introduction to design of experiments brian cunningham jennifer horner 03-26-11 ua college of...
TRANSCRIPT
Introduction to Design of Experiments
Brian CunninghamJennifer Horner03-26-11
UA College of Engineering
Agenda
• Short History: Design of Experiments (DOE)• How to use DOE to optimize the
design of a spinning parachute• In-Class DOE activity with your team
DOE: A Short History
• Design of Experiments (DOE) was first introduced in the 1920s when a scientist at an agricultural research station in England, Sir Ronald Fisher, showed how valid experiments could be conducted in the presence of many naturally fluctuating conditions such as temperature, soil condition, and rainfall.
• In the past decade or two, the application of DOE has gained acceptance in the U.S. as a valuable tool for improving the quality of goods and services.
Sir Ronald Fisher
Design of Experiments Examples
A plastic molding workshop wants to reduce injection molding rejects; performs a set of experiments which change injection pressure, mix temperature and setting time. • Analysis of the results shows a combination of
temperature and setting time as the most significant factor.
• Further experiments find the optimum combination of these.
Design of Experiments Examples
A yacht design team aims to improve speed through changing the shape of the boat's sail. • Rather than try random shapes, they identify the key
sail parameters and then design and perform a set of experiments with each factor set at two levels.
• They follow this up with multi-level experiments for the two most significant factors found in the first experiment set.
• The result: a new sail that increases speed by 5%.
Why DOE?
Minimize time and cost required to obtain the most information possible at the earliest possible stage in a product’s life cycle
Cost of changesTo product design
Production Consumptiontime
$
Preliminary design
Detaileddesign
The Spinning Parachute Company
Current best-selling model:
Fold forward
Fold backward
cut
attach paper clip here
The Spinning Parachute Company
• The Spinning Parachute Company wants to improve this product– Better customer satisfaction ratings– Improved sales
• Goal: Improve in-flight time (fall time)
The Spinning Parachute Company
Preliminary research: 3 main factors impact the flight time
A. Blade widthB. Blade lengthC. Body length
Fold forward
Fold backward
cut
attach paper clip here
Blade width
Body length
Blade length
The Spinning Parachute Company
Blade widthBlade lengthBody length
Process
Increase average fall timeDecrease standard deviation of fall time
P-Diagram* to represent the design improvement of the Spinning Parachute
“Factors” “Performance characteristics”
* P-diagram is short for Parameter diagram
Noise
The Spinning Parachute Company
• Purpose of DOE in this example: Uncover statistical relationships that connect
design factors to performance characteristics Allow the design team to select the factor
settings that generate the desired performance
The Spinning Parachute Company
The company’s engineers want to test new dimensions for the three factors:
Fold forward
Fold backward
cut
1
21
21
212
1
2
attach paper clip here
Wider blade
Longer blade
Longer body
The Spinning Parachute Company
There are 23, or 8, possible configurations to examine:Combination A, Blade
WidthB, Blade Length
C, Body Length
1 1 1 1
2 1 1 2
3 1 2 1
4 1 2 2
5 2 1 1
6 2 1 2
7 2 2 1
8 2 2 2
This is called a “full factorial analysis.”
The Spinning Parachute Company
Rather than gather the data for all eight combinations, you can obtain valuable information from considering just four carefully chosen combinations:
A
B
C
(2, 1, 2)
(1, 1, 1)
(2, 2, 1)(1, 2, 2)
Two Orthogonal Sets of Four Combinations
Combination
A B C
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1
Combination
A B C
1 1 1 2
2 1 2 1
3 2 1 1
4 2 2 2
Set 1
Set 2
Each set is called a “half factorial analysis.”
Orthogonal Set 1 Applied to the Spinning Parachute
Combination
A B C
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1
Combination
ABlade width
BBlade length
CBody
length
1 narrow short short
2 narrow long long
3 wide short long
4 wide long short
Note that for each factor, each level is tested in two combinations
Gather Data for Ten Replicates
Use Set 1, the four half-factorial combinations (treatments)Combination 1 2 3 4
(A, B, C) (1, 1, 1) (1, 2, 2) (2, 1, 2) (2, 2, 1)1.22 1.69 1.63 1.781.22 1.59 1.50 1.691.12 1.53 1.56 1.841.15 1.62 1.54 1.751.25 1.60 1.59 1.751.22 1.60 1.56 1.691.28 1.63 1.56 1.751.19 1.56 1.50 1.721.38 1.63 1.50 1.751.22 1.63 1.59 1.87
Avg. fall time 1.225 1.610 1.553 1.759Stand. Dev. 0.068 0.043 0.042 0.055
Record the fall time, in seconds
Summarize Spinning Parachute Data
Depicting this data another way:
Combination(Treatment)
ABlade width
BBlade length
CBody
length
Avg. fall time
Stand. Dev.
# of Reps.
1 1 1 1 1.225 0.068 10
2 1 2 2 1.610 0.043 10
3 2 1 2 1.553 0.042 10
4 2 2 1 1.759 0.055 10
Grand Average
1.537
Analyze Results of Spinning Parachute Experiment
• Calculate the effects of each of the 3 factors on the two performance characteristics
• Example: Factor A, Blade width: There are two combinations where A is at level 1 The average fall time when factor A is at level 1 is
(1.225 + 1.610) / 2 = 1.418 seconds Now consider factor A at level 2:
(1.553 + 1.759) / 2 = 1.656 seconds So the overall effect of varying Factor A between level 1
and level 2 is |1.656 – 1.418| = .238 seconds These results also determine the level setting that leads
to the longest fall time
Analyze Results of Spinning Parachute Experiment
Now repeat this analysis for Factors B and C
ABlade width
BBlade length
CBody
length
Level 1 Avg. Fall time 1.418 1.389 1.492
Level 2 Avg. Fall time 1.656 1.685 1.582
Effect (difference) 0.238 0.296 0.090
Optimum level 2 2 2
Optimum configuration wide long long
Representing the Results Graphically
The steepest slope between the two factor levels indicates a greater effect on the performance characteristics
1.71.61.51.41.3
1 2 Level Factor A
Blade width
1 2 Level Factor B
Blade length
1 2 Level Factor C
Body length
Grandaverage
Develop a Prediction Equation for the Best Fall Time
Max Fall Time = 1.537 + (1.656 – 1.537) + (1.685 – 1.537) + (1.582 – 1.537) = 1.849 seconds
ABlade width
BBlade length
CBody
length
Level 1 Avg. Fall time 1.418 1.389 1.492
Level 2 Avg. Fall time 1.656 1.685 1.582
Effect (difference) 0.238 0.296 0.090
Optimum level 2 2 2
Optimum configuration wide long long
Grand average = 1.537 seconds
Final Step: Run a Verification Experiment
• Set all factors at their optimal levels• Collect data for ten replicates• Calculate the average of the ten trials• Compare with the predicted value
Your Turn!
With your team:1. Carefully cut out four parachutes according to the four
combinations from the half factorial orthogonal array
2. Conduct ten experiments for each of the four combinations (treatments)
Usually need to randomize the experiments, but time = limited here Minimize the effects of other factors, such as:
Same person always drops Perform each drop using same technique Same position of chair (minimize air current variability) Same person times the fall Same cue for start and finish of fall
Your Turn! Continued
3. Collect your data: fill in the fall times for the ten x four = 40 drops
Two team members can fill in the data – preferably using the Excel spreadsheet
4. Perform analysis on your data5. Construct your prediction equation for the longest fall
time6. Carefully cut one more parachute model to your
recommended factor levels for the longest fall time7. Run ten replicates of a verification experiment to test
your prediction Goal: within 2 standard deviations of the predicted value
Conclusion
• I ask the teams to turn in or email me their data collection and analysis form
• Additional information on DOE: http://www.moresteam.com/toolbox/t408.cfm
http://www.isixsigma.com/index.php?option=com_k2&view=itemlist&task=category&id=199:teaching-doe&Itemid=156
http://www.stat.psu.edu/online/courses/stat503/01_intro/02_intro_history.htmlhttp://www.camo.com/rt/Resources/design_of_experiment.html
http://thequalityportal.com/q_know02.htm