Abstract
Partial least squares regression models were optimized for the quantification of trace elements including lanthanides (e.g., Ce, Nd), transition metals (e.g., Fe, Cr, Ni, Zr), post-transition metals (e.g., Al, Pb), alkali/alkaline earth metals (e.g., Na, Mg), metalloids (e.g., Si, As) and nonmetals (e.g., P) in uranium (U) by analyzing inductively coupled plasma-optical emission spectra. Chemical separations are commonly used to separate U from trace elements to enable highly reliable measurements by removing low lying spectral interferences from U in optical emission spectra. Here, an innovative multivariate regression approach was tested to circumvent the need for separations under relevant trace concentration ranges (20–5000 μg per g U). An I-optimal design was used to efficiently select training set samples, which were validated against several quality control samples with root mean square error of the prediction values ranging from 1% to 3% for 30 elements. The methodology was validated by the analysis of reference materials CRM 124-1 and CUP-2 and compared to partial least squares regression predictions from experimental values. The exemplar results indicate that the multivariate regression approach can account for covarying and overlapping spectral features better than standard software protocols. This unique approach provides a powerful tool for measuring trace elements in U without the time and waste associated with separations or matrix matched calibration standards and may be adapted to other systems.
