Design of Experiment
11 likes972 views
This document discusses the design of experiments methodology. It begins by outlining the general research process and different approaches to experimentation such as trial and error, one factor at a time, and design of experiments. It then provides more details on design of experiments, including why it is used, important terminology, examples, different types of designs, and software for designing experiments. The document guides the reader through the process of designing an experiment using Response Surface Methodology and the Design-Expert software package.
Design of Experiment
- 2. Research Process 1. Formulating the Research Problem 2. Extensive Literature Survey 3. Developing the Research Hypothesis 4. Designing the Research 5. Collecting the Research Data 6. Statistical Analysis of Data 7. Interpretation 8. Result Presentation 9. Report/Paper Writing
- 3. Approaches to Experimentation 1. Trial and Error Method • Multiple attempts are made to reach a solution 2. One Factor at a Time (OFAT) • One factor change at a time while others are kept fixed 3. Design of Experiments (DOE)
- 4. Design of Experiment (DoE) • Statistical techniques for improving process/product designs • Maximum realistic information with the minimum number of well designed experiments • Example of software Design Expert Minitab
- 5. Why DOE? Reduce time Minimum sample size Improve performances & Reliability Less resources Interaction between factors Perform evaluation of materials and system
- 6. Important Terminology • Factors – Input variables (control Or uncontrol Factors ) • Temperature, Concentration, Contact time • Levels – Specific values of factors (inputs) • Continuous or Catergorical • Contact time (1 to 3 hours), Temp. (10 – 20 ℃), Pip Height (8 – 9 mm) • Response variable – Output of the Experiment • Adsorption Efficiency, Tensile Strength • Replication – Completely re-run experiment with same input levels – Used to determine impact of measurement error • Interaction – Possible interaction between two or more factors
- 7. Example of DOE in Real Life… Factors Levels Responses Variable Inputs Settings Outputs Sugars Beans Grind Time Cups 10 – 50 g Type A or B 1 to 4 min 1 to 4 min Example of Characteristics Taste Bitterness
- 8. DOE Process Define Problem Determine Objectives Brainstorm Design Experiment Conduct experiment & Collect Data Analyze data Interpret results Verify Predicted Results
- 9. Type of DOE 1. One Factorial 2. Full Factorial 3. Fractional Factorial 4. Screening Experiment 5. Response Surface Analysis
- 10. DOE Only one or more factors having an impact on output at different factor levels Qualitative or Quantitative Qualitative Type of material, Type of Column Quantitative Temperature, Voltage, Load
- 11. Selection Guide Design No of Factors Levels 1 Way ANOVA 1 Factorial Design (Randomized) 2 Level Factorial Level NF= 2-21 2 Minimum-Run Resolution V Characterization Design NF= 6- 50 2 Minimum-Run Resolution IV Screening Design NF= 5- 50 2 Multilevel Categorical Design CF= 1- 12 Different Level Optimal (Custom) Design NF=2- 30 2
- 12. Selection Guide Design No of Factors Levels Miscellaneous Resolution V Irregular Fraction Design 4- 11 2 Plackett-Burman Design 2-47 2 Taguchi OA Design Orthogonal array designs – L4 –L64 2
- 13. Selection Guide Design No of Factors Levels Factorial Design (Split Plot) Regular Two-Level Design 2-15 2 Multilevel Categoric Design 2- 12 Different level Optimal (Custom) Design 2- 30 (Category) 2
- 14. Selection Guide Design No of Factors Levels Response Surface (Randamized ) Central Composite Design NF =2- 50 CF = 0- 10 5 Box-Behnken Design NF= 3-21 CF = 0-10 3 Optimal (Custom) Design NF= 1-21 CF = 0-10 2 Response Surface (Split -Plot ) 5- 50 2 Central Composite Design NF =2- 21 5 Optimal (Custom) Design NF =1- 30 CF = 0- 10 2
- 15. Design- Expert Software DESIGN-EXPERT® VERSION 12 SOFTWARE TRIAL https://www.statease.com/t rial/
- 16. Design- Expert Software What are you supposed to do before you start designing your experiment with Design Expert? 1. Choose your Operating Parameters 2. Decide on your range (min and max) 3. Identify the appropriate design for your research
- 17. Design- Expert Software What are you supposed to do before you start designing your experiment with Design Expert? 1. Choose your response 2. Select your factors to be investigated 3. Select level of each factors (minimum and maximum values) 4. Identify the appropriate design for your research
- 18. Design the experiment -RSM Part 1 1. Select the program 2. Click the blank Sheet icon 3. Click RSM 4. Choose Central Composite Design (CCD) 5. Select the ‘numerical factors’ (if you have 3 factors, then you have to click 3) 6. Insert the details for low and high levels. 7. Complete response form 8. Click finish and save your file
- 19. Design the experiment -RSM Part 2 Enter the Response Data
- 20. Part 3 – Analyze the data 1. Click analysis 2. Then the response 3. Click fit summary tab (top of the screen) Sources Sequential p - value Lack of Fit p- value Adjusted R2 Predicted R2 Linear 0.29235 0.00030 0.06165 -0.3259 2FI 0.90496 0.00020 -0.04086 -0.8062 Quadratic 0.00010 0.63799 0.98570 0.9722 Suggested Cubic 0.84890 0.28939 0.98125 0.8555 Aliased Model : p< 0.05 (Significant ) Lack of fit : p> 0.05 (Not Significant) – compares Residual error with ‘Pure Error’ R2 : Near to 1 Low Standard Deviation PRESS : Low
- 21. Part 3 – Analyze the data ANOVA 1. P-values less than 0.05- model is significant , greater than 0.1 the model is not significant 2. If too many insignificant , model reduction may improve the model 3. Non significant Lack of fit is good– Model is fit 4. Adequate Precision – greater than 4 ( can use to navigate the design space ) Model : p< 0.05 (Significant ) Lack of fit : p> 0.05 (Not Significant) – compares Residual error with ‘Pure Error’
- 22. Part 4 – Examine model Graph 1. Model graph 2. 2D Contour or 3D Surface Plot
- 23. Part 5 – Numerical Optimization 1. Maximize, minimize, target , in Range or Equal to 2. Running the optimization – click Solution 3. Choose the satisfactory solution 4. Do the Confirmation Run 5. Validate the Experimental Result with the Predicted Values