Using JMP® to Develop Glass Formulations for High-level Nuclear Waste Conditioning: Data Visualization, Statistical Analysis and Predictive Models
D. Perret, PhD, Research and Development Scientist – Atomic Energy Commission, France
Long-term storage of radioactive waste requires its stabilization into a form that will neither react nor change for extended periods of time. Glass formulation for the vitrification of high-level nuclear waste elements has been under investigation at the French Alternative Energies and Atomic Energy Commission (CEA) for many years. Besides the complexity of its formulation, nuclear glass also needs to meet requirements specific to the industrial vitrification process. Viscosity, density, electrical and thermal conductivities, and of course, long-term durability of the glass, are properties that have to be perfectly understood and controlled. As a consequence, we continuously have to deal with large amounts of data, including formulation data (glass compositions), physical and chemical properties, and data related to the vitrification process. JMP software has been recently implemented in our R&D environment by teams who develop nuclear glass formulation. The graphical platform provides very useful and easy-to-use tools such as the Scatterplot Matrix and the Mixture Profiler platforms, which enable the visualization and the analysis of large amounts of formulation data. Besides these convenient features, the main reason for using JMP is its powerful statistical analysis platform, which facilitates the comparison of glass composition domains with a high degree of complexity. For this purpose, PCA and Cluster platforms are very relevant. Since our studies are also focused on building property-to-composition predictive models, such tasks can be efficiently carried out by using the Fit Model and Stepwise platforms. Finally, considering the complexity of the glass formulation in question, we necessarily have to use a design of experiments (DOE) approach. Although the Mixture Design platform does not exactly meet all of our specifications, JMP does provide statistical information (PRESS, hat values, leverage plots) to analyze the results coming from our designs of experiments.