Anal. Chem. 2000, 72, 3392-3395
Determination of Water in NIST Reference Material for Mineral Oils Anders Cedergren* and Ulrika Nordmark
Department of Chemistry, Analytical Chemistry, Umeå University, S-901 87 Umeå, Sweden
The accuracy of the reference concentrations of moisture in electrical insulating oil RM 8506 and lubricating oil RM 8507 (both of mineral type) and specified by the National Institute of Standards and Technology (NIST) as containing 39.7 and 76.8 ppm (w/w) water, respectively, has recently been the subject of debate in this journal. To shed some further light on this controversy, we report in this correspondence results for these oils obtained by two additional methods, one based on specially designed reagents for diaphragm-free Karl Fischer (KF) coulometry and the other based on the concept of stripping at elevated temperature/continuous KF coulometry. A positive interference effect was shown to take place for RM 8506 when the direct coulometric method was used. If the results are corrected for this, the values including six different procedures varied in the range 13.5-15.6 ppm (w/w). For RM 8507, all values were between 42.5 and 47.2 ppm (w/w), which means that the values recommended by NIST for both reference oils using volumetric titration are about twice as high as those obtained with the other techniques. A possible explanation for this discrepancy is presented. The accuracy of the reference mass concentration of moisture in transformer oil, RM 8506,1 and lubricating oil, RM 8507,2 as specified by the National Institute of Standards and Technology (NIST) has recently been questioned by Jalbert et al.3 in this journal. According to Reports of Investigation1,2 from NIST, the correct values for these oils were stated to be 39.7 ( 2.8 (w/w) and 76.8 ( 2.3 ppm (w/w), respectively, although the consensus amounts of moisture resulting from two round robin studies among 14 laboratories based on the coulometric method recommended by ASTM4 are 21 ( 3 (w/w) and 47 ( 4 ppm (w/w), respectively. In a series of papers,5-8 Margolis, representing NIST, * Corresponding author: (e-mail)
[email protected]. (1) National Institute of Standards and Technology. Reports of Investigation on Reference Material 8506. Moisture in Transformer Oil; revision of June 13, 1997. (2) National Institute of Standards and Technology. Reports on Investigation on Reference Material 8507. Moisture in Mineral Oil; revision of June 13, 1997. (3) Jalbert, J.; Gilbert, R.; Te´treault, P. Anal. Chem. 1999, 71, 3283-3291. (4) ASTM D 1533-88 Standard Test Method for Water in Insulating Liquids (Karl Fischer Method). ASTM Annual Book of Standards; ASTM: West Conshohocken, PA, 1993; Vol. 10.03, p 189. (5) Margolis, S. A. Anal. Chem. 1995, 67, 4239-4246. (6) Margolis, S. A. Anal. Chem. 1997, 69, 4864-4871. (7) Margolis, S. A. Anal. Chem. 1998, 70, 4264-4270.
3392 Analytical Chemistry, Vol. 72, No. 14, July 15, 2000
argued that the results from the round robin study are biased low because the solvents used do not adequately dissolve the oil matrix. A prerequisite for this, according to Margolis, is that the titration is carried out in a medium containing at least 65% (v/v) chloroform. In a recent paper,8 the same author showed that water in oil that is not measured by the coulometric method, when the oil is incompletely dissolved, resides in the oil phase of the heterogeneous (multiphasic) suspension. It should be mentioned that this conclusion was based on determinations of residual water in as much as 30 mL of oil using the same volumetric method as is being questioned in this paper. For the above-mentioned reference oils, Jalbert et al. reported3 values in the range 13.0-14.8 ppm (w/w) for RM 8506 and 42.546.4 ppm (w/w) for RM 8507 using three different techniques: direct coulometry with the Mitsubishi instrument (diaphragm), azeotropic distillation/continuous coulometry (diaphragm), and headspace gas chromatography. It should be mentioned that these results were all based on systems that were calibrated by means of standards prepared in mineral oil. Jalbert et al.3 stated that the argument used by Margolis,5-8 namely, that low results are a consequence of insufficient dissolution of the oil, does not hold for their results since special attention was paid to the selection of conditions for complete dissolution of the oil matrix by means of a coulometric reagent based on a mixture of pentanol and chloroform. To shed some further light on this controversy, we report in this paper results for these oils obtained by two additional methods. One is based on direct coulometry using a modified reagent designed for diaphragm-free coulometry9 and the other on the concept of stripping at elevated temperature/continuous coulometry.10,11 In addition, the influence of the oil matrix on the Karl Fischer (KF) system was studied by using reagents without any sulfur dioxide present, i.e., in a medium in which no reaction with water takes place. EXPERIMENTAL SECTION Chemicals. Imidazole (pa) and sulfur dioxide (>99.9%) were from Fluka. Iodine (pa), trichloroacetic acid (TCA) (pa), imidazolium iodide (pa), and hexanol (99%) were from Riedel-de-Hae¨n. Methanol was from KeboLab. Chloroform (pa) was from Merck. Safety Considerations. Methanol is highly flammable and is toxic by inhalation, in contact with skin, and if swallowed. (8) Margolis, S. A. Anal. Chem. 1999, 71, 1728-1732. (9) Nordmark, U.; Cedergren, A. Anal. Chem. 2000, 72, 172-179. (10) Rosvall, M.; Lundmark, L.; Cedergren, A. Anal. Chem. 1998, 70, 53325338. (11) Cedergren, A.; Lundstro ¨m, M. Anal. Chem. 1997, 69, 4051-4055. 10.1021/ac9913006 CCC: $19.00
© 2000 American Chemical Society Published on Web 06/03/2000
Table 1. Compositions of the Methanolic Reagents Used reagent
[Im] (M)
[TCA] (M)
[SO2] (M)
[I2] (M)
[ImH+I-] (M)
[hexanol] (%, v/v)
[chloroform](%, v/v)
A B C D E F
3 5.5 6.5 1.4 6.5 1.2
0 0 0 0.5 0.5 0.5
0.25 0.51 0.6 0 0 0
0.1 0.1 0.1 0 0 0
0 0 0 0.2 0.2 0.2
0 0 10 0 0 14
0 40 30 40 40 30
Chloroform is harmful by inhalation and ingestion and may be fatal. Inhalation of vapor may cause headache, nausea, vomiting, and dizziness. Prolonged skin contact may result in dermatitis. The liquid is readily absorbed through the skin. Imidazole is harmful by inhalation, in contact with skin, and when swallowed. Sulfur dioxide is intensely irritating to eyes and the respiratory tract. Iodine is a poison and may be fatal if swallowed; avoid contact with eyes, skin, or clothing, do not breathe the vapor, and use with adequate ventilation. Trichloroacetic acid is a poison and causes severe burns; avoid contact with skin and eyes. Reagents. Except for the Hydranal Coulomat AG oven (RiedeldeHa¨en), all reagents were homemade, the compositions of which are given in Table 1. The reagents containing sulfur dioxide were decolored with water directly after preparation. Instrumentation and Methods. Direct Coulometry. For all experiments using direct coulometry, 100 mL of reagent B, C, or D was transferred to the diaphragm-free cell of the Metrohm 756 coulometric system. The maximum generating current, 400 mA, was used in combination with the maximum titration rate, 2240 µg min-1. Drift values between 0.2 and 1.8 µg min-1 were obtained for both reference oils when reagents B and C were used. The preset ac bipotentiometric end point potential was normally 70 mV, corresponding to an iodine excess concentration of (1-5) × 10-5 M. To ensure that the entire titration curve was registered by the computer, in which the data program Vesuv 3.0 (Metrohm; Herisau, Switzerland) had been installed, an extraction time of 600 s was normally selected. The compositions of reagents B and C were chosen in accordance with recently reported findings9 in order to minimize the extent of formation of oxidizable reduction products formed in the cathodic reaction. Relative errors in the range 0.2-1% were expected for water determinations in RM 8506 and RM 8507 using direct diaphragm-free coulometry for such reagents. Self-oxidation of iodide to iodine may cause a negative drift (leading to an overtitration) when the SO2-free reagent is used. To cope with this problem, a reference electrode,12,13 in which a fine porous Vycor glass constituted the liquid junction, was always kept in contact with the titration vessel solution. The reference electrode was filled with Hydranal Coulomat AG oven, which contains ∼1 M sulfur dioxide. This compound slowly diffuses into the reagent where it reacts with iodine in the presence of water (∼0.03%) and this resulted in a desired small positive drift between 0 and 0.1 µg min-1 (calculated as water). Stripping/Continuous Coulometry. The strip cell (5 mL) used has been described before,11 and this was used in combination with a diaphragm-free cell12 connected to a computer-controlled coulometric titrator.10 Preheated, dried air at a flow rate of 50 mL (12) Cedergren, A.; Jonsson, S. Anal. Chem. 1997, 69, 3100-3108. (13) Cedergren, A.; Luan, L. Anal. Chem. 1998, 70, 2174-2180.
min-1 was passed through the stripper held at 105 °C and bubbled through the coulometric cell containing 13 mL of reagent A. A 150-400 mg sample of reference oil was injected into the stripper. Before the syringe was filled with the sample, it was carefully rinsed with a dry oil. Contamination due to adsorbed water on the surface of the needle was determined in separate experiments to be less than 0.1 µg water, which corresponds to ∼0.3 ppm (w/w) water in the oil. Direct Zero-Current Potentiometry. The instrumentation and procedure for the direct potentiometric measurements of interference effects caused by RM 8506 and RM 8507 were the same as described before.13 A 0.4-0.8 g sample of oil was injected into the cell containing 22 mL of reagent E or F. The change in redox potential in the range 5 × 10-5-10-4 M was followed (1 measurement/s) as a function of time by help of a Fluke dual-display multimeter 45 connected to a computer in which the data collection program Fluke QS was available. By using the theoretical Nernst slope value and the actual temperature and volume, the total consumption or production of iodine equivalents could be calculated and this value was then transformed into a value corresponding to ppm water interference expected for the oil samples. Investigation of the NIST Volumetric Method. These experiments were undertaken in order to understand the reason for the much higher values reported by Margolis/NIST1,2,5 using their concept volumetric titration/amperometric detection in reagent mixtures with chloroform concentrations above 65% (v/v). We used the cell designed for direct zero-current potentiometric measurements13 equipped with a biamperometric indicating system including two identical platinum electrodes (area 0.08 cm2; polarization voltage 20 mV). In this way, the signals from both systems could be followed simultaneously. The same type of reagent mixture as described by Margolis5 was prepared by mixing 70 mL of 6:1 (v/v) chloroform/methanol with 20 mL of methanolic pyridine-buffered KF reagent containing 0.1 M iodine, 1.5 M sulfur dioxide, and 3 M pyridine (reagent G). Before use, this reagent was nearly decolored by careful addition of pure water. To compensate for the well-known relatively slow reaction rate between water and pyridine-buffered KF reagents, calibrations were carried out at somewhat higher concentrations of iodine than normal, (2-5) × 10-4 M. Water standards containing 50 µg of water were injected into reagent volumes between 23 and 26 mL, and a few seconds were sufficient to obtain stable potential readings. RESULTS AND DISCUSSION The existence of a positive bias for trace water determinations in oils such as naphthenic and aromatic oils, 3 and 9 ppm (w/w), Analytical Chemistry, Vol. 72, No. 14, July 15, 2000
3393
Table 2. Results Obtained for RM 8506 and RM 8507 Using Different Combinations of SO2-Free Reagents and Electrochemical Techniques reagent
method
D E F
coulometry/ac bipotentiometrya zero-current potentiometryb zero-current potentiometryc
RM 8506 RM 8507 (ppm H2O) (ppm H2O) 6.3 ( 0.5 7.3 ( 1.2 8 ( 1.0
negative negative negative
a 1 g samples, 100 mL medium, 2 × 10-5 M iodine end point concentration. b 0.4-0.5 g samples, 22 mL medium, 5 × 10-5-10-4 M iodine excess. c 0.7-0.8 g samples, 22 mL medium, 5 × 10-5-10-4 M iodine excess.
respectively, has been demonstrated by Jones and Mayne14 for coulometric as well as volumetric KF titrations. To investigate the influence of the oil matrixes of RM 8506 and RM 8507 on the KF titration system, three different types of reagents without any sulfur dioxide were designed. Since 1 mol of sulfur dioxide produces ∼1 mol of protons in a methanolic KF reagent at pH 7 and above, this was compensated for in the SO2-free reagents by including the relatively strong trichloroacetic acid (pKa ) 4.7). By selecting different initial concentrations of imidazole (pKa for protonated imidazole is ∼9), the quotient of free imidazole and protonated imidazole could be varied in such a way as to resemble the conditions prevailing in modern types of imidazole-buffered reagents.9 The results obtained with two principally different end point techniques, zero-current potentiometry and ac bipotentiometry (Metrohms system), are summarized in Table 2, where it can be seen that all values for RM 8506 are within the range 6-8 ppm (w/w) despite a relatively large variation in the experimental conditions. A small but significant increase in the values was observed at the higher iodine concentrations within the interval studied ((2-10) × 10-5 M). From this we conclude that the interference effect is mainly caused by a rapid quantitative chemical reaction consuming iodine rather than an electrochemical phenomenon. The results for RM 8507 are not that clear-cut since for this oil a negative interference effect was observed in all experiments. This indicates that iodine is produced in the SO2free reagent as a consequence of a reaction between iodide and some constituents present in the oil. This reaction was found to be relatively slow, which made quantification difficult. However, it was interesting to notice that when this oil was injected into the ordinary KF reagent C there was no indication of an interference reaction as reflected by the excellent stability of the baseline. One possible explanation for the absence of the negative interference effect in the ordinary KF reagent may be that the sulfur component (methyl sulfite) reduces the interfering agent much more rapidly as compared to a possible reaction between iodide and the interfering agent resulting in the formation of iodine. This is a speculation and such a scenario requires that the reduced interfering agent is not oxidizable by iodine. In earlier work on water determinations in vegetable oils,11 the same negative interference effect was observed for most oils, and in this case as well, there was no indication that it affected the results obtained by direct coulometry. In Figure 1, typical signals obtained (14) Jones, C. F.; Mayne, A. Applications and Limitations of the Karl Fischer Technique of Moisture Analysis in Electrical Plant Insulation. Proc. 4th Int. Conf. Properties Appl. Dielectric Mater., 1994, 895-898.
3394 Analytical Chemistry, Vol. 72, No. 14, July 15, 2000
Figure 1. Recorded titration curves obtained with the Metrohm 756 coulometer and diaphragm-free cell containing 100 mL of reagent D (free of sulfur dioxide). The end point concentration of iodine was 2 × 10-5 M.
for both mineral oils using the Metrohm coulometric system and the SO2-free reagent D are presented. A summary of the analytical results obtained in this work using two different techniques is given in Table 3 along with those reported by others for RM 8506 and RM 8507. If the values for direct coulometry obtained by us and those reported in the round robin study1,2 are corrected for the positive interference effect discussed above for RM 8506 (7.5 ( 1.5 ppm (w/w)) all values, except for that reported by Margolis,5 are within the range 13.515.6 ppm (w/w). In view of the relatively large number of different techniques, these results strongly indicate that the value of Margolis is biased high. It should be pointed out that the value reported by Jalbert et al.3 was obtained with a coulometric instrument calibrated with mineral oil standards, which means that any interference effect is likely to be already compensated for. The same trend is also seen for RM 8507; the value reported by Margolis (representing NIST), 76.8 ppm (w/w), is significantly higher than those obtained by the other techniques, 42.5-47.2 ppm (w/w). The good agreement between the different methods, except for the volumetric procedure, indicates that the negative interference effect observed in the experiments with the SO2-free reagents does not show up when direct KF coulometric methods are used. It is obvious that the 20-30 ppm (w/w) higher values reported by Margolis/NIST using their volumetric method are not caused by a chemical interference effect. This method involves the use of a biamperometric end point detection system but unfortunately no experimental details are given in the paper of Margolis.5 Nevertheless, we tried to repeat the experiments using exactly the same reagent mixture (reagent G). The results are shown in Figure 2, where it can be seen that the biamperometric response curve is markedly changed upon addition of the RM 8506 to the reagent mixture. It should be emphasized that the key for the establishment of the response curves given in the figure was the
Table 3. Summary of Results Obtained with Different Techniques for Determination of Water in RM 8506 and RM 8507 Sample Assays this worka direct coulometry mean value (ppm)
std dev (ppm)
23.1d 22.2e 47.2f
this work stripping/coulometry
Jalbert et al.3 direct coulometryb
mean value (ppm)
std dev (ppm)
mean value (ppm)
std dev (ppm)
0.6 0.5
14.1
0.1
13.0
0.8
1.2
45.2
0.3
42.5
1.9
Jalbert et al.3 azeotropic distillation/ coulometry mean value (ppm)
Jalbert et al.3 headspace GC
round robin1,2 direct coulometry ASTM4
Margolis5 volumetricc
std dev (ppm)
mean value (ppm)
std dev (ppm)
mean value (ppm)
std dev (ppm)
mean value (ppm)
std dev (ppm)
RM 8506 14.8
0.6
13.0
0.4
21
3
39.7
3.3
RM 8507 44.5
1.1
46.4
0.8
47
4
76.8
4.8
a Metrohm 756 with diaphragm-free cell. b Mitsubishi Moisturemeter CA-06 (with diaphragm). c Amperometric titration with standard pyridine reagent added to a pretitrated solution containing 65% (v/v) chloroform in methanol. d Sample sizes between 0.7 and 0.9 g (four injections); 100 mL of reagent B (40% (v/v) chloroform). e Sample sizes between 0.7 and 0.8 g (three injections); 100 mL of reagent C (10% (v/v) hexanol, 30% (v/v) chloroform). f Sample sizes used: 0.4, 0.8, and 1.9 g (three injections). Same reagent as was used for footnote e.
Figure 2. Influence of RM 8506 on the biamperometric response curve.
use of a zero-current potentiometric indicating system inserted in the same cell. Since this system showed a perfect Nernst slope value ((0.05 mV), in the absence as well as in the presence of the oil matrix in the range (5-50) × 10-5 M iodine excess, it was possible to establish the relation between the amperometric signal (µA) and the excess concentration of iodine (M). Since the volumetric titration of Margolis is carried out using a preselected end point (i.e., preselected value of the diffusion current), the expected error at different iodine excess concentrations can be
calculated from the curves. For example, if 3 mL of RM 8506 is added to 23 mL of reagent G at point A in Figure 2, it can be seen that the concentration of iodine has to increase from 1.8 × 10-4 to 2.7 × 10-4 M in order to produce the same value of the diffusion current. The sensitivity of the biamperometric indicating electrode system is decreased after addition of the oil sample and this may be due to the increased viscosity of the solution. The expected error will then be (26 × 10-3 × 2.7 × 10-4 × 18.015) - (23 × 10-3 × 1.8 × 10-4 × 18.015) g ) 51.9 × 10-6 g of water. Since the weight of the oil was 2.59 g, the error will be ∼20 ppm (w/w). It should be mentioned that the pure dilution effect was not considered in the work of Margolis5 and this corresponds to 3.7 ppm (w/w) water for this iodine end point concentration (1.8 × 10-4 M). As is well known, pyridine-buffered reagents react relatively slowly as compared to the more commonly used imidazole reagents and therefore Margolis might have used a higher iodine end point concentration than normal, (2-7) × 10-5 M, which could explain the discrepancy between the volumetric method and the coulometric method. On the basis of the results shown in this paper, the reference concentration values of moisture published by NIST1,2 and Margolis5 should be seriously questioned. ACKNOWLEDGMENT The authors thank Dr. Sam Margolis for kind support of the reference materials, Mr. Dieter Strohm for support of equipment, and Dr. Michael Sharp for linguistic revision of the paper. Received for review November 11, 1999. Accepted April 20, 2000. AC9913006
Analytical Chemistry, Vol. 72, No. 14, July 15, 2000
3395
