Paint, salad dressing, soap. What do they have in common? They are all mixtures. To optimize a mixture, you must change the proportions of the ingredients or components while keeping the total mass or volume the same. If this sounds like a system you are working with, watch this video to learn how to create and analyze mixture DOEs.
Q&A / Discussion
Big thanks to @Victor_G for help with the Q&A!
Q: What modeling techniques can we use for mixture DOE?
A: In a model-specified mixture design, linear regression models will test the specified model effects and make acceptable models. In model-agnostic mixture designs, machine-learning modeling methods (neural networks, SVM, Gaussian process) may also be quite helpful.
Q: How did you create the cost column?
A: The cost column is a simple column formula that sums up the cost for each component (saved as a table variable) multiplied by the proportion of that component in the mixture.
Q: How can we set a screening mixture design in Custom Design?
A: Create a custom design and add the mixture factors and the range for each factor in the formulation. For a screening design, you may only be interested in main effects, which is the default setting for this. If you want to look for non-linear blending terms, you will need to add interactions to the model.
Q: How we can plot four or five component mixtures/formulations?
A: When you add more than three variables to the ternary plot, you will get a scatter plot of ternary plots. These plots will be arranged so that two axes are two different variables; the third axis is the sum of all the other components.
Q: Can you use a definitive screening design for mixtures?
A: If you have true mixture factors, then you cannot use a DSD. However, if your formulation components are truly independent, you can use a DSD.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.