|
Abstract:
|
A goal of the Hanford Waste Treatment and Immobilization Plant is to make glass containing a significant fraction of nuclear waste while satisfying glass quality and processing property constraints. To do this, the Pacific Northwest National Laboratory and others have conducted years of research to quantify the property-composition relationships and glass quality prediction uncertainties. This talk describes how nonlinear logistic regression was used to predict if nepheline crystals would form based on historical glass composition data. The talk also explains how the nepheline-composition relationship was used to design an experiment focused on the most uncertain part of the composition region. Because the new experimental data was to be combined with historical data for the next phase of research, we chose an augmented space-filling design subject to several linear and non-linear constraints. However, there were no currently available methods or software to develop such a design, so we created a way to generate the design using existing capabilities in the analytical software JMP. Finally, we show how the initial regression performed on the new data and how it can be updated.
|
