(415c) Decomposition of Nuclear Waste Spectra Using Blind Source Separation: Application to Hanford Waste

Authors: 
Kocevska, S. - Presenter, Georgia Institute of Technology
Grover, M., Georgia Tech
Rousseau, R., Georgia Institute of Technology
The clean-up efforts at the Hanford Site have been complicated by the nature of the waste, which includes many species and three different waste categories: supernatant liquid, saltcake solids and water-insoluble sludge [1]. Each of the 177 underground waste tanks has a unique combination of all the species and categories of waste, with lower variability for tanks in the same farm.

The supernatant liquid low-activity waste will be processed and stabilized via vitrification, as part of the Direct-Feed Low-Activity Waste (DFLAW) initiative. In our research, we use in situ infrared (attenuated total reflectance-Fourier transform infrared or ATR-FTIR) and Raman spectroscopy to measure the composition of the supernatant liquid, since in situ monitoring techniques can facilitate continuous operation of the DFLAW melter. In previous work [2], we applied Blind Source Separation (BSS) methods to quantify the composition of simpler mixtures containing six species.

In this contribution, we work with more complex mixtures, seeking to quantify the concentration of key species in the simulated waste. Examples of key nonradioactive species include nitrate and nitrite due to their prevalence in low-activity waste [3]. In order to obtain accurate estimates of the concentrations of these species, without performing lengthy calibration that span multiple levels of all the possible species, we developed a four-step procedure that couples BSS techniques with a standard calibration.

In the first two steps, we used BSS to: (1) identify all the independent components (i.e. species) in a mixture spectrum and (2) correlate the independent components with reference spectra from our library. At the end of step 2, we identified the contributions of the key species to the original mixture, which lets us proceed to: (3) correct the mixture spectrum by subtracting the additional information that pertains to the additional components. Now that we have the corrected mixture spectrum, we proceed to: (4) quantify the concentrations of the key species using a classical regression algorithm, such as Partial Least-Squares Regression (PLSR). The regression algorithm is trained with the species of interest only, thus reducing the training data set by hundreds of measurements.

The four-step procedure was applied to both simulated and experimental data, which include species of interest (nitrate, nitrite, carbonate and sulfate), as well as additional species (aluminate, phosphate, oxalate, chromate and acetate). Our research has shown that this method can be used to decompose the signal of complex mixtures to quantify specific components. The results are comparable to predictions from calibration approaches, which would have included lengthy experiments to collect spectra of all the possible combinations of major and minor components.

[1] Peterson, R. A.; Buck, E. C.; Chun, J.; Daniel, R. C.; Herting, D. L.; Ilton, E. S.; Lumetta, G. J.; Clark, S. B. Review of the Scientific Understanding of Radioactive Waste at the U.S. DOE Hanford Site. Environmental Science & Technology 2018, 52, 108381–396

[2] Maggioni GM, Kocevska S, Grover MA, Rousseau RW. Analysis of Multicomponent Ionic Mixtures Using Blind Source Separation: A Processing Case Study. Ind. Eng. Chem. Res. 2019, 58, 50, 22640-22651

[3] Howe AM. WTP Real-Time In-Line Monitoring Program Tasks 4 and 6: Data Quality and Management and Preliminary Analysis Plan. 2017

Topics: