Glass formulation for the vitrification of high-level nuclear wastes has been under investigation for many years. Glass material has been chosen in the beginning of the 1960’s by many countries, such as Canada, France, USA, Germany and Russia, for high-level nuclear waste conditioning. Because glass is hard to dissolve and is chemically stable, radionucleides can be confined in a glass matrix for long periods of time. During the industrial process, liquid waste is calcinated and mixed with crushed glass (frit) in a furnace. Molten glass is then poured into stainless steel canisters where it solidifies into a stable form.
Beside the complexity of its formulation, nuclear glass also needs to meet requirements which are specific to the industrial vitrification process. Consequently, large amounts of data have to be processed continuously, including formulation data, physical and chemical properties, and data related to the vitrification process.
In 2010, the JMP software was implemented at the French Atomic Energy Commission by R&D teams who develop nuclear glass formulation. At the Discovery Summit Conference held in Denver in 2011, we showed how we use the JMP statistical analysis platforms to compare glass composition domains with a high degree of complexity. Considering the complexity of the glass formulation in question, a Design of Experiments (DOE) approach has to be used. In 2011, we explained why the JMP 9 statistical analysis platforms did not exactly meet all of our specifications at that time. However, since then many improvements have been made by JMP developers, and JMP 11 provides powerful methods for generating and analyzing Mixture DOEs, in order to investigate highly constrained experimental domains. During the presentation, we give some basics on mixture designs and we show how we use the JMP Pro 11 Mixture DOE platform, by taking the example of a 7-component mixture with linear constraints. We compare the former “distance-based design” approach we have been using for about ten years by using other commercial software packages, with optimal designs generated with JMP. Various optimality criteria are considered, such as D-efficiency and average variance of prediction for example.
Efforts have also been made by JMP developers to improve the Fit Model and Stepwise platforms, which enable to build even more accurate property-to-composition predictive models. During the presentation, we show how JMP enables to analyze results obtained from the JMP DOE platform, as well as results coming from external data sets. We present the example of a viscosity-composition model, and explain why the Prediction Profiler tool is very useful for this application. Finally, we explain how the JMP Multivariate platform (PCA and Clustering) is very relevant to analyze big data sets we have been generating for decades in the field of the nuclear glass formulation.
Below, you'll find papers, posters and selected video clips from Discovery Summit 2014.