1.Introduction

The high temperature electrochemical treatment of used nuclear fuel (UNF), commonly termed pyroprocessing, is currently being developed by several countries— including non-nuclear weapons states (NNWS) such as Japan and the Republic of Korea. If a commercial pyroprocessing facility were to be built and operated in an NNWS, two parties would have a need to monitor inventories of special nuclear materials (SNM) within various material balance areas (MBA) in such a facility: (1) the host country’s operators and engineers and the (2) International Atomic Energy Agency (IAEA). Of particular interest in a pyroprocessing facility is a unit operation called an electrorefiner (ER) where SNM is partitioned and accumulates in amounts greater than the IAEA’s definition of a significant quantity. The ER consists of an anode basket containing used nuclear fuel, a molten eutectic LiCl-KCl salt bath which serves as the electrolyte, and a cathode where pure uranium or a mixture of uranium and transuranics (TRU) is collected depending on the type of cathode employed.

Monitoring the concentration in the ER salt is the key to maintaining process control, optimizing process parameters, and meeting IAEA safeguards requirements. However, the current method for monitoring the concentration in the ER is to sample the salt and wait weeks for analytical chemistry to be performed and reported. This turnaround time would be insufficient to meet the timeliness goals of the IAEA and the feedback provided to the operators would be too slow to allow them to make appropriate adjustments to process parameters in an industrial-scale facility. Real time (RT) or near-real time (NRT) concentration measurements could assist both the host country and the International Atomic Energy Agency (IAEA) by providing a continuity of knowledge of salt composition, which could be relatively uncertain otherwise. The host country’s operators could use RT/NRT measurements to determine optimal process parameters and the frequency of salt cleanup or TRU drawdown operations. The IAEA could use RT/ NRT concentration measurements to help close mass balances and detect the abnormal operating conditions. A RT/ NRT concentration measurement sensor would help close a large information gap and would be beneficial to development of pyroprocessing technology and safeguards.

Because the ER is already an electrochemical environment, electroanalytical techniques utilizing electrode probes (cyclic voltammetry, chronoamperometry, etc.) can be readily applied to make in situ, RT/NRT quantitative measurements of ER salt composition. Cyclic voltammetry (CV) is a popular method because of its reliability over a broad range of concentrations in molten LiCl-KCl eutectic [1]. Additionally, normal pulse voltammetry (NPV) and square wave voltammetry (SWV) have been applied to the measurement of actinide ion concentrations in molten LiCl-KCl eutectic, but were found to be only suitable at low concentrations [2]. In addition to these two studies, the electrochemical behavior of several actinides in LiCl-KCl eutectic have been well studied. Zhang provided an extensive review of these studies [3]. The studies reviewed by Zhang support the development of electrochemical sensors for NRT concentration measurements by providing important parameters and properties of actinide elements. Unfortunately, these studies have mostly focused on single actinides in eutectic LiCl- KCl with the exception of the work by Iizuka et. al. [2]. In an actual ER, several of the actinides are present at varying concentrations. It is anticipated that the actinides can interfere with each other’s electrochemical responses.

Typically, the concentration in an electrolyte such as a molten salt is estimated by relating an electrochemical signal such as steady-state current, current peak height, current peak area, or cumulative charge to concentration. However, if multiple actinides are present, their associated signals could overlap making it difficult to establish baselines for the peaks or to confidently attribute the measured signal to a particular ion. For example, in CV, current peaks are formed by scanning the potential of a sensing electrode. As long as the peak heights are linearly related to the concentration of the corresponding ion, they can be reliably used to estimate concentration. However, if another ion is present and forms a peak in close proximity to the other ion, the peaks will overlap. The peaks will need to be deconvoluted in order to extract reliable concentration estimations for each individual ion. This could involve pausing the scan just after the peak potential or semi-differentiating the CV curve, but these methods require a certain amount of separation between the peak potentials and discard a majority of the data [4, 5]. Alternatively, multivariate analytical techniques, which utilize the complete CV curve rather than a couple data points at the peaks, could provide more reliable and accurate estimations, potentially even when peaks almost completely overlap [6, 7]. Both of these methods, peak separation and multivariate analysis, were utilized and evaluated in this work for the GdCl₃-LaCl₃-LiCl-KCl system.

The GdCl₃-LaCl₃-LiCl-KCl quaternary system was selected based on the commercial availability of the individual salts and spacing of their respective apparent reduction potentials from literature data. In Zhang’s review, the spacing between Gd³⁺/Gd and La³⁺/La reduction potentials was reported to be 0.158 V at 773 K. The rationale is that 0.158 V is smaller than the spacing of U³⁺/U and Pu³⁺/Pu apparent standard potentials, 0.234 V. Therefore, if accurate estimations can be achieved for the quaternary system of GdCl₃ and LaCl₃ in eutectic LiCl-KCl, then it is highly likely that accurate estimation can be made for quaternary system of UCl₃ and PuCl₃ in eutectic LiCl-KCl using the same methods investigated in this work.

2.Experimental Procedures

All electrochemical tests were performed in a Kerr Electro- melt furnace (Model No. 35224) which heated the salt to 500°C +/- 2°C. The interior of the furnace consists of a graphite crucible and a heating element. A 6-cm diameter alumina (99.6%) crucible (AdValue Technology, Part No. AL-2250) was inserted into the furnace to act as a liner in case of cracking in the primary crucible containing the molten salt. The primary crucible was an alumina (99.6%) 4-cm diameter crucible (AdValue Technology, Part No. AL-2100). A 5-cm diameter hole was drilled into the lid of the furnace. Then an alumina plug was manufactured to hold the electrodes and fill the hole in the lid. A picture and drawing of the experimental setup are displayed in Figure 1. The reference electrode (RE) consisted of a thin-walled borosilicate NMR tube (Sigma-Aldrich, Part No. Z274771-1PAK), a 1-mm diameter silver wire (99.9%, Alfa-Aesar) and salt mixture consisting of eutectic LiCl-KCl (99.99%, Sigma-Aldrich) and 1-wt% AgCl (99.9%, Strem Chemicals). The working electrode (WE) was 1-mm diameter Mo wire (99.94%, Alfa- Aesar) partially sheathed in alumina to above the level of the salt and immersed 0.9-1.8 cm. The specific immersion depth for each test is listed in Table 1. The WE was rinsed in 3-N HCl, polished by 2000-grit silicon carbide paper and anodically cleaned before conducting electrochemical tests. The counter electrode (CE) was a stainless steel rod and basket containing a 0.25” x 1” Gd metal rod (99.9%, Alfa-Aesar). GdCl₃ (99.99% anhydrous, Alfa-Aesar) and LaCl₃ (99.99% anhydrous, Alfa-Aesar) were added to eutectic LiCl-KCl (99.99% anhydrous, Sigma-Aldrich) salt to form mixtures of various compositions. The compositions of GdCl₃ and LaCl₃ in eutectic LiCl-KCl as determined by ICP-OES are listed in Table 2 for each mixture. Additionally, two ternary mixtures were made—one containing GdCl₃ at 1.71wt% and the other containing LaCl₃ at 0.96wt%. Each mixture was analyzed electrochemically using cyclic voltammetry (CV), chronoamperometry (CA) and chronopotentionmetry (CP).

3.Results and Discussion

3.1.Experimental Data

A representative CV of the quaternary mixtures is plotted in Figure 2. On the forward (reducing) scan, the current is flat until the reduction potential of Gd³⁺ ions which forms the first peak at about -1.94 V. In previous work [8, 9] with GdCl₃ in eutectic LiCl-KCl, Gd³⁺ ions were identified to reduce to Gd metal in a single step. Before the Gd³⁺ reduction peak decays, La³⁺ ions begin to deposit and the peak corresponding to La³⁺ reduction closely follows the Gd³⁺ reduction peak. Again, a single step reduction mechanism has been identified for La³⁺ to La metal [10, 11]. Lastly, a sharp decrease in the current near -2.4 V is indicative of Li⁺ ions reducing establishing the lower limit of the electrochemical window for eutectic LiCl-KCl. Thus, the scan is reversed and Li metal is oxidized forming the peak at -2.3 V. The expected oxidation peak for the oxidation of La metal is absent and a prominent peak only forms where Gd metal oxidizes. The phenomenon is better illustrated in the plot in Figure 3 which overlays CVs from the two ternary mixtures on the CV from test no. 10. It is clear in Figure 3 that the La oxidation peak is missing or unnoticeable on the black curve which corresponds to test no. 10, and the remaining oxidation is in the position of the Gd oxidation peak.

Two possible causes are speculated for the absences of the La oxidation peak. First, it is feasible that Gd and La form a metal alloy when both are present on the WE. Indeed, phase diagrams show that at 500°C alloys readily form between Gd and La metals [12]. Attempts were made to outpace the supposed alloying reaction by increasing the CV scan rate incrementally up to 2 V/s. No indication of a second oxidation step was observable from the CV at any of the scan rates. Alternatively, La metal could begin to oxidize, but could become covered by Gd metal since Gd deposition is still favored at La oxidation potentials. This could be the cause of the slight bump around -2.1 V in Figure 3. This bump is more prominent when LaCl₃ exceeds GdCl₃ in LiCl-KCl eutectic which is the case for test no. 3. A CV for test no. 3 is displayed in Figure 4. A clear bump is observed between -2.1 and -2.0 V. Semi-differentiation of the CV reveals two distinct peaks indicating the possible convolution of two oxidation peak in the CV. In either case, further investigation into the cause of the absence of La oxidation peaks is warranted. Thus, only the cathodic peaks were investigated in this work.

The absence of the La metal oxidation peak was further investigated using CP. A current of -80 mA was applied to the cell for 4 seconds, then the cell was held at open circuit and the potential was measured. The resulting potential profiles are plotted in Figure 5 with a magnification of the first 0.75 seconds. There is a slight bend in the plot when a reducing current of -80 mA is charged indicating the onset of the deposition of La from La³⁺ ions. However, when the open-circuit potential is measured afterward, there is no step or bend in the potential profile to indicate oxidation of La metal. Ideally, three potential plateaus should appear for Li, La and Gd metal oxidation while the cell is held at open-circuit. In Figure 5, this is not the case. Only two plateaus appear corresponding to the open-circuit potentials measured for Li/Li+ and Gd/Gd³⁺ redox couple.

As shown in Figures 2-5, the electrochemical signals of Gd³⁺/Gd and La³⁺/La redox couples overlap significantly. Thus, in order to make concentration estimations, a method needs to be developed to separate the signals or analyze them together. The signals recorded for CA and CP had very minimal or no separation between the two redox couples making them poor candidates for signal separation. Some separation between reduction peaks could be achieved in CVs using slower scan rates. However, additional treatment was necessary to obtain the individual peaks.

3.2.Peak Separation

The peak separation technique involved both the semidifferentiation and curve fitting as detailed by Palys et. al [5]. First, the reduction peaks of the CVs of the quaternary salt mixtures in Table 1 were semi-differentiated. Semi-differentiation creates sharper and more symmetrical peaks. A representative plot of semi-differentiated peaks is shown in Figure 6. While the peaks in Figure 6 look sharper and are more symmetrical than those in Figure 3, there is still some overlap and asymmetry to them, unlike the peaks fitted by Palys et. al., who fitted symmetrical peaks with the inverse of the hyperbolic cosine squared. In this case, due to asymmetry, a Bifurcated Gaussian (Bigaussian) distribution was fitted to the semi-differentiated peaks using Fityk 0.9.8 [13], as shown in Figure 6. The fitted Bigaussian peaks are then semi-integrated back to obtain individual CV reduction peaks, as shown in Figure 7. In most cases, a good fit can be made similar to those in Figures 6 and 7. However, occasionally the fit was poor. There were two types of poor fits. First, statistically the curves did not match closely. Second, the curves fit well statistically, but were not true to the physical nature of the system. For example, the fit in Figure 8 matches the measured data well, but does not preserve the physical features of the system. Both La and Gd deposition commences at the same potential. This results in a far too large of a peak for La with a shallow rise, as opposed to the typical sharp rise. In some cases, like test no. 4, the fit could be altered to preserve the physical features while still maintaining a good statistical fit. However, for tests nos. 5 & 6, statistically good fits that preserved the physical features of the reduction peaks could not be achieved. These sets of data were excluded from further analysis.

Once the peaks had been separated, their peak heights for Gd and La from test nos. 1-9 (excluding 5 and 6) were correlated to concentration to form calibration curves. These curves are displayed for Gd and La in Figure 9 with GdCl₃ and LaCl₃ concentration displayed on the left and right vertical axes, respectively. The regressed linear parameters, coefficient of determination (R²) and standard error of the y estimates (SE_y) for both the Gd and La peak height correlations are given in Table 3. The parameters are only valid within the concentration range of 1.49-4.69wt% GdCl₃ and 0.89-2.45wt% LaCl₃. Furthermore, these correlations were developed based on CVs with a scan rate of 100 mV/s. The calibration curve for Gd is more tightly correlated than the calibration curve for La. This is most likely due to the fact that La³⁺ reduction occurs after Gd³⁺ reduction, making it more difficult to decipher the true La peak height. Using the regressed curves in Figure 9, the composition of Gd and La were estimated for tests nos. 10 and 11 and compared to the composition measured by ICPOES in Table 4.

The values in Table 4 match on average within 6.7%, indicating that voltammetry is likely to be very useful for process control by the operator. In Table 3, the relative errors between the estimated and ICP-OES measured concentrations are quite low except for GdCl₃ in test no. 10. But the error is still too high to be useful for safeguards measurements by the IAEA. This method may prove more effective for the UCl₃-PuCl₃-LiCl-KCl quaternary system because the overlap will be significantly less for Pu and U peaks. However, there may be other species in the ER that will interfere with U and Pu peaks such as NpCl₃. Thus, resolving the issues associated with this closely spaced GdCl₃-LaCl₃- LiCl-KCl system is still an important objective.

Error may have been introduced into the measurement by the peak separation method. For example, too much or too little current may have been attributed to either species in some cases. In Figure 9, the lowest concentrations (~1.5wt% for GdCl₃, ~1wt% for LaCl₃) appear to have 3 separate peak height values. This may be an artifact of peak separation or unintentional experimental variations unrelated to concentration. Additionally, the peaks in Figure 6 appear to have a shifting baseline. Efforts were made to remove the baseline by assuming a linear baseline, subtracting it and fitting peaks. However, the correlation between peak height and concentration decreased when the baseline correction was made (i.e. R² of 0.678 and 0.688 for GdCl₃ and LaCl₃, respectively). This is possibly due to the fact that two peaks are present resulting in two different baselines. It is highly unlikely that both baselines have the same slope. Thus, the baseline correction may have removed from or added to the actual peak. For example, if the Gd peak has a shallow baseline and La peak has steep baseline, then some of actual signal is being removed. Thus, further refinement of the peak separation technique is merited and may result in better concentration estimations. Alternatively, the process of peak separation could be bypassed by using the multivariate technique of principle component regression (PCR) as discussed in the following section.

3.3.Principle Component Regression

PCR is a combination of principle component analysis (PCA) and least-squares regression. PCR uses a training set of data to analyze the variance of data over a range of compositions and then regress the main contributors, principle components (PCs), to variance with concentration. Using this regression the PCs of an unknown sample can be used to estimate the composition. An advantage of PCR is that all of the data collected in the region of interest is used. The application of PCR to electrochemical data is discussed and demonstrated, in detail, elsewhere [6, 7, 14]. This work used test nos. 1-9 (excluding 5 and 6) as the training set. The compositions estimated by PCR for tests nos. 10 & 11 are displayed in Table 5.

The compositions estimated by PCR are overall less accurate than those estimated using the separated peak height, the lone exception being GdCl₃ estimation in test no. 10. There may be a couple of causes for PCR to underperform. First, the pretreatment of the data may need to be adjusted. Before PCR can be applied to experimental electrochemical data, it needs to be prepared. This can involve weighting, mean-centering, scaling, etc. In this work, only an interpolation was performed to obtain uniform spacing of the potentials for each data set. Some form of centering or scaling may help improve the accuracy of PCR. However, caution needs to be used to keep the data physically accurate so that the data transformations do not remove or standardize CV features, such as peak height and potential, which are concentration dependent. In the future, a standard pretreatment procedure that preserves the physical features of the CVs will be developed. Second, each test was its own experiment performed one day at a time. These experiments were performed over a couple months. During this time RE tubes broke, different WE wires were used and other experimental conditions not related to concentration may have varied. These variations may not be as pronounced in separated peak heights since only one data point is used. The most important condition to replicate in each experiment was the WE surface. The WE was prepared as described in the procedures. Despite the uniform treatment of the WE, it still proved difficult to obtain completely identical CVs in separate tests using the same materials and conditions from a previous test. The electrochemical cell is currently being redesigned to ensure more repeatable electrode configurations and to allow the addition of analyte salts while the salt is still molten and electrodes are in place. This will allow CVs to be measured at multiple compositions in one experiment under the same conditions which will help reduce variations unrelated to concentration.

3.4.Comparison of Peak Height and PCR

Peak height may utilize less data than PCR, but in this case produced more accurate estimations of the concentration. This may be due to the fact that the peak height is less sensitive than PCR to changes in other variables. However, in order to use peak heights to estimate concentrations, significant data processing needed to be performed. Automating this process may prove difficult since the fitting statistically may appear to be good, but may be a poor representation of the physical process. PCR does not require the separation of signals and potentially could be applied to mixture containing more than two analytes. It would be fairly straightforward to automate PCR, even with the additional pretreatment of the data. However, fluctuations in the process temperature, WE surface condition and other variables would need to be tightly controlled or their effect characterized and taken into account.

4.Conclusions

Several electrochemical methods (CV, CA and CP) were applied to quaternary mixtures containing GdCl₃ and LaCl₃ in eutectic LiCl-KCl. The electrochemical signals of each species were inseparable and, in some cases, indecipherable using CA and CP, but could be separated for CV. Two methods were used to estimate the composition of GdCl₃ and LaCl₃ in the molten salt mixtures—peak separation and PCR. Peak separation resulted in more accurate estimations overall. However, the pretreatment of the data for PCR can be improved. Also, there may be significant variations in the data unrelated to concentration which may have introduced inaccuracy to both methods. Further work is underway to design an electrochemical cell that will result in more repeatable measurements. Additionally, PCR and peak separation methods need to be further refined to better preserve the physical data and improve estimations.