Analysis of Multicomponent Ionic Mixtures Using Blind Source Separation: A Processing Case Study

Abstract

Management and remediation of complex nuclear waste solutions require identification and quantification of multiple species. Some of the species forming the solution are unknown, and they can be different from vessel to vessel, thus limiting the utility of standard calibration approaches. To cope with such limited information, we propose a procedure based on blind source separation (BSS) techniques, in particular independent component analysis and multivariate curve resolution, with a one-point calibration library. Here we show the applicability and reliability of our procedure for online measurements of aqueous ionic solutions by proposing an automatic procedure to identify the number of species in the mixture, estimate the spectra of the pure species, and label the spectra with respect to a library of reference components. We test our procedure against simulated and experimental data for mixtures with six species (water plus five sodium salts) for the case of Raman and attenuated total reflection-Fourier transform infrared spectroscopy.

Spectroscopic Quantification of Target Species in a Complex Mixture Using Blind Source Separation and Partial Least-Squares Regression: A Case Study on Hanford Waste

“…Figure shows how the nitrate–carbonate region is affected when the source is classified as a non-target species and removed. Previous work has shown a blue shift of the main Raman peak of nitrate at 1050 cm –1 toward higher wavenumbers and, thus, shifting it closer to the main carbonate peak in the presence of increasing concentrations of certain anions and cations. − A similar source was identified in our previous work using mixtures of nitrate, nitrite, carbonate, sulfate, and phosphate in water . Considering this information, the source was reclassified as an additional target species by introducing a threshold value of 0.70 for the coefficients of correlation.…”

“…Since MCR–ALS requires an initial guess, in this work, as well as previous works, we combined ICA and MCR–ALS using the results from ICA to initiate MCR–ALS. Detailed explanations and derivations of BSS techniques can be found elsewhere . To perform the ICA algorithm, we used FastICA, which maximizes the non-Gaussianity of the sources.…”

“…The Python code for BSS–PLSR can be found on GitHub: . Identifying the independent components (sources): first, BSS methods outlined in Section , ICA and MCR–ALS, are used to analyze the mixture data X and identify the independent components or sources, . The number of independent components, n̂ K , is determined using an automated analysis of the singular value decomposition of the mixture data, as reported in our previous work, which also includes more details on the ICA and MCR–ALS procedures . Typically, BSS is applied to the mixture data that need to be quantified, which in our case are the test data set, X ts .…”

“…Detailed explanations and derivations of BSS techniques can be found elsewhere. 13 To perform the ICA algorithm, we used FastICA, 10 which maximizes the non-Gaussianity of the sources. The MCR−ALS algorithm was performed using pyMCR, developed by NIST.…”

“…When applied to spectroscopy data, the goal of BSS is to approximate the signal of the original components found in the mixtures as well as their contributions to the mixture. However, since both the mixture identity and composition are unknown at the start, BSS is limited in quantifying the concentrations of the components in the mixtures. − …”