 2 1 2 1 2 1 2 6 2 1 2 2 1 2 1 7 2 2 1 1 2 2 1 8 2 2 1 2 1 1 2 9
Step 3: Select design (cont.) 10
Step 3: Select design (cont.) Next select the design for the noise factors. 11
Step 3: Select design (cont.) Column 1 2 3 Run 1 1 1 1 2 1 2 2 3 2 1 2 4 2 2 1 12
Step 3: Select design (cont.) 13
Step 8: Analyze data Signal-to-Noise Ratio (Smaller the Better) - For situations in which the response is to be minimized, Taguchi proposed analyzing SNR S 10log n = number of rows in outer array n i1 y 2 i n n i1 y n 2 i y 2 1 1 s n 2 Filename: solder2.sgx 14
Step 8: Analyze data (cont). Standardized Pareto Chart for defects(snrs) C:flux density + - A:temperature E:wave height D:preheat temperature B:speed 0 4 8 12 16 20 24 Standardized effect 15
defects(snrs) Step 8: Analyze data (cont.) Main Effects Plot for defects(snrs) -41-43 -45-47 -49 temperature speed flux density wave height preheat temperature 16
Step 9: Optimize response 17
wave height Step 9: Optimize response (Cont.) Estimated Response Surface Mesh speed=7.2,preheat temperature=200.0 0.6 0.58 0.56 0.54 0.52 0.5 480 0.98 485 490 0.96 495 500 505 temperature 510 0.9 0.92 0.94 flux density 1 defects(snrs) -50.0-48.8-47.6-46.4-45.2-44.0-42.8-41.6-40.4-39.2-38.0 18
Example #2 Combined Array Myers, Montgomery, and Anderson-Cook (2009) provide an example optimizing the quality of color television images. Response: Y = reception quality in decibels Controllable factors: X1 = filter tabs X2 = sampling frequency Noise factors: Z1 = image bits Z2 = voltage 19
Step 1: Specify response 20
Step 2: Specify factors 21
Step 3: Select design 22
Step 3: Select design (cont.) Select a single design for both the controllable and noise factors. 23
Step 3: Select design (cont.) 24
Step 3: Select design (cont.) 25
Step 4: Select model The default model has two-factor interactions and quadratic effects for the 3-level factors. 26
Step 8: Analyze data Standardized Pareto Chart for reception quality C:image bits A:filter tabs B:sampling frequency D:voltage BC AB AC BD AD AA BB CD + - 0 10 20 30 40 Standardized effect Filename: tvsignal.sgx 27
reception quality Step 8: Analyze data (cont.) Interaction Plot for reception quality 37 sampling sampling 33 frequency=6.25 frequency=13.5 sampling frequency=13.5 29 25 21 17 5.0 21.0 sampling frequency=6.25 filter tabs Reception quality is good at high sampling frequency, or at low sampling frequency and low filter tabs. 28
reception quality Step 8: Analyze data (cont.) Interaction Plot for reception quality 38 35 image bits=256.0 image bits=256.0 image bits=512.0 32 29 26 23 20 image bits=512.0 6.25 13.5 sampling frequency Reception quality is less affected by changes in image bits at high sampling frequency. 29
reception quality Step 8: Analyze data (cont.) Interaction Plot for reception quality 38 35 32 image bits=256.0 image bits=512.0 image bits=256.0 29 26 23 20 5.0 21.0 image bits=512.0 filter tabs Reception quality is less affected by changes in image bits at low filter tabs. 30
Step 9: Optimize response Desirability on a scale of 0 to 1, how desirable are the values of the response variables at a selected combination of the controllable factors. In this case, desirability is a combination of the mean response and the standard deviation of the response at that combination. The transmission of error formula is used to estimate the std. deviation: se r i1 y( x, z i 2 z) z 2 31
Analysis Options 32
Step 9: Optimize response (cont.) Transmission of error formula used to estimate the std. deviation of the response at the optimal conditions. se r i1 y( x, z i 2 z) z 2 33
Desirability Step 9: Optimize response (cont.) Desirability Plot image bits=512.0,voltage=200.0 1 0.8 0.6 0.4 0.2 0 12.2 14.2 Desirability 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 4 8 12 16 filter tabs 20 24 6.2 10.2 sampling frequency 8.2 34
sampling frequency Step 9: Optimize response (cont.) Overlay Plot 14.2 12.2 1.5 35.0 3.0 4.5 30.0 6.0 25.0 7.5 reception quality Std. deviation 10.2 8.2 6.2 0 4 8 12 16 20 24 filter tabs 9.0 20.0 10.5 15.0 12.0 10.0 35
More Information (1) Myers, R. H., Montgomery, D. C. and Anderson-Cook, C. M. (2009). Response Surface Methodology: Process and Product Optimization Using Designed Experiments, 3 rd edition. New York: John Wiley and Sons. (2) Statgraphics Centurion PDF file: Doe Wizard Inner-Outer Arrays (3) Statgraphics Centurion PDF file: Doe Wizard Robust Parameter Designs (4) www.statgraphics.com 36