科研之友
微信
新浪微博
Facebook
分享链接
摘要
Novel statistical analysis and machine learning algorithms are proposed for the deconvolution and interpretation of Raman spectra of silicate glasses in the Na2O-CaO-SiO2 system. Raman spectra are acquired along diffusion profiles of three pairs of glasses centered around an average composition of 69.9 wt.% SiO2, 12.7 wt.% CaO, and 16.8 wt.% Na2O. The shape changes of the Raman spectra across the compositional domain are analyzed using a combination of principal component analysis (PCA) and sparse non-negative matrix factorization (NMF). This procedure yields components accounting for the observed changes, as well as their mixing proportions, without any direct prior assumption as to their actual shape, number or position. These methods are applied separately to the Q band (wavenumbers in the range 850-1400 cm(-1)), the main band (200-850 cm(-1)), and to the whole spectra (200-1400 cm(-1)). Compositional profiles obtained by Electron Probe Micro-Analysis (EPMA) are then used to relate spectral components to structural entities. Spectral components extracted from a Q band analysis and a complete spectral analysis show significant similarities both in terms of shape of the components and their mixing proportions. This result implies a link between Q(n) species and the shift of the medium-range network features in the main band of the Raman spectra. To illustrate the possibilities of the method, a linear regression model is used to relate the proportions of spectral components derived from the Raman spectra to chemical composition. This model can be used to determine the composition of different glasses inside the investigated compositional domain with reasonable accuracy.
- 出版日期2015-11-15
