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

One of the challenges associated with multicomponent mixture analysis using chemometrics models is collecting calibration data. Depending upon the number of constituents, the size of the calibration set can be quite large. In some cases, the mixtures may contain numerous species, but only a small subset is central to the process for which quantification is being undertaken. For example, nuclear waste at the Hanford site contains a large number of radioactive and non-radioactive species, which complicates remediation efforts. However, only the concentrations of a few target species may need to be quantified in real time to facilitate operation of the cleanup process. In this paper, we introduce a preprocessing procedure that reduces the need for extensive model calibration. The preprocessing framework uses blind source separation (BSS) to identify the independent components in the mixture, which is followed by a correlation to classify them as either target species (part of the critical quality attributes that need to be measured during waste processing) or non-target species. The classification is used to preprocess the original mixture data: the signals of the target components are retained, while those of the non-target components are removed. Since the preprocessed spectra only contain the target components, the spectra-to-concentration regression model can be trained with a smaller calibration set. The approach is tested for Raman and infrared spectroscopy using simulated and experimental data sets based on simulant mixtures of nuclear waste.