See how to:
- Understand the sequence of basic DOE capabilities (set up design, save/load factors and responses, simulate responses, select D- A- I- or Alias-Optimal criteria, ID number of starts, add interaction terms, modify number of JMP-recommended runs, run the design, make the design table)
- Understand the basic design types (classical, mixture, optimal, supersaturated screening, definitive screening, fractional factorial, augmented, covering arrays, space filling, Balance Incomplete Block (BIBD)
- Understand added value of using JMP experimental design options over one factor at a time (OFAT) approaches
- Understand basic terminology (randomization, replication, orthogonality, interactions, blocking)
- Build a Classical Full Factorial design and model for confirmatory analysis
- The goal: design the product to have lower fat and maintain adequate sheet strength
- Create a full factorial design with two levels, no center points, no replicates
- Build the model using the JMP DOE interface
- Evaluate the Classical Full Factorial design
- Determine if number of runs and factors are sufficient
- Examine Power, where values closest to one indicates that if the displayed factors are active and important, then the probability is high that you will learn something about them when running the experiment and building the model
- Examine Prediction Variance Profile, which shows midpoints of designs, and where low values are best
- Examine Color Map Correlations, where value of 0 means factors are completely independent
- Examine D-, G- and A- Efficiency, which compares your design to the best D-, G- or A-Optimal design, respectively
- Examine the Prediction Variance for the Fraction of Design Space Plot, where low values are best
- With the same goal of designing a product with lower fat and adequate sheet strength, build a Custom Full Factorial Design and modify it to include quadratic terms using Response Surface Methodology (RSM)
- Use 5 factors, include 2-way interactions
- Understand Custom Design factor options (continuous, discrete numeric, categorical, blocking, covariate, mixture, constant, uncontrolled)
- From the DOE interface, construct and evaluate a model based on Custom Full Factorial design
- Examine Full Factorial pattern, which summarizes the values for each factor and where minus sign (-) indicates factor with low value and plus sign (+) indicates high value.
- Examine Summary of Fit for each response to see if the mean explains the result better than the model
- Remove terms that are >.05, unless it is a main effect related to a higher order term that is significant
- Assure that the Residual by Predict Plot is random
- Assure Probability>F is < .05, which indicates that the the model predicts better than the mean
- Optimize factors using Prediction Profiler Desirability function to attempt to match targets for fat and sheet strength
- Save values of optimization attempts using Maximize and Remember option
- Simulate optimized model to try to find more experiments to run to help determine best settings
- Use JMP Pro to compare Full Factorial and OFAT designs and determine Full Factorial at optimized settings is the best experiment to run to yield the required fat and sheet targets
Note: Q&A is included at times 27:18 and 50:06.