Quantifying Impurity Effects on the Surface ... - ACS Publications

Jun 28, 2019 - level.6 While the specifications for U ore concentrates are stringent, trace ... incorporation and the final concentration of the impur...
0 downloads 0 Views 10MB Size
Download PDF
Article Cite This: Anal. Chem. XXXX, XXX, XXX−XXX

pubs.acs.org/ac

Quantifying Impurity Effects on the Surface Morphology of α‑U3O8 Alexa B. Hanson,† Rachel Nicholls Lee,† Clement Vachet,‡ Ian J. Schwerdt,† Tolga Tasdizen,‡ and Luther W. McDonald IV*,† †

University of Utah, Department of Civil and Environmental Engineering, Nuclear Engineering Program, 201 President’s Circle, Salt Lake City, Utah 84112, United States ‡ Scientific Computing and Imaging Institute, 72 South Central Campus Drive, Room 3750, Salt Lake City, Utah 84112, United States Downloaded via UNIV OF SOUTHERN INDIANA on July 31, 2019 at 13:45:10 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

S Supporting Information *

ABSTRACT: The morphological effect of impurities on α-U3O8 has been investigated. This study provides the first evidence that the presence of impurities can alter nuclear material morphology, and these changes can be quantified to aid in revealing processing history. Four elements: Ca, Mg, V, and Zr were implemented in the uranyl peroxide synthesis route and studied individually within the α-U3O8. Six total replicates were synthesized, and replicates 1−3 were filtered and washed with Millipore water (18.2 MΩ) to remove any residual nitrates. Replicates 4−6 were filtered but not washed to determine the amount of impurities removed during washing. Inductively coupled plasma mass spectrometry (ICP-MS) was employed at key points during the synthesis to quantify incorporation of the impurity. Each sample was characterized using powder X-ray diffraction (p-XRD), high-resolution scanning electron microscopy (HRSEM), and SEM with energy dispersive X-ray spectroscopy (SEM-EDS). p-XRD was utilized to evaluate any crystallographic changes due to the impurities; HRSEM imagery was analyzed with Morphological Analysis for MAterials (MAMA) software and machine learning classification for quantification of the morphology; and SEM-EDS was utilized to locate the impurity within the α-U3O8. All samples were found to be quantifiably distinguishable, further demonstrating the utility of quantitative morphology as a signature for the processing history of nuclear material.

M

oxides will provide a realistic and innovative approach in understanding the processing history of nuclear materials. For example, U3O8 is a precursor of nuclear fuel, often transferred in large quantities between commercial facilities, and is routinely studied in nuclear forensics investigations.11 Nuclear material outside of regulatory control will likely be transferred between intermediaries and/or synthesized in an undeclared facility prior to confiscation. These processes are very likely to introduce impurities. Characterizing the effect of every possible impurity and mixtures of impurities on the surface morphology of nuclear material will require many years of experimentation. However, determining the morphological changes of a select few key impurities is an essential first step as it will bring the scientific community closer to developing predictive models as a function of the impurities. Therefore, the morphological effect of impurities on α-U3O8 from the uranyl peroxide route was investigated. The uranyl peroxide precipitation route is a commonly used commercial process known for its environmental subsidies.7 In the following study, four elements, Ca, Mg, V, and Zr, were

orphological features of U oxides have been previously shown to be a result of processing parameters including dependence on the starting material,1 intermediate material,2 oxidation rates,3 precipitation conditions,4 and thermal history.5 These were the first studies to successfully quantify morphological analysis of nuclear material and were essential to advancements in nuclear forensics and the nuclear fuels community. They demonstrated that surface morphology could serve as a signature of the processing history of nuclear material. However, these previous research efforts have focused solely on high-purity starting materials. In commercial practices, the purity of U ore concentrates is important as price penalties are assessed in terms of purity level.6 While the specifications for U ore concentrates are stringent, trace elements, or impurities, are still allowed in appreciable quantities7,8 and may be present in highly variable concentrations. This may be the result of residuals from U ore processing or feed material. Impurities may also be intentionally added during fuel pellet manufacturing to attain desired material properties, which dramatically change the nuclear fuel’s physical behavior.9 Therefore, impurities could serve as a useful fingerprint of the processing history of nuclear material. Current nuclear research efforts have focused only on qualitative morphology effects of impurities.10 A quantifiable understanding of trace impurities on the morphology of U © XXXX American Chemical Society

Received: April 27, 2019 Accepted: June 28, 2019 Published: June 28, 2019 A

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

Powder X-ray Diffraction (p-XRD). p-XRD data was collected for each of the 15 washed and 15 unwashed α-U3O8 samples. Prior to analysis, all samples were ground with 2 mL of n-pentane in a high-purity aluminum oxide mortar and pestle. Characterization was performed on a Bruker D2 Phaser with a starting position of 10° 2θ and an ending position of 70° 2θ. Each scan was recorded for 103 continuous minutes at 2 s/ step, a step size of 0.02°, and rotated at 15 rotations/min to account for any preferential orientation of the crystals. All scans were completed under the same settings. Scanning Electron Microscopy (SEM) and SEM with Energy Dispersive X-ray Spectroscopy (SEM-EDS). For SEM analysis, samples were prepared by dusting 5−10 mg of each sample onto a 12.7 mm aluminum stub mount with an adhesive 12 mm conductive carbon tab. The stub mounts were lightly tapped to remove any loose material and avoid any loss of material throughout the analysis. Images were acquired utilizing an FEI Nova NanoSEM 630 high-resolution scanning electron microscope (HRSEM). The Through the Lens (TLD) secondary electron (SE) detector was used at a voltage of 5 kV. As particle size varied with the added impurity, the HRSEM-SE image magnification ranged from 5 000× to 90 000× to ensure individual surface features could be captured for quantitative analysis. Over 650 HRSEM-SE images were collected. Energy dispersive X-ray spectroscopy (SEM-EDS) was completed on a FEI Quanta 600 FEG SEM. Images were overlaid with elemental maps corresponding to the added impurity to determine the elemental distribution of the impurities. Quantitative Morphological Analysis. Manual quantitative analysis on SEM-SE surface morphology was done using Los Alamos National Laboratory’s Morphological Analysis of MAterials (MAMA) version 2.1 software.17 The procedure for manual particle segmentation was described in detail previously.5 Over 6 000 discrete particles were manually segmented and statistically analyzed utilizing JMP Pro version 14.2.0.18 Machine learning classification was additionally employed on the SEM-SE images for quantitative analysis. The automated methods remove any human bias introduced during the manual segmentation and enable a much larger set of images to be analyzed. The binary classification was utilized to distinguish between washed and unwashed samples for each impurity. Multilabel classification was used to differentiate impurities for washed samples only, unwashed samples only, or all samples. To ensure a homogeneous data set, the collected images were filtered by magnification. Images with magnifications falling within the interquartile range for each impurity were used for analysis. Figures detailing the data set can be found in the Supporting Information. Initial image acquisitions of 1024 × 943 pixels were split into nine smaller overlapping samples of 512 × 440 pixels, effectively increasing the number of images for analysis. This facilitates the training of deep learning models as they rely on large sample sizes. Standard data augmentation techniques were additionally used to increase data set variability. This entailed flipping along the main axes, small random rotation, small balance, contrast adjustments, and random cropping of 224 × 224 pixels, a standard image size to use pretrained networks.19,20 The direct classification was performed using a deep learning method based on Convolutional Neural Network (CNN). The

implemented and examined as impurities within the U oxide. These elements were chosen to represent commonly encountered impurities in the nuclear fuel cycle. Ca and Mg are common impurities from metal working of U,12 and V and Zr are common byproducts of U mining.7,13 Ca, Mg, V, and Zr are all recognized as common impurities in the Standard Specification for Uranium Ore Concentrate by ASTM International.8 Inductively coupled plasma mass spectrometry (ICP-MS) was employed at key points during the synthesis to quantify incorporation and the final concentration of the impurity, while powder X-ray diffractometry (p-XRD) was used to observe any crystallographic changes due to the incorporation of the impurities. Scanning electron microscopy with secondary electron detectors (SEM-SE) was used in conjunction with the Morphological Analysis of MAterials (MAMA) software for quantification of the morphology, and SEM with energy dispersive X-ray spectroscopy (SEM-EDS) was utilized to locate the impurity within the α-U3O8. Further quantification was performed by machine learning. Results from this study demonstrate the power of quantitative morphology for yielding knowledge of the processing history on realistic samples. Furthermore, the quantitative morphological results provide insights into how impurities may influence desired properties in nuclear fuels.

■

EXPERIMENTAL SECTION Synthesis. Impurities were separately introduced to UO2(NO3)2·6H2O (UNH) from stock solutions of Ca(NO3)2·4H2O, Mg(NO3)2·6H2O, V2O5, and ZrO(NO3)2· 2H2O. UNH and Ca and Mg impurities were dissolved in Millipore (18.2 MΩ) water to form 0.1 M solutions. As the V and Zr compounds used had low water solubility, the stock solutions were thoroughly mixed to form a slurry and pipetted into the Millipore water and UNH to form 0.1 M solutions. Control samples were synthesized through the same procedure without the addition of impurities. The synthesis of studtite, (UO2)O2(H2O)2·2H2O, from UNH and a molar excess of H2O2 was described previously.5 Six total replicates were synthesized, and replicates 1−3 were filtered and washed with three aliquots of 50 mL of Millipore water to remove any residual nitrates.14 Replicates 4−6 were filtered and unwashed to determine the amount of impurities removed during washing. All replicates were dried for 24 h at room temperature. Samples were then placed in 5 mL platinum crucibles contained within an aluminum oxide boat for calcination. Synthesis of α-U3O8 was based on previous work done by Tamasi et al.15,16 The studtite was held at a calcination temperature of 800 °C for 20 h under 500 mL/min of purified air to form α-U3O8. When samples were not being used for data analysis, they were stored in a vacuum chamber at 24 in Hg. Inductively Coupled Plasma Mass Spectrometry (ICPMS). Incorporation of the impurity throughout synthesis was quantified utilizing ICP-MS. Samples were taken postprecipitation from the dried precipitate and post-calcination from the α-U3O8. The solids were dissolved in 5% trace metal grade HNO3 and stored in 14 mL polypropylene centrifuge tubes. Determinations of Ca, Mg, V, and Zr concentrations were done by the University of Utah’s Department of Geology and Geophysics ICP-MS Metals Lab using an Agilent 7500ce quadrupole mass spectrometer. B

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

Table 1. Impurity Concentrations Pre-precipitation,a Postprecipitation,b and Post-calcination,b Reported by Impurity wt % in U and 2σ Standard Error

technique was fine-tuned via ResNet50, a 50-layer residual neural network pretrained on the image data set.19 A small dense network was added to the convolutional component of ResNet50, including adaptive pooling and one hidden layer of 512 features. The final dense layers were updated according to the classification tasks, namely, adapting the number of final output features to match the number of classes. A two-stage transfer learning was applied; first, training only the dense network for 15 epochs with a default learning rate of 0.01, followed by training the full network for 30 additional epochs with a default learning rate of 0.001. Learning rates were modified slightly for each specific classification task. The fast.ai library, an open-source deep learning platform that sits on top of pytorch, was utilized to perform the machine learning analysis. Stochastic gradient descent (SGD) was used with restart as an optimizer.21 Additional information about the network itself, training phase, and transfer learning are described in further detail by Heffernan et al.22 A 5-fold cross-validation scheme was applied to estimate how accurately the model will perform in practice and generalize to an independent data set. During each fold, the data set was split between training data (80%) and validation data (20%). An additional component includes the use of majority voting during interference, the prediction on the test data. Each image was again split into nine smaller samples, the prediction was performed at each sample level, and majority voting was used to acquire a single prediction at the acquisition level. Validation results from the cross-validation scheme including accuracy, F1 score, precision, and recall were averaged to estimate the model’s predictive performance. Precision, or positive predictive value, is the number of true positives over the number of true positives plus false positives. Recall, or sensitivity, is the number of true positives over the number of true positives plus false negatives. The average F1 score is the harmonic mean of precision and recall weighted by the number of true positives per label for multilabel classification.

a

Pre-precipitation values were calculated and normalized based on the impurity compound and resulting stock solution density, impurity and U molarity in solution, and the total weight of the solution. bPostprecipitation and post-calcination values were measured by ICP-MS and were normalized to the concentration of U present in each sample.

ICP-MS was also utilized to quantify Ca, Mg, V, and Zr concentrations within the control samples post-precipitation and post-calcination. These results are summarized in the Supporting Information in Table S-2 and reported in wt %. Nearly all unwashed samples had concentrations of Ca, Mg, V, and Zr below the detection limit of the ICP-MS. However, the washed samples had measurable concentrations of Ca, Mg, and Zr. The Ca concentrations in the washed control samples are within 2σ error of the washed samples containing the Ca impurity. Mg and Zr values in the controls are much lower than the samples with added impurities and are not believed to have a discernible effect. This is further discussed in the p-XRD and SEM analysis sections below. Powder X-ray Diffraction (p-XRD). Crystallographic changes were investigated by p-XRD. All samples were analyzed using PANalytical X’Pert Highscore Plus v2.2d software.23 The spectra illustrated in Figure 1 are representa-

■

RESULTS AND DISCUSSION Inductively Coupled Plasma Mass Spectrometry (ICPMS). Impurity concentrations added during synthesis were calculated and normalized based on the stock solution density, impurity and U molarity in solution, and the total weight of the solution. Impurity concentrations in the dried precipitate and α-U3O8 were measured by ICP-MS. These results were normalized to the concentration of U in each sample (wt %). The calculated and measured impurity concentrations are reported in Table 1 to 2σ standard error. Post-precipitation and post-calcination Ca and Mg concentrations largely differ from the added pre-precipitation concentrations. This indicates that Ca and Mg will not precipitate with the U upon addition of H2O2. Furthermore, what little Ca and Mg that did precipitate with the U was readily removed when the sample was washed with water. In contrast, V and Zr readily precipitate with the U following addition of H2O2; >60% of the V and >90% of the Zr were precipitated with the U. Washing the samples removed some of the V and Zr impurities, but they were still present at concentrations of 1.1 ± 0.4 and 0.9 ± 0.1 wt %, respectively. Almost no changes were observed in the wt % of impurities following calcination of the precipitate to U3O8. This was expected as none of the impurities were anticipated to form volatile species at the calcination temperature.

Figure 1. p-XRD data representative of washed samples compared to the α-U3O8 reference pattern.24 The sample with the addition of V shows additional peaks throughout the spectrum, indicative of at least one additional complex forming. The sample with the addition of Zr shows broad peaks due to the small crystallite size.28 C

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry tive of washed replicates 1−3 and compared against the reference peaks of α-U3O8 (Powder Diffraction File 01-073629324). Spectra for the remaining replicates are presented in the Supporting Information. All samples match the α-U3O8 reference pattern with the most significant peaks occurring at 21.4° and 26.0° (2θ). Samples with the addition of V exhibit additional peaks most dominantly at 19.3° and 24.8° (2θ) that are not characteristic of the α-U3O8 crystallinity. This is indicative of the formation of another complex, and these additional peaks match with an analogue of uranyl pyrovanadate, (UO2)2V2O7.25,26 This is further supported by (UO2)O2(H2O)2·2H2O as the starting calcination material and previous work by Yáñez et al., in which similar experimental methods were used to obtain and classify uranyl pyrovanadate.27 Samples with the addition of Zr show peaks corresponding to nanocrystalline materials as smaller particle sizes will result in increasingly broad peaks.28 Similarly, the p-XRD results for the washed samples are comparable to the unwashed samples, in which the only qualifiable p-XRD difference in the unwashed samples are broader peaks due to smaller particle sizes. Scanning Electron Microscopy (SEM) and SEM with Energy Dispersive X-ray Spectroscopy (SEM-EDS). Figure 2 illustrates the differences in the α-U3O8 morphology observed between samples. Each sample was qualitatively analyzed based on a lexicon developed by Tamasi et al. for describing nuclear material imagery.29 All control samples match α-U3O8 morphology described previously,5,30 in which the samples are clumped/massive agglomerates with rounded/ subrounded subparticles with semirounded grains. They are arranged in irregular clumps containing somewhat rough surface features. Samples with the addition of Ca and Mg are comparable to these descriptors, with the exception that all features are qualitatively smaller in the unwashed samples and contain fewer agglomerates. Samples with the addition of Zr have the smallest features but can still be characterized by the same descriptors. In contrast, samples containing V have the largest features. These samples are characterized as having clumped/massive conglomerates. Two different subparticle morphologies were observed, with one having rounded/subrounded particles and the second with elongated, visibly darker in color particles. They are arranged in irregular clumps comprised of somewhat smooth, glasslike surface features. Samples containing V were further examined by SEM-EDS to locate the impurity within the different surface features. SEM images were overlaid with an elemental map highlighting the position of V within the sample in yellow (Figure 3). These results indicate V is more concentrated in the elongated surface features. The different subparticle morphology alongside the dense regions of V suggests the presence of another complex. This compliments the p-XRD results, which showed (UO2)2V2O7 present in addition to α-U3O8. Quantitative Morphological Analysis. Qualitatively, V and Zr can be readily distinguished from the controls. Nonetheless, the washed and unwashed samples have no discernible difference in the qualitative morphology. This is supportive of the ICP-MS results which showed very little change in the wt % of V or Zr incorporated post-precipitation and post-calcination. While the samples with Ca and Mg impurities show no distinguishable differences from the controls, washed to unwashed samples are discernible. This also supports the ICP-MS results, which showed quantifiable

Figure 2. SEM image comparison of washed (left column) and unwashed (right column) α-U3O8 samples. Unwashed samples with the addition of Ca or Mg are qualitatively smaller than the washed samples.

differences in the Ca and Mg samples with or without washing. To further elucidate morphological changes due to the impurities, quantitative analysis was pursued to establish statistically significant differences between all impurities and washed and unwashed replicates. Two methods were used for quantitative analysis, MAMA software and machine learning classification. With the MAMA software, over 6 000 discrete particles were manually segmented and classified according to attributes such as pixel area, circularity, and ellipse aspect ratio. This data was D

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

Figure 4. Comparison of mean pixel area at a 95% confidence interval between replicates 1−6 (1−3 washed and 4−6 unwashed). Samples with the addition of V largely offset the data curve due to large particle size (mean pixel area of 3.5 and 2.9 μm2 for washed and unwashed samples, respectively) and are not pictured. Results indicate washed samples containing the addition of Ca cannot be distinguished from the washed controls. All other samples are discernible.

have similar morphologies, albeit the unwashed sample particles are slightly larger. In contrast, washing samples that contained Mg resulted in a lower impurity concentration in the final product. Thus, the morphology of samples containing Mg varies between the washed and unwashed replicates. Samples with the addition of V were not included in Figure 4 as they offset the data due to large particle sizes. The mean pixel area for the washed and unwashed samples were 3.5 and 2.9 μm2, respectively, and cannot be distinguished from each other within a 95% confidence interval. This also corresponds with the ICP-MS data above in which the V concentrations had little change following washing the samples. While manual segmentation was more effective at classifying samples than qualitative analysis, two sets of samples could not be differentiated (i.e., washed control from washed Ca and washed from unwashed V). Therefore, the second method of quantitative analysis, machine learning classification, was employed. The multilabel classification model achieved an average accuracy of 83.84% for all samples and is illustrated in Figure 5 by the confusion matrix. False positive predictions occurred in each sample. Examining washed samples only resulted in an accuracy of 85.22% and samples containing V and Zr were predicted without any false positives. Analyzing unwashed samples only returned an accuracy of 99.07%, and the control and samples containing V were predicted without any false positives. This data is illustrated in additional confusion matrixes provided in the Supporting Information. The binary classification distinguished between washed and unwashed control, Ca, Mg, V, and Zr samples at 100%, 98.95%, 100%, 91.06%, and 89% accuracy, respectively. The multilabel classification gave high accuracy predictions for V and Zr samples, while binary classification gave high accuracy predictions for control, Ca, and Mg samples, utilizing both methods of classification provided high accuracy predictions for each sample.

Figure 3. SEM-EDS images of α-U3O8 before (top) and following (bottom) the addition of a V overlay map. Regions highlighted in yellow contain V, which was found to be more concentrated in the elongated surface features. This is indicative of an additional complex forming. Images are representative of washed samples with the addition of V.

processed using JMP, in which principle component analysis determined pixel area was the most distinguishing particle attribute. Figure 4 depicts a comparison of the mean pixel area at a 95% confidence interval for the washed and unwashed samples containing Ca, Mg, and Zr. Washed samples with Mg and Zr impurities are statistically discernible from the washed controls. Likewise, all unwashed samples with added impurities are statistically different than the unwashed controls. Furthermore, all washed samples are statistically discernible from their corresponding unwashed sample. However, washed samples with Ca impurities are not statistically distinct from the washed control samples. This data correlates to the observed ICP-MS data above. Specifically, the washed control samples contained small concentrations of Ca which were statistically the same as those with added Ca impurities. This resulted in the washed controls and washed Ca samples having similar particle morphology. In this case, unwashed control samples represent the “true control” as Ca concentrations were below the limit of detection of ICP-MS in these samples. Likewise, washing samples containing Zr had very little change in the total concentration of Zr in the final product. Hence, the Zr samples E

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

■

matrix for multilabel classification model of washed and unwashed samples (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: 1-801-581-7768. ORCID

Luther W. McDonald, IV: 0000-0001-6735-5410 Author Contributions

The manuscript was written through the contributions of all authors. All authors have given approval to the final version of the manuscript. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This synthesis of U-oxides and their subsequent characterization by p-XRD, SEM, and ICP-MS was supported by the U.S. Department of Homeland Security, Domestic Nuclear Detection Office, under Grant Award No. 2015-DN-077ARI092. The same DHS grant, along with the Defense Threat Reduction Agency, under Grant Award No. HDTRA1-16-10026 supported the quantitative analysis via MAMA. The machine learning analysis was supported by the U.S. Department of Homeland Security, Domestic Nuclear Detection Office, under Grant Award No. 2016-DN-077ARI102. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U.S. Department of Homeland Security or Defense Threat Reduction Agency. This work made use of University of Utah Shared facilities of the Surface Analysis and Nanoscale Imaging Group sponsored by the College of Engineering, Health Sciences Center, Office of the Vice President for Research, and the Utah Science Technology and Research (USTAR) Initiative of the State of Utah. This work made use of the Physics Department and Materials Characterization Lab at the University of Utah. The authors would also like to thank University of Utah’s Department of Geology and Geophysics Metals Lab for the ICP-MS analysis.

Figure 5. Confusion matrix for multilabel classification model of all samples. Results indicated an average accuracy of 83.84%.

■

CONCLUSION This study examined the morphological effect of Ca, Mg, V, and Zr impurities on α-U3O8. Quantitative analysis by ICP-MS showed Ca and Mg do not readily precipitate with U upon addition of H2O2 and are further removed from solution when washed with water. V and Zr will precipitate and are not as affected by washing. The washed control samples and samples containing the Ca impurity were found to have statically similar concentrations of Ca. Samples with the addition of V were found through p-XRD and SEM-EDS to have formed another complex in addition to α-U3O8, (UO2)2V2O7. This was all further supported by HRSEM imagery and quantitative MAMA software analysis. While MAMA analysis alone could not statistically differentiate between the morphology of all samples, machine learning classification could distinguish between all samples with high accuracy. This study quantitatively validates the proof of concept ability of impurities to alter the surface morphology of α-U3O8. Realistically, impure samples will likely contain mixtures of several impurities, and this study is an essential first step to the further investigation of impurities in uranium oxides as it demonstrates the power of quantitative morphology on the processing history of nuclear material.

■

■

REFERENCES

(1) Schwerdt, I. J.; Hawkins, C. G.; Taylor, B.; Brenkmann, A.; Martinson, S.; McDonald, L. W., IV Radiochim. Acta 2019, 107 (3), 193−205. (2) Abbott, E. C.; Brenkmann, A.; Galbraith, C.; Ong, J.; Schwerdt, I. J.; Albrecht, B. D.; Tasdizen, T.; McDonald, L. W., IV Radiochim. Acta 2019, DOI: 10.1515/ract-2018-3065. (3) Olsen, A. M.; Schwerdt, I. J.; Richards, B.; McDonald, L. W., IV J. Nucl. Mater. 2018, 508, 574−582. (4) Schwerdt, I. J.; Olsen, A.; Lusk, R.; Heffernan, S.; Klosterman, M.; Collins, B.; Martinson, S.; Kirkham, T.; McDonald, L. W., IV Talanta 2018, 176, 284−292. (5) Olsen, A. M.; Richards, B.; Schwerdt, I.; Heffernan, S.; Lusk, R.; Smith, B.; Jurrus, E.; Ruggiero, C.; McDonald, L. W. Anal. Chem. 2017, 89 (5), 3177−3183. (6) Caropreso, F. E.; Kreuz, D. F. Production of Uranium Peroxide. United States Patent 4,024,215, May 17, 1977. (7) International Atomic Energy Agency. Uranium Extraction Technology, Technical Report Series No. 359, International Atomic Energy Agency: Vienna, Austria, 1993.

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.9b02013. Raw and normalized ICP-MS data post-precipitation and post-calcination for control samples; multilabel and binary classification results; SEM images representative of all replicates; SEM image comparison of control samples and samples with the addition of Ca, Mg, V, and Zr; p-XRD data of all replicates; comparison of mean particle pixel area at 95% confidence between all samples; SEM images before and after MAMA segmentation; machine learning classification image and magnification data set overview; and confusion F

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry (8) ASTM Committee C26. Standard Specification for Uranium Ore Concentrate. In Annual Book of ASTM Standards; ASTM International: Conshohocken, PA, 2013; pp 233−235. (9) Mayer, K.; Wallenius, M.; Varga, Z. Chem. Rev. 2013, 113 (2), 884−900. (10) Manna, S.; Roy, S. B.; Joshi, J. B. J. Nucl. Mater. 2012, 424 (1− 3), 94−100. (11) Fongaro, L.; Lin Ho, D. M.; Kvaal, K.; Mayer, K.; Rondinella, V. V. Talanta 2016, 152, 463−474. (12) Durazzo, M.; Saliba-Silva, A.M.; Martins, I.C.; de Carvalho, E.F. U.; Riella, H.G. Ann. Nucl. Energy 2017, 110, 874−885. (13) By-Product Recovery with Uranium Mining, WISE Uranium Project, Mining & Milling. (14) Cordfunke, E. H. P.; Van Der Giessen, A. A. J. Inorg. Nucl. Chem. 1963, 25 (5), 553−555. (15) Tamasi, A. L.; Boland, K. S.; Czerwinski, K.; Ellis, J. K.; Kozimor, S. A.; Martin, R. L.; Pugmire, A. L.; Reilly, D.; Scott, B. L.; Sutton, A. D.; Wagner, G. L.; Walensky, J. R.; Wilkerson, M. P. Anal. Chem. 2015, 87 (8), 4210−4217. (16) Tamasi, A. L.; Cash, L. J.; Tyler Mullen, W.; Ross, A. R.; Ruggiero, C. E.; Scott, B. L.; Wagner, G. L.; Walensky, J. R.; Zerkle, S. A.; Wilkerson, M. P. J. Radioanal. Nucl. Chem. 2016, 309 (2), 827− 832. (17) Ruggiero, C. E.; Bloch, J. J. Morphological Analysis for Material Attribution (MAMA), version 2.1; Los Alamos National Laboratory 2016. (18) SAS Institute Inc. JMP Pro, version 14.2.0; 2018. (19) He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016; pp 770−778. (20) Simonyan, K.; Zisserman, A. Very Deep Convolutional Networks for Large-Scale Image Recognition. In International Conference on Learning Representations (ICLR 2015), San Diego, CA, May 7−9, 2015; pp 1−14. (21) Loshchilov, I.; Hutter, F. SGDR: Stochastic Gradient Descent with Warm Restarts. In International Conference on Learning Representations (ICLR 2017), Toulon, France, April 24−26, 2017; pp 1−16. (22) Heffernan, S. T.; Ly, N.-C.; Mower, B. J.; Vachet, C.; Schwerdt, I. J.; Tasdizen, T.; McDonald, L. W., IV Radiochim. Acta 2019, DOI: 10.1515/ract-2019-3140. (23) Degen, T.; Sadki, M.; Bron, E.; König, U.; Nénert, G. Powder Diffr. 2014, 29 (S2), S13−S18. (24) ICDD. International Centre for Diffraction Data (ICDD), PDF-2 Database Release, 2008. (25) Saadi, M.; Dion, C.; Abraham, F. J. Solid State Chem. 2000, 150 (1), 72−80. (26) Chippindale, A. M.; Dickens, P. G.; Flynn, G. J.; Stuttard, G. P. J. Mater. Chem. 1995, 5 (1), 141−146. (27) Yanez, J. L. R.; Cortes, M. R.; Moye, E. T.; Lardizabal, D.; Riveros, H.; Cabrera, M. E. M. Revista Mexicana de Fisica 2012, 58 (3), 253−257. (28) Scardi, P.; Leoni, M.; Beyerlein, K. R. Zeitschrift fur Kristallographie 2011, 226 (12), 924−933. (29) Tamasi, A. L.; Cash, L. J.; Eley, C.; Porter, R. B.; Pugmire, D. L.; Ross, A. R.; Ruggiero, C. E.; Tandon, L.; Wagner, G. L.; Walensky, J. R.; Wall, A. D.; Wilkerson, M. P. J. Radioanal. Nucl. Chem. 2016, 307 (3), 1611−1619. (30) Tamasi, A. L.; Cash, L. J.; Mullen, W. T.; Pugmire, A. L.; Ross, A. R.; Ruggiero, C. E.; Scott, B. L.; Wagner, G. L.; Walensky, J. R.; Wilkerson, M. P. J. Radioanal. Nucl. Chem. 2017, 311 (1), 35−42.

G

DOI: 10.1021/acs.analchem.9b02013 Anal. Chem. XXXX, XXX, XXX−XXX