DOE Presentation
DOE Presentation
DOE Presentation
12/12/2007
Bryce Gantt Igor Uroic
Design of Experiments
CASE: Traffic Times
Bryce Gantt, Igor Uroic Agenda: Definition
Experiment Types Terminology
Procedure
Full-Factorial Experiment Graphs
Definition
Design of Experiment (DOE) is a structured method for determining the relationship between process factors (x) and the process output (Y) A structured DOE has several benefits:
Identifies important factors Establishes process stability Allows simultaneous factor interaction analysis Determines optimal operating conditions
Experiment Types
Trial and Error
Random changes to factors
Full-Factorial
Evaluate all possible combinations of factors at levels
Fractional-Factorial
Evaluate only a fraction of possible combinations Less information than a Full-Factorial
Terminology
Factor
Process variable (x)
Level
Settings at which a factor is being evaluated e.g. pressure settings (3psi, 5psi OR 10psi)
Interaction
Failure of a factor to produce identical results at ALL experiment levels (all three psi levels)
Response
A measurable process (Y)
Terminology (contd.)
Main Effect
(Avg. at high level) (Avg. at low level) Factor impact on the response variable (Y)
Interaction Effect
Combined effect of multiple factors on the response variable
Experimental boundaries
The range of all possible values
DOE Procedure
Establish and objective (e.g. minimize cost) Identify all factors (x)
Pick several factors (pick 2 or 3 are common) Select levels
Determine the sample size for each experiment Randomize order Run experiment Analyze the results
Full-Factorial Experiment
The nx experiment n=number or levels x=number of factors As the number of factors and levels increases, the complexity of the experiment increases exponentially
e.g of increasing complexity (22 vs. 33 vs. 37) 4 combinations, 27 combinations, and 2,187 combinations respectively
DOE Graphs
Main Effect
No effect
(-) Factor 1 (x) (+)
-- Strong Effect
(-) Factor 1 (x) (+)
Interaction Effect
No effect
Factor 2 (x)
Factor 2 (x)
-- Strong Effect
(-)
Factor 1 (x)
(+)
(-)
Factor 1 (x)
(+)
9
Description
Category for day of work week Time leave for work School holiday or summer break
Levels
- Beginning (M-W) - Ending (R-F) - 8:00AM - 8:30AM - Yes - No Minutes from driveway to parking lot
10
Input
B C
Output
Morning Commute
Example (contd.)
Determine the sample size
8 samples per run
Coding: -1 = low level +1) = high level
Randomize order
Standard order Trial Factor A
1 2 3 4 5 6 7 8 -1 1 -1 1 -1 1 -1 1
Coding:
Factor B
-1 -1 1 1 -1 -1 1 1
Factor C
-1 -1 -1 -1 1 1 1 1
11
Example (contd.)
Run experiment
Trial/ Run Response Factor A
1 2 3 4 5 6 7 8 -1 -1 -1 -1 -1 -1 -1 -1
Factor B
-1 -1 -1 -1 -1 -1 -1 -1
Factor C
-1 -1 -1 -1 -1 -1 -1 -1
Duration
37 31 23 32 32 20 16 28
Factor B
-1 -1 1 1 -1 -1 1 1
Factor C
-1 -1 -1 -1 1 1 1 1
Avg. R
24.24 21.80 21.97 18.42 21.93 18.75 18.60 18.63
12
Example (contd.)
Analyze the results
Important? Main effects have steep slopes. Interactions almost parallel.
13
Example (contd.)
Analyze the results
No significant factors
Main Factors
14
Minitab Procedure
Open Minitab and enter the data
15
16
17
Excel Procedure
Enter data
18
References
Dataset: http://www.omninerd.com/articles/attachments/Brandon/a50/commute_data.xls MGS 8020: http://www2.gsu.edu/~wwwkem/Lectures/DOE_Overview.pdf The Lean Six Sigma Pocket Toolbook
by M. George, D. Rowlands, M. Price, & J. Maxey
19
Questions