Multivariate chemometric methods and Vis-NIR spectrophotometry for monitoring plutonium-238 anion exchange column effluent in a radiochemical hot cell

• Published in: Talanta open Vol. 5; no. C; p. 100120
• Main Authors: Sadergaski, Luke R., Myhre, Kristian G., Delmau, Laetitia H.
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01-08-2022; Elsevier
• Subjects: Anion exchange; Hot cell; Kennard-stone; Machine learning; Multivariate analysis; Multivariate analysis Spectrophotometry; Neptunium; Partial least squares; Plutonium; Principal component analysis; RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEAR CHEMISTRY; Spectrophotometry; Supervised; Unsupervised; ORNL; Unsupervised; Multivariate analysis; Supervised; Neptunium; Anion exchange; Kennard-stone; Machine learning; Plutonium; Spectrophotometry; Partial least squares; Hot cell; Principal component analysis
• ISSN: 2666-8319; 2666-8319
• DOI: 10.1016/j.talo.2022.100120

•Novel unsupervised multivariate analysis data modeling method.•Effective means to guide process decisions and optimize separations.•Practical alternative to deploy optical spectroscopy for hot cell use.•Discovered neptunium and plutonium dynamic redox chemistry.
The Plutonium-238 Supply Program at Oak Ridge National Laboratory has developed the capability to inform process decisions during full-scale 238Pu anion exchange column runs using multivariate chemometrics and visible-near infrared (Vis-NIR) absorption spectroscopy. Multivariate analytical methods provided a suitable option for real-time analysis of complex neptunium (Np) and plutonium (Pu) absorption spectra. Thousands of spectra from multiple production campaigns were analyzed by principal component analysis and Kennard-Stone sample selection to select a training set composed of 60 spectra. Multivariate curve resolution–alternating least squares analysis (MCR-ALS) identified spectral components and component concentrations. A partial least squares regression (PLSR) model was built using the spectra and concentration values determined by these virtually unsupervised methods. Next, a supervised PLSR model was built by scaling the MCR-ALS–selected concentration matrix using molar epsilon values. Both PLSR models predicted the concentrations of Np(IV), Np(V), Pu(III), Pu(IV), and Pu(VI) in the column effluent and identified nearly identical product cut decisions. The time-dependent Np and Pu concentration profiles provided valuable insight into column dynamics and the predictions agreed with traditional methods including alpha spectrometry and inductively coupled plasma mass spectrometry. These results establish a resourceful avenue for modeling multicomponent systems with convoluted absorption bands without significant user input. It is particularly advantageous for production-oriented radiochemical applications in restrictive hot cell environments.
[Display omitted]