An Uncertainty Quantification Case Study Using Space-Filling Designs With Mixture Variables for Deterministic Data
Sep 7, 2017 1:15 PM
Melissa Jablonski, Statistician, US Army ARDEC Douglas Ray, PhD, Statistician, US Army ARDEC Edward Cooke, Chemist, US Army ARDEC
This presentation will focus on a deterministic computer experiment containing seven mixture type variables. Although the Army application for this data cannot be shared in a public setting, a bread baking analogue will be shared to make the data more relatable as the process for designing the experiment, fitting models to each of the response variables, visualizing the results, and narrowing down the design space is presented interactively with JMP. Specific topics that will be covered include using mixture constraints within the Space-Filling Design platform, fitting Gaussian process models, visualizing and interpreting the results (including using the Prediction Profiler, ternary surface plots and data filtering), and finding an optimum subset of the design space satisfying certain response criteria where the design of a follow-up physical experiment should focus.