Design of Experiments (DOE) is often used in industries such as medical care, biotechnology, pharmaceuticals, and high-tech manufacturing as a method to reduce the number of experiments and costs. What are the important principles and steps when conducting experimental design? What are the concepts and applications of classic screening design and full factorial design in traditional experimental methods? This article will introduce the basic principles of DOE, the concepts and application cases of classic screening design and full factorial design methods.
What is Design of Experiments (DOE)
Design of Experiments (DOE) is to design experiments by screening experimental conditions and reducing external factors to reduce the impact on the experiment. The characteristic is that it can be efficient through fewer experiments, less experimental costs and time. It can be used to collect data or find correlations and problems between factors, etc. It has a wide range of applications.
Three major drivers of experimental design
In the 1920s, Ronald Fisher introduced the concept of experimental design to agriculture. Now, experimental design DOE is not only used in agriculture, but also in medical, biotech and pharmaceutical, high-tech manufacturing and other industries. Later, in the 1950s, Genichi Taguchi published the Taguchi method, introduced orthogonal arrays, and published the concept of quality loss function in the early 1970s; in 1951, British statistician George EP. Box and K. B. Wilson introduced response surface methodology, George EP in 1960. Box continues with Hunter, J. Stuart published the two-level partial factorial design, which allowed experiments and researchers to quickly find influencing factors and introduced the concept of resolution.
Traditional experimental methods: trial and error, one factor at a time, and traditional design
- Trial and error method: based on rules of thumb and limited to simple or less variable processes
- One factor at a time: other factors are fixed, and one factor is tested in one experiment
- traditional design
- Stage1 partial factorial design
- The purpose is to screen & identify important factors. It is usually used when there are too many factors or when the process is unknown.
- Stage2 full factorial design
- The purpose is to consider whether there are main effects and other effects, which will affect the estimation accuracy; estimate the main effects and interaction effects; and understand the system characteristics.
- Stage3 reaction surface
- Optimization (optimization): finding the best settings for significant primers
- Estimate main effects, interaction effects, and nonlinear effects
- Focus on the accuracy and precision of predictions
Experimental design guidelines
- Define the problem and goals
- Identify response variables
- * important! Confirm measurement system (MSA) – too much error will have an insignificant factor effect
- Identify the number of factors and their level numbers
- Identify whether there are restrictions or other restrictions
- Create design
- Preparation before experiment
- Perform experiments
- Analyze data
- draw conclusions and decide next steps
Purpose of experimental design
Screen --> Explore --> Optimize --> Validate --> Robust
- Screening: Find the factors that most significantly affect the reaction variables. Usually there are too many factors or the process is unknown.
- Explore: Find out if there are new factors or new levels
- Optimization: Find the best setting values for significant factors that correspond to the desired response variables
- Verification: Confirm whether the system or process conforms to the expected behavior. For example, use different batches of raw materials to verify that the results are the same model, but different machines to verify whether the parameter settings can be directly applied.
- Robust: Minimize the variation of reaction variables, including control conditions and factor settings
Using JMP Six steps to conduct experiments
- Step1. Describe Define responses, factors and levels
- Step2. Specify determines which effect terms are included in the model
- Step3. Design determines whether to add center points or repetitions to the number of experiments, and provides relevant experimental evaluation items
- Step4. Collect collects data according to experimental conditions and order
- Step5. Fit Model Fit the model and find out the key factors
- Step6. Predict finds a suitable model and optimizes it according to the goal
Classic screening design VS full factorial design
(1) Classic screening design Screening Design
- Detect whether the main effect is significant, or linear effect. Although the center point can detect whether there is a surface effect, it cannot know exactly which factor it is.
- In partial factorial design, it is necessary to deal with the confounding effect. Usually the resolution is set to four or more.
- It may be that it is impossible to distinguish whether the alias term or the effect term is significant due to the cross-talk, and additional amplification experiments are needed to distinguish.
When to use classic screening design and main effects design
Case sharing: Comparison of partial factor and main effect designs
(2) Full factorial design Full Factorial Design
- Consider the level of each factor to generate each combination
- The more levels or factors, the lower the efficiency and the higher the cost.
- Advantages: All interaction terms can be estimated and considered more comprehensively than the one-factor-at-a-time and trial-and-error methods.
- Disadvantages: Too many experiments, too high cost
Case Sharing: Battery Life Experiment
Thousands of engineers participated [ DOE Introductory course]
The above is a brief introduction to the classic screening design and full factorial design in DOE experimental design.
If you want to fully understand the case application of classic screening design, welcome to watch the DOE introductory course prepared for you by JMP professional consultants.
Click below link , you can watch it for free.
