Article pubs.acs.org/ac

Development of Online Spectroscopic pH Monitoring for Nuclear Fuel Reprocessing Plants: Weak Acid Schemes Amanda J. Casella,* Laura R. H. Ahlers, Emily L. Campbell, Tatiana G. Levitskaia, James M. Peterson, Frances N. Smith, and Samuel A. Bryan* Energy and Environment Directorate, Pacific Northwest National Laboratory, Richland, Washington 99352, United States S Supporting Information *

ABSTRACT: In nuclear fuel reprocessing, separating trivalent minor actinides and lanthanide fission products is extremely challenging and often necessitates tight pH control in TALSPEAK (Trivalent ActinideLanthanide Separation by Phosphorus reagent Extraction from Aqueous Komplexes) separations. In TALSPEAK and similar advanced processes, aqueous pH is one of the most important factors governing the partitioning of lanthanides and actinides between an aqueous phase containing a polyaminopolycarboxylate complexing agent and a weak carboxylic acid buffer and an organic phase containing an acidic organophosphorus extractant. Real-time pH monitoring would significantly increase confidence in the separation performance. Our research is focused on developing a general method for online determination of the pH of aqueous solutions through chemometric analysis of Raman spectra. Spectroscopic process-monitoring capabilities, incorporated in a counter-current centrifugal contactor bank, provide a pathway for online, real-time measurement of solution pH. The spectroscopic techniques are process-friendly and can be easily configured for online applications, whereas classic potentiometric pH measurements require frequent calibration/maintenance and have poor long-term stability in aggressive chemical and radiation environments. Raman spectroscopy discriminates between the protonated and deprotonated forms of the carboxylic acid buffer, and the chemometric processing of the Raman spectral data with PLS (partial least-squares) regression provides a means to quantify their respective abundances and therefore determine the solution pH. Interpretive quantitative models have been developed and validated under a range of chemical composition and pH conditions using a lactic acid/lactate buffer system. The developed model was applied to new spectra obtained from online spectral measurements during a solvent extraction experiment using a counter-current centrifugal contactor bank. The model predicted the pH of this validation data set within 11% for pH > 2, thus demonstrating that this technique could provide the capability of monitoring pH online in applications such as nuclear fuel reprocessing.

R

tion from Aqueous Komplexes (TALSPEAK) reprocessing was used as a model process. One approach to reducing the long-term radiotoxicity of irradiated nuclear fuel includes transmutation of the minor actinides, americium and curium, to short-lived or stable isotopes in nuclear reactors. To be most effective, these elements must be separated from the lanthanide fission products which in general are neutron poisons that greatly reduce the efficiency of the destruction of the minor actinide elements.1 TALSPEAK focuses on separating trivalent actinides from lanthanides. The aqueous phase of the TALSPEAK process typically consists of a lactic acid buffer and diethylenetriamine-N,N,N′,N″,N″-pentaacetic acid (DTPA) as the actinide hold-back reagent, while the organic phase contains the extractant bis(2-ethylhexyl)phosphoric acid (HDEHP).2 The TALSPEAK separation is strongly affected by solution pH and requires careful control to maintain optimal

enewed effort is focusing on the nuclear fuel cycle including treatment and reuse of irradiated fuel. For reprocessing, solvent extraction methods often contain various steps tailored to the separation of specific radionuclides, which are highly dependent upon solution properties. Acid strength/pH is critical for process quality and control, as it affects speciation of the target analytes and thus their extraction efficiency and selectivity. In a full-scale nuclear fuel reprocessing facility, classic potentiometric pH measurements are not suitable for obtaining real-time, continuous data due to the need for frequent calibration and maintenance and their poor long-term stability in aggressive chemical and radiation environments. To address this need, this work focused on developing a general method for determining pH of an aqueous carboxylate buffer by Raman spectroscopy, which can be used for online, real-time monitoring in a reprocessing system. This unique analytical capability is needed to remotely measure pH in highly radioactive and chemically hostile environments as encountered within a nuclear fuel reprocessing facility. To develop this method, the Trivalent Actinide-Lanthanide Separation by Phosphorus reagent Extrac© 2015 American Chemical Society

Received: December 9, 2014 Accepted: April 14, 2015 Published: April 14, 2015 5139

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry

Figure 1. (A) Raman spectra of 1 M lactate with a variable pH from 1.7 and 5.0 at an ionic strength of 1 M in the 600 to 2000 cm−1 range. (B) Raman spectra of 1 M lactate for pH 1.7 and 5.0 at an ionic strength of 1 M in the 2000 to 4000 cm−1 range. Alterations to the spectra due to changes in the pH can be utilized to develop online pH-monitoring capability during process operations.

operating conditions.3 Extraction mechanisms of this system include Ln 3 + + n (HA)2 ↔ 3H+ + LnA3(HA)2n − 3

(1)

LnBb3 − b + n (HA)2 ↔ (3 − b)H+ + LnBbA3 − b(HA)2n − (3 − b)

Na + + n (HA)2 ↔ H+ + NaA(HA)2n − 1

(2) (3)

where Ln3+ is a lanthanide metal ion, HA is HDEHP, and B is an anion such as nitrate.4−6 Spectroscopic techniques, such as Raman spectroscopy, have been used extensively for quantifying and analyzing solution compositions for both aqueous and organic phases,7−16 and they have been implemented into flow systems that provide real-time solution information.8,17 As shown in Figures 1 and S1, as lactate and lactic acid are Raman active, they provide a pathway for solution pH correlation and prediction. The lactate/lactic acid Raman spectra contain several bands18 that can be used for the determination and prediction of solution pH by online real-time monitoring. Figure 2 shows the speciation diagram for the relative percentage of lactic acid and lactate in the system as a function of solution pH. The relative abundance is a function of solution pH and the pKa for the lactic acid/lactate buffer system shown in eq 4. The speciation plots for lactic acid with the ionic strength (I.S.) varying from 0.1 to 0.5, 1.0, and 2.0 M are shown in Figure 2. The pKa values for 0.1, 0.5, and 1.0 M I.S. are nearly identical, but the curve for 2.0 I.S. is shifted by about 0.2 pH units.19 On the basis of this speciation diagram, for the pH = 1.7 solution shown in Figure 1 (blue spectrum), the main species present are 99% lactic acid and 1% lactate; for the pH = 5.0 solution (red spectrum in Figure 1), the main species present are 4% lactic acid and 96% lactate.

Figure 2. Speciation diagram for the lactic acid/lactate buffer system showing the speciation, HLac = lactic acid and Lac− = lactate ion. This speciation diagram is calculated on the basis of the ionic strength varying from 0.1 to 0.5, 1.0, and 2.0 M.

■

EXPERIMENTAL SECTION Methodology. To develop the pH-monitoring method utilizing online Raman spectroscopy, a database was constructed from the spectra of solutions of known composition. From this data set, a chemometric model was created on the basis of a multivariate analysis of the spectra. Validation of the method was performed by comparing measured values to those predicted by the model using an independent data set of spectra acquired during online measurements using a counter-current centrifugal contactor extraction system, similar to those used in nuclear fuel reprocessing. Database solutions were systematically varied in pH, ionic strength, and lactate/lactic acid concentrations. Feed solutions in the solvent extraction experiment were systematically varied in pH, with half containing 25 mM Nd3+, a realistic concentration for lanthanide metals within the TALSPEAK system. Neodymium was a model M(III) lanthanide/actinide ion for testing with a nonradioactive system. Materials. Lactic acid solutions were prepared from DL-lactic acid (85% w/w), purchased from Sigma-Aldrich. Sodium hydroxide (NaOH) (50% w/w solution in water), sodium chloride (NaCl) (with maximum 0.00002% Al), and neodymium(III) chloride hexahydrate (NdCl3·6(H2O)) were also obtained from Sigma-Aldrich. The NaCl was used to achieve 5140

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry desired ionic strengths. All aqueous solutions were prepared with ≥18 MΩ cm deionized water. The HDEHP (97% purity) and ndodecane (99%+ purity) used in the extraction experiment were purchased from Sigma-Aldrich and Alfa Asear, respectively, and the extractant mixture was water washed prior to use. Centrifugal Contactor System. For independent validation of the spectroscopic pH measurement method, aqueous solutions were generated using a solvent-extraction contactor bank. This bank consisted of four in-series, 2 cm, counter-current centrifugal contactors fed by Fluid Metering Incorporated Model QVG50 lab pumps. The inlet and outlet lines of both aqueous and organic phases were instrumented with Brooks Instrument Quantim Precision mass flow/density meters. The system contained 12 resistance temperature detectors; one on each interstage line, one on each phase inlet and outlet, and two recording the atmospheric temperature around the equipment. The system was equipped with National Instruments LabVIEW software for continuous data acquisition of the flow rate, density, and temperature. All inlet and outlet lines (aqueous and solvent pre- and postcontact) contained a Raman probe in a hightemperature, stainless steel sleeve with a quartz window, shown in Figure 3. For model validation, only the postcontact aqueous raffinate data was used to validate pH prediction.

sample. Raman spectra were acquired from all the samples at room temperature (20.4−20.9 °C), with the probe directly immersed into the solution. The Raman probe is enclosed within a stainless steel sleeve with a quartz optical window allowing direct solution measurement. Measurements used to validate the model were obtained through insertion of the Raman probe, with a protective sleeve, directly into the raffinate stream. A schematic showing the flow of the aqueous feed and the organic feed through the contactor with a description of the location of the probe within the raffinate stream is included as Figure S2 in the Supporting Information. The Raman probe and sleeve were sealed in the solution stream and configured with stainless steel lines on both sides of the probe assembly to eliminate ambient light interference. The inlet and outlet lines of both aqueous and organic phases of the contactor system were instrumented with an individual Raman probe. During the solvent extraction experiment, each spectrum was collected for a duration of 10 s with a 30 s delay between scans for each location. An Inphotonics/DiCon GP700 General Purpose Fiber-optic Switch allowed multiple Raman probes to obtain independent samples at various locations. MoleCue acquisition software (InPhotonics) with GRAMS 32 data manipulation software (Galactic Industries Corporation) was used to process all Raman data. Analytical Determination. The commercial lactic acid/ lactate stock was diluted with water to approximately 3 M and boiled for 24 h to eliminate esterification/polymerization effects.20 The lactic acid concentration was measured by titration using a Metrohm 800 Dosino Automatic Titration device equipped with Tiamo 1.2 software. The pH was adjusted as needed by addition of NaOH and pH was measured using an ORION 3 STAR meter along with the ORION 8103BNUWP ROSS Probe. A mV-calibration linear fit was created using either Ricca Chemical Company or Orion pH standards of 2, 4, 7, and 10, with the fit verified using independent pH standard measurements, which measured within 0.4% of the reported values. Solution molar ionic strength was calculated, per eq 5, on the basis of measured concentrations of the solution components: ionicstrength =

1 2

∑ Cizi 2

(5)

where C is the molar concentration of the ith ion present and z is the charge of the ion. Where calculated, ionic strength measurements included all solutes in solution. Ionic strength calculations were based upon speciation predicted by Hyperquad Simulation and Speciation (HySS)21 for the given solution compositions. Equations for the protonation of lactic acid and the 1:1, 1:2, and 1:3 lanthanide/lactate complexes were entered into the model set. Chemometric Model Development. Multiple solutions were used as a training set for predictive model construction. In total, 98 individual lactate solutions were prepared and analyzed using Raman spectroscopy to incorporate variations in pH, ionic strength, and lactate concentrations expected within the solvent extraction validation experiment. The solution compositions are detailed in Table S1 in the Supporting Information. The solutions used to develop the chemometric model contained approximately 0.25, 0.5, and 1.0 M lactate concentrations, with pH varying from 1.3 to 5 and ionic strengths ranging from 0.25 to 3 M. Each solution used was measured spectrally 10 times, then the spectra were averaged. Sample IDs 85 through 98 are stock solutions used as feed solutions in the contactor tests (Table S1,

Figure 3. Raman probe is enclosed within a stainless steel sleeve with a quartz optical window allowing for direct solution measurement. This system provides a means for online, real-time measurements during the solvent extraction experiment. The probe used to validate the method in this experiment was the postcontact aqueous stream exiting the countercurrent centrifugal contactor bank.

Raman Spectroscopic Measurements. Raman measurements were performed with an InPhotonics RS2000 Raman spectrometer containing a thermoelectrically cooled-charged coupled device (CCD) detector operating at −55 °C; a 670 nm, 150-mW, visible diode laser as the excitation source; and a focused fiber-optic InPhotonics RamanProbe operated in a 180°back reflection mode. The laser beam focal point was 5 mm beyond the end of the laser probe quartz window and into the interrogated solution. The measured laser intensity at the sample was typically 50 mW. For the solutions used to develop the chemometric model, an integration time of 10 s was used for each acquisition, and 10 acquisitions were taken and averaged per 5141

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry

final analysis, a pretreatment including a first derivative followed by mean centering was used. Figure 4 shows a subset of the training set data, comparing the raw Raman spectra to that subjected to a first derivative, followed

Supporting Information); three independent Raman measurements were taken of each of these feed stocks for use in the training set and averaged. Raw spectral files were collected into a matrix database within a MATLAB environment (version 7.9, Mathworks Inc., Natick, MA, USA). The data set contained 98 rows (samples) and 10 000 columns (variables, Raman intensity values at different wavenumbers). The selection of one or more spectral ranges containing chemical information, as well as the elimination of spectral ranges that do not contain analyte spectral information, are widely used strategies to improve partial least-squares (PLS) regression models.22,23 To construct the pH model, the spectral regions associated with the lactic acid/lactate bands in the Raman fingerprint region (500−2000 cm−1) and the OH region (3000− 4000 cm−1) were independently considered for inclusion within the model. Multivariate analysis of the spectroscopic data was performed using commercial software (PLS Toolbox version 6.2.1, eigenvector Research Inc., Wenatchee, WA, USA). The PLS regression method used in this study was applied to the data matrices to develop a quantitative predictive model for the concentration of each component by correlating the spectral data with concentration data within the database. The PLS method has been used extensively in the field of chemistry23,24 including modeling of spectroscopic data as detailed elsewhere.25 PLS analysis was performed using the statistically inspired modification of PLS, or SIMPLS, algorithm26 within the PLS Toolbox. An external validation data set, generated from the centrifugal contactor run, was used to evaluate the models’ performance. The quality of the models to predict the new data were assessed by evaluating (1) the regression coefficient (R2) of the fit of prediction; (2) the root-mean-square error of calibration (RMSEC); (3) the root-mean-square error of cross validation (RMSECV), where every tenth sample from the training set was selected to be used for prediction; and (4) the root-mean-square error of prediction (RMSEP). Before chemometric analysis, several pretreatment and transform steps were performed on the spectral data. A first derivative of the spectral data was utilized to reduce baseline offset effects. The derivatives of all the Raman spectra were computed by the Savitsky-Golay method,27 using second-order smoothing through a 15-point moving average. This step was followed by mean centering of the samples.23 Mean centering subtracts the mean absorbance value from each sample, placing the centroid of the data set at the origin and removing an overall bias from the data set. For systems where a zero point of the measurement scale is arbitrary (e.g., temperature measured in degrees Celsius) or a nonzero intercept is expected (such as pH), mean centering is recommended. The use of multiplicative signal correction (MSC),28 a method to reduce differences in spectra based on variable scattering not generally associated with spectral features, was also investigated. This technique was considered to compare reducing differences in training set data composed of dissolved analytes within a homogeneous aqueous phase (similar to feed solutions) to those used in the prediction data set composed of aqueous (raffinate) solutions contacted with an organic extraction phase. Such an evaluation is justified because the raffinate phases may contain dissolved organic extractant, as well as possible organic phase carryover, causing variable scattering due to fluorescence and light scattering from micellar organic particles within the aqueous phase. In practice, employing MSC as a pretreatment step did not improve the errors associated with the models; thus, it was not used. In the

Figure 4. Comparison of Raman spectra before and after application of preprocessing steps. (A) Raman spectra of solutions prior to preprocessing, (B) first derivative preprocessing applied to spectra, and (C) first derivative and mean centering applied to spectra. Solution compositions have variable pH (1.53−4.2), at total lactate concentration of 2.0 M and ionic strength of 2.0 M.

by mean centering of the data. The first derivative preprocessing step eliminates most of the effect of a changing baseline as can be seen in Figure 4B. The addition of mean centering to the data preprocessing is shown in Figure 4C. Solvent-extraction centrifugal contactor tests included a series of 1.0 M lactate aqueous feed solutions of ionic strength 2.0 M and the organic phase containing 0.5 M HDEHP in n-dodecane inside the bank of contactors. Half of the feed solutions contained 25 mM Nd3+, as this would cause larger shifts in pH between the feed and postcontact aqueous stream due to extraction of the metal (eqs 1 and 2). The test began with the most acidic feed stock (pH = 1.53). Once steady state was reached with the first feed stock, it continued running in the system for 50 min. After 35 min, collections were taken from the postcontact aqueous stream (raffinate) in increments of 5 min. After 15 min of collection, the next feed was introduced into the system. For all feeds, an organic/aqueous volume ratio of 1 was maintained with a flow rate of ≈10 mL/min for both streams. Testing was performed at ambient temperature, where the 5142

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry

of the protonated lactate to the 860 cm−1 band (νC−CO2−) of the anion form as the pH increases. Additional pronounced alterations also result from the dissimilarities between the protonated and anion forms in the bending and stretching modes of the carboxylate group.18 The bending modes for the protonated and deprotonated carboxalate group (δ O−C−O) changes from 750 to 775 cm−1 transitioning from the lactic acid to latate forms, respectively. Large changes are also observed in the carboxalate stretching modes; the strong ν CO band for lactic acid is replaced by the strong νas CO2− and νs CO2− bands for the lactate species observed at 1510 and 1420 cm−1, respectively. For the subset of samples containing 25 mM Nd3+ (Table S1, Supporting Information, Sample IDs 92−98), there were no changes in the Raman bands associated with the carboxylate based on adding Nd3+ when compared to samples of similar composition not containing this ion. As expected, Figure 5A shows the increase in intensity of all the lactate Raman bands as the lactate concentration increases within

laboratory room temperature varied from 21 to 24 °C throughout the experiment (i.e., 21 °C in the morning and increasing throughout the day). Temperatures recorded by the resistance temperature detectors in-stream next to the Raman probe showed the solution temperature began at 21 °C in the morning, elevated to 30 °C over the first 3 h of the test, and then remained constant for the remaining time (approximately 5 h). This increase in temperature was attributed to heat input from the motor of the contactors. Feeds 1 through 7 were measured in a single day, with feeds 8 through 14 measured on the following day. Similar temperature trends were observed on both days. To determine the temperature dependence of the lactate bands within the Raman spectra over the limited temperature range observed in this study (∼21 to 30 °C), a series of temperature dependent measurements were performed using lactic acid with variable pH. Measurements were made at temperatures ranging from 25 to 55 °C in 10 degree intervals. The greatest temperature dependence was observed in the water band region (3000 to 4000 cm−1) over the full temperature range. However, when limiting the temperature range between 25 and 35 °C, the variations in the spectra in the 500 to 2000 cm−1 range or 3000 to 4000 cm−1 range were virtually the same. This indicates that over the limited temperature range (∼10 °C range) used in this study, there was little or no effect to the resulting pH model. The samples collected from the postcontact raffinate stream throughout the experiment were used to independently verify the solution pH (by pH meter measurement) for comparison with the model predictions. As shown in Table 1, for each feed Table 1. Feed Solution Compositions Used in the SolventExtraction Experiment for Method Validationa

a

feed

pH

Nd3+ (mM)

1 2 3 4 5 6 7 8 9 10 11 12 13 14

1.53 1.88 2.11 2.64 3.18 3.63 4.20 1.29 1.49 1.74 2.37 3.07 3.57 4.16

0 0 0 0 0 0 0 25 25 25 25 25 25 25

Figure 5. (A) Raman signatures of 0.25 to 2 M lactate at an ionic strength of 2 M with pH between 3.5 and 3.7. (B) Raman signatures of 0.25 M lactate varying in ionic strength from 0.25 to 3 M with a pH between 2.9 and 3.1. Direction of the arrow indicates the trend with increasing lactate concentration and ionic strength in each respective figure. I.S. = ionic strength.

All solutions contained 1 M lactate and an ionic strength of 2 M.

solution, Raman spectral measurements were obtained online from the aqueous, postcontact, raffinate stream (effluent stream from contactor system) and averaged over three successive 5 min intervals after system steady state was obtained.

■

aqueous solution with the ionic strength of 2 M, with the pH held relatively constant between 3.5 and 3.7. Figure 5B shows the effects of variable ionic strength on the Raman signatures of 0.25 M lactate; here, the ionic strength was varied from 0.25 to 3 M with the pH held relatively constant, between 2.9 and 3.1. The largest alteration in the spectra in Figure 5B is in the water bending mode centered at 1650 cm−1, which overshadows the carboxylate bands at this wavelength, due to the low total concentration of lactate in these solution. There are small changes in the bands located at 830 to 860 cm−1, associated with

RESULTS AND DISCUSSION Effect of Solution pH, Lactate Concentration, and Ionic Strength on Raman Spectra. The changes in the Raman spectra, shown in Figure 1, reflect the molecular effects from proportioning between the protonated and anionic forms of lactic acid and lactate ion, respectively. These changes correlate directly to the pH of the solution and provide the basis for monitoring pH using this method. A prominent change in the spectra results from a shift from the 830 cm−1 band (νC−CO2H) 5143

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry

Figure 6. Raman water region (2700−3900 cm−1) with variations in (A) pH with 0.25 M lactate at 1.0 M ionic strength, (B) ionic strength with 0.25 M lactate and a pH between 2.0 and 2.5, (C) lactate concentration with 2.0 M ionic strength and a pH between 3.5 and 3.8, and (D) NaCl from 0.001 to 5 M (no lactate in solution). For (C), the decrease at ≈3400 cm−1 most likely results from decreasing Cl− to maintain an ionic strength of 2 M.

the C−CO2H and C−CO2− stretching modes, which can be explained by a change in pKa due to the change in I.S. of the system. On the basis of the speciation diagram in Figure 2, the relative contribution of the protonated to deprotonated forms of lactic acid to lactate at pH = 3 changes from 20% (protonated) and 80% (deprotonated) at I.S = 0.1 M to 13% (protonated) and 87% (deprotonated) at I.S = 2.0M. The effects of solution pH, lactate concentration, and ionic strength on the Raman water stretching region (3100−3800 cm−1) are shown in Figure 6. In the spectra of solutions containing constant lactate concentration (0.25 M) and ionic strength (1.0 M), there are only minor alterations in the water region as a result of varying the pH (ranging from 2.3 to 4.6), as shown in Figure 6A. However, large changes in the water region are observed with variable ionic strength (0.25−3.0 M), as shown in Figure 6B. By increasing lactate concentration from 0.25 to 2.0 M (Figure 6C), the bands resulting from the CH and CH3 bending18 appear in the 2800−3100 cm−1 region. Additional spectral variations are observed in Figure 6C in the region between 3100 and 3900 cm−1 and attributed to the differing Cl− concentrations added to maintain the 2.0 M ionic strength.11 For comparison, the effects on the water spectrum due solely to NaCl are shown in Figure 6D. The Cl− anion produces a strong effect on the water region due to its breaking the tetrahedral-hydrogen-

bonded water structure as an effect of ion hydration, which affects the intermolecular ion−water stretching vibration and polarization of the water molecules and influences hydrogen bond strength.29−32 PLS Regression and Model Development. A total of 98 individual lactate solutions were measured by Raman spectroscopy. The resulting spectra along with their related compositions were used to develop a model for pH prediction. The solution compositions are given in Table S1, Supporting Information. The measured pH values are apparent values and will be affected by ionic strength on the junction potential of the pH probe. To address this point, we have calculated the actual pH for the training set data (solutions 1−98 in Table S1, Supporting Information) using the HySS (Hyperquad simulation and speciation) program,21 on the basis of the known composition of these solutions and accepted pKa values. The measured pH values are lower than the calculated pH values for these 98 samples by an average value of 0.341 pH units (std. dev. = 0.188). If the calculated pH rather than the measured pH is included within the regression model, this would yield a predicted pH value related to the true (calculated) pH rather than the measured pH value. However, the data set used to validate the model is a raffinate stream of an extraction process with an unknown chemical composition. Therefore, the pH value was 5144

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry Table 2. Parameters of PLS Models Used for Predicting Solution Compositiona preprocessingb

wavenumber range (cm−1)

original variables (J)

latent variable (LV)

# of samples (I)

RMSEC

RMSECV

R2 (cal)

# of validation samples (I)

RMSEP

first der, MC first der, MC

500−2000 500−4000

3100 7490

7 8

98 98

0.108 0.113

0.277 0.269

0.987 0.985

39 39

0.408 0.916

a

The number of individual sample measurements (I), number of original variables (J, wavelengths), latent variables (LV, PLS principal components) used in the model, root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV), and root mean square error of prediction (RMSEP) calculated by the model performance are included. bPreprocessing steps: first der = application of first derivative to spectrum; MC = mean centering of data.

Table 3. Samples Taken during the Solvent Extraction Experiment with Measured Solution Compositions and Acid Concentrations Predicted by the Developed Chemometric Model through Analysis of Real-Time Raman Spectraa pH of raffinate (post-contact aqueous stream after solvent extraction) first collection

second collection

third collection

feed #

pH measured before solvent extraction

measured

predicted

measured

predicted

measured

predicted

1 2 3 4 5 6 7 8 9 10 11 12 13 14

1.53 1.88 2.11 2.64 3.18 3.63 4.2 1.29 1.49 1.74 2.37 3.07 3.57 4.16

1.38 1.61 1.82 2.25 2.86 3.28 3.68 0.83 0.97 1.19 2.03 2.81 3.30 NM

1.85 2.04 2.10 2.20 2.55 3.19 3.89 1.62 1.79 2.03 2.00 2.38 3.08 NM

1.39 1.63 1.83 2.27 2.86 3.31 3.71 0.83 0.98 1.20 2.08 2.82 3.32 NM

1.88 1.92 2.14 2.24 2.59 3.18 3.71 1.52 1.65 1.92 2.10 2.64 3.16 NM

1.39 1.64 1.85 2.29 2.87 3.31 3.70 0.83 0.99 1.20 2.10 2.91 3.31 NM

1.77 2.08 2.06 2.37 2.75 3.32 3.86 1.48 1.88 1.96 2.25 2.62 3.21 NM

a

These solutions were used as validation for the model. All feeds contained 1 M lactate and an ionic strength of 2 M. Feeds 1−7 contained no Nd3+ while feeds 8−14 contained 25 mM Nd3+; multiple samples were taken during each feed. NM = not measured.

wavenumber regions, with results showing that, by excluding the water region, the model errors (RMSEC, RMSECV, and RMSEP) are all less than when using the water region, with the error in prediction (RMSEP) being larger by a factor of approximately three. The increased error in prediction of pH by including the water region can be explained by a study of the data shown in Figure 6. The Raman O−H stretching bands show no change in position or intensity over the pH range of 2.3 to 4.6, under constant lactic acid and ionic strength concentrations of 0.25 and 1.0 M, respectively, as seen in Figure 6A. However, the variation of the intensities of the Raman bands due to changes in ionic strength and lactic acid concentrations (shown in Figure 6B,C, respectively) is large. Therefore, this spectral region is much more sensitive to changes in ionic strength effects versus changes in pH alone. For this reason, the model using the Raman region confined to the 500−2000 cm−1 region was chosen. All pH values predicted by the model and their measured values are listed in Table 3. Figure 7 contains the plot of the data showing the predicted pH (determined by the model) as a function of the pH measured by a potentiometric pH meter. Reduction of pH in the postcontact solutions compared to the feed solutions was observed for all samples due to the occurrence of two equilibrium processes: deprotonation of HDEHP and its conversion to the sodium salt (eq 3) upon contact with aqueous lactate buffer solution and Nd3+ extraction by an ion exchange mechanism and formation of the Nd-HDEHP complex (eqs 1 and 2). Prediction of the raffinate pH near or in the lactate buffer zone (pH 2−5) remained within the 95% confidence interval of the

unable to be calculated on the basis of the HySS model for comparison. While the measured pH values have a bias due to ionic strength effects, both the training set and validation set are able to be measured in the same manner. Since the bias between the training set and validation set are assumed to be the same (similar ionic strengths in both sets of solutions), the measured pH vs model prediction pH values are then able to be compared on an equal basis. To validate the techniques, variations in pH, ionic strength, and lactate concentrations expected within the solvent extraction experiment were incorporated into the experimental design. The parameters used in the various PLS models to predict solution pH are displayed in Table 2. The preprocessing steps included first derivative followed by mean centering of the data. The models were validated using the Raman spectra taken from the online contactor measurement of the 39 independent solution measurements listed in Table 3. When using the 500 to 2000 cm−1 region, the error of the model calibration (RMSEC) was 0.108, and the model prediction (RMSEP) error was 0.408, with the RMSEP approximately four times greater than the RMSEC. For comparison, the error of prediction based on a cross validation (RMSECV) method utilizing a subset of the training set yielded errors approximately 2.5 times greater than the RMSEC values. During model development, the 500−2000 cm−1 region, which includes the lactic acid/lactate fingerprint region, was used. We also investigated using the entire Raman spectrum (500−4000 cm−1), which includes the water stretching bands. Table 2 shows a comparison of the model using these two 5145

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry

emulsion began to form and the test was ended prematurely to prevent complications due to plugging of the centrifugal contactors. As a result, solution measurements and pH predictions were not performed on the raffinate solutions resulting from this feed. A photograph of the emulsion is shown in Figure S3 in the Supporting Information.

■

CONCLUSIONS This work examined the applicability of using Raman spectroscopy to measure pH in a used nuclear fuel reprocessing separation system. The testing evaluated pH, lactate concentration, and ionic strength effects on the Raman spectrum. Data analysis showed dependence of the Raman spectrum on these solution properties, which was utilized to develop a predictive model for solution pH. The developed PLS model was applied to online spectra obtained during solvent extractions performed using a centrifugal contactor bank. The model predicted the pH within 11% for pH > 2, thus validating the model and demonstrating that this technique could provide the capability of monitoring pH online in nuclear fuel reprocessing. This technique can be expanded for determining solution pH in other buffer systems potentially applied in the advanced TALSPEK processes such as citrate or malonate. The effect of aqueous complexant agents (e.g., DTPA and HEDTA) on the performance of the online Raman pH-monitoring method should be investigated. The model could readily be expanded through incorporation of additional spectra or coupling with additional physiochemical measurements to further enhance predictive accuracy.

Figure 7. Chemometric model results for the prediction of pH, including solutions from model development along with calculated fit, 95% confidence intervals, and independent samples from solvent extraction with and without Nd3+ in the feed.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org/.

model and within 11% of the measured value. For pH < 2, greater deviations were observed between the model-determined and measured values. Below pH ∼ 2, the lactic acid is expected to be ≥98% fully protonated and therefore the lactic acid/lactate system is outside of its buffer region. In this low pH region, the pH dependent changes in ratios of lactic acid to lactate ion are not a sensitive indicator for pH prediction. The minimal change in speciation of lactic acid/lactate in this pH region can be verified in the speciation plot shown in Figure 2 indicating that lactic acid is essentially fully protonated for pH values ≤2. Deviation from the 95% confidence interval for raffinate solutions is particularly pronounced for samples with measured pH ≤ 1.2. The raffinate samples with measured pH ≤ 1.2 all correspond with solutions that contain Nd within the original feed (Feed #s 8−10 in Table 3). In general, the raffinate solutions containing Nd (Feed #s 8−13, Table 3) have a lower measured pH than for their counterpart solutions containing no Nd (Feed #s 1−7, Table 3). This difference in pH is more pronounced for the lower pH samples (compare solutions 8−10 with Nd to solutions 1−3 without Nd). The main reason for the lower measured pH values for solutions 8−13 than for their counterpart solutions (1−3) is derived from the ion exchange reaction, where Nd3+ is exchanged for H+ in the solvent extraction process (eq 1). This significantly lowers the pH of the feeds originally containing Nd3+ (solutions 8−13), while the feed solutions containing no Nd are changed to a lesser degree when these solutions’ starting pH levels are at or below the buffer region of the lactic acid/lactate system. During testing, when the final feed (Table 3, Feed # 14: pH of 4.16 with 25 mM Nd3+) was initiated in the system, an interfacial

■

AUTHOR INFORMATION

Corresponding Authors

*E-mail: [email protected]. *E-mail: [email protected]. Author Contributions

All authors gave approval to the final version of the manuscript. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This research was supported by the Laboratory Directed Research and Development Program at the Pacific Northwest National Laboratory operated by Battelle for the U.S. Department of Energy under Contract DE-AC05-76RL01830 and by the U.S. Department of Energy’s Fuel Cycle Research and Development (FCR&D), Separation Campaign (NE).

■

REFERENCES

(1) Nash, K. L. Solvent Extr. Ion Exch. 2015, 33 (1), 1−55. (2) Nilsson, M.; Nash, K. L. Solvent Extr. Ion Exch. 2007, 25 (6), 665− 701. (3) Nilsson, M.; Nash, K. L. Solvent Extr. Ion Exch. 2009, 27 (3), 354− 377. (4) Peppard, D. F.; Mason, G. W.; Maier, J. L.; Driscoll, W. J. J. Inorg. Nucl. Chem. 1957, 4 (5−6), 334−343. (5) Lundqvist, R.; Lu, J. F.; Svantesson, I. Acta Chem. Scand., Ser. A 1983, 37 (9), 743−753. 5146

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147

Article

Analytical Chemistry (6) Melnik, M. I.; Spiryakov, V. I.; Filimonov, V. T.; Karelin, E. A. J. Alloys Compd. 1998, 275, 863−867. (7) Baumgartner, F.; Ertel, D. J. Radioanal. Chem. 1980, 58 (1−2), 11− 28. (8) Bryan, S. A.; Levitskaia, T.G.; Sinkov, S. Schlahta, S.N. Raman Based Process Monitor For Continuous Real-Time Analysis of High Level Radioactive Waste Components. In Waste Management Conference, Phoenix, AZ, February 24−28, 2008. (9) Bryan, S. A.; Levitskaia, T. G.; Johnsen, A. M.; Orton, C. R.; Peterson, J. M. Radiochim. Acta 2011, 99, 563−571. (10) Burck, J. Anal. Chim. Acta 1991, 254 (1−2), 159−165. (11) Casella, A. J.; Levitskaia, T. G.; Peterson, J. M.; Bryan, S. A. Anal. Chem. 2013, 85 (8), 4120−4128. (12) Colston, B. J.; Choppin, G. R. J. Radioanal. Nucl. Chem. 2001, 250 (1), 21−26. (13) Guillaume, B.; Begun, G. M.; Hahn, R. L. Inorg. Chem. 1982, 21 (3), 1159−1166. (14) Madic, C.; Begun, G. M.; Hobart, D. E.; Hahn, R. L. Radiochim. Acta 1983, 34 (4), 195−202. (15) Madic, C.; Begun, G. M.; Hobart, D. E.; Hahn, R. L. Inorg. Chem. 1984, 23 (13), 1914−1921. (16) Meinrath, G. J. Radioanal. Nucl. Chem. 1997, 224 (1−2), 119− 126. (17) Armenta, S.; Garrigues, S.; de la Guardia, M. TrAC, Trends Anal. Chem. 2007, 26 (8), 775−787. (18) Cassanas, G.; Morssli, M.; Fabregue, E.; Bardet, L. J. Raman Spectrosc. 1991, 22 (7), 409−413. (19) Martell, A. E.; Smith, R.M. Critical Stability Constants, Vol. 3: Other Organic Ligands; Plenum Press: New York, 1977. (20) Yao, J. Y.; Lim, L. T.; Li, Y. J. Preparation and Characterization of Poly(Lactic Acid) with Different Molecular Weights by Thermal Hydrolysis. In Advanced Materials and Sports Equipment Design; Yang, D., Zhang, T. B., Luo, Q., Eds.; Trans Tech Publications Ltd: StafaZurich, 2014, p 19−24. (21) Alderighi, L.; Gans, P.; Ienco, A.; Peters, D.; Sabatini, A.; Vacca, A. Coord. Chem. Rev. 1999, 184, 311−318. (22) Beebe, K. R.; Pell, R. J.; Seasholtz, M. B. Chemometrics, A Practical Guide; John Wiley and Sons: New York, 1998. (23) Sharaf, M. A.; Illman, D. L.; Kowalski, B. R. Chemometrics; Kolthoff, I. M., Ed.; Wiley and Sons: New York, 1986; Vol. 82. (24) Martens, H.; Naes, T. TrAC, Trends Anal. Chem. 1984, 3 (8), 204−210. (25) Levitskaia, T. G.; Bryan, S. A.; Creim, J. A.; Curry, T. L.; Luders, T.; Thrall, K. D.; Peterson, J. M. Drug Dev. Res. 2012, 73 (5), 252−273. (26) de Jong, S.; Wise, B. M.; Ricker, N. L. J. Chemom. 2001, 15 (2), 85−100. (27) Savitzky, A.; Golay, M. J. E. Anal. Chem. 1964, 36 (8), 1627−1639. (28) Martens, H.; Naes, T. Multivariate Calibration; John Wiley & Sons: New York, 1989. (29) Barnett, R. N.; Landman, U. Phys. Rev. B 1993, 48 (4), 2081− 2097. (30) Bergstrom, P. A.; Lindgren, J. J. Mol. Struct. 1991, 245 (3−4), 221−229. (31) Carey, D. M.; Korenowski, G. M. J. Chem. Phys. 1998, 108 (7), 2669−2675. (32) Li, R. H.; Jiang, Z. P.; Guan, Y. T.; Yang, H. W.; Liu, B. J. Raman Spectrosc. 2009, 40 (9), 1200−1204.

5147

DOI: 10.1021/ac504578t Anal. Chem. 2015, 87, 5139−5147