Energy & Fuels 2004, 18, 1851-1854
1851
Validation of the European Union’s Reference Method for the Determination of Solvent Yellow 124 in Gas Oil and Kerosene Thomas Linsinger,*,† Ger Koomen,‡ Håkan Emteborg,† Gert Roebben,† Gerard Kramer,† and A. Lamberty† European Commission, DG-JRC, Institute for Reference Materials and Measurements, Retieseweg 111, 2440 Geel, Belgium, and Dutch Customs Laboratory, Kingsfordweg 1, 1043GN Amsterdam, The Netherlands Received July 27, 2004
The European Union’s reference method for the determination of a common fiscal marker of gas oil for heating purposes, Solvent Yellow 124 (SY124), was validated. A total of 12 different batches of samples using various commercially available gas oils and various colorants with SY124 concentrations from 0.12 to 9 mg L-1 were prepared. Various other dyes were added to check for potential interferences in the separation and detection of SY124. A total of 26 laboratories participated in the validation study. Outliers were identified using the Cochran and Hawkins test, and the resulting datasets were checked for normal distribution using normal probability plots. At a level of 6 mg L-1, the relative repeatability and reproducibility standard deviations were 0.68% and 3.8%, respectively. At 0.12 mg L-1, the repeatability and reproducibility standard deviations were 5.4% and 13.5%. A limit of detection of 0.020 mg L-1 and a limit of quantification of 0.065 mg L-1 were estimated. The method was found to be without significant bias and is therefore suitable as a reference method for the determination of SY124 in gas oil.
1. Introduction Gas oils used for transport and heating purposes are taxed differently in most European countries, resulting in price differences. To prevent the use for transportation of gas oil destined for heating, many countries perform roadside controls, checking for various colorants added to the heating oil. Increased European unification has led to increased cross-border traffic, thus making the detection of fraud more difficult. In 1995, the European Commission decided to introduce a common marker to facilitate harmonization of excise duties on mineral oils.1 In a study that lasted from 1996 to 2000,2 Solvent Yellow 124 (SY124), or N-ethyl-N-[2-(1-isobutoxyethoxy)ethyl]-4-(phenylazo)aniline, CAS Registry No. 34432-92-3, was selected as the most suitable candidate for such a common marker. It was already used without problems in several European countries as a marker, although SY124 is partially UV unstable and can be laundered. On the basis of this study, SY124 was introduced as a common marker in the European Union from August 1, 2002,3 with a lower marking level of at least 6 mg L-1. In 2003,4 the upper marking level was set to 9 mg L-1. †
Institute for Reference Materials and Measurements. Dutch Customs Laboratory. (1) European Council Directive 95/60/EC. (2) Umlauf, G., private communication (Hiller, B.; Umlauf, G. Selection of a fiscal marker for rebated gas oil and kerosene according to Council Directive 95/60/EC, JRC Ispra Internal Report No. 1219696-09AICA ISP B, 28/01/2000). (3) European Commission Decision 2001/574/EC. (4) European Commission Decision 2003/900/EC. ‡
Following this decision, the Excise Duty Committee decided that the establishment of a community reference method was desirable that should have a limit of quantification of a maximum of 0.12 mg L-1 (2% of the initial concentration or 50 times lower than the marking limit). In this paper, we describe the validation and final establishment of such a reference method for the determination of SY124 in gas oil. 2. Experimental Section 2.1. Samples. A total of 12 batches representing different combinations of gas oils and additives were prepared at the Institute for Reference Materials and Measurements, European Commission’s Directorate General JRC, Geel, Belgium. Pure SY124 was obtained from John Hogg Technical Solutions Ltd., Manchester, U.K. The purity was checked using a highperformance liquid chromatograph with a diode array detector (HPLC-DAD) and differential scanning calorimeter (DSC) and found to be above 99%. Dyeguard Red 161 (purity 50%), Dyeguard Red C (purity 60%), Dyeguard Blue 79R (purity 48%), Solvent Red 24 (purity >90%), Quinizarin (purity 100%), Dyeguard Green DL (33) (purity 74%), and Coumarine (purity 100%) were obtained from John Hogg Technical Solutions Ltd. Nine different gas oils and kerosene representing a wide variety of compositions (distilling at 250 °C: from 30 to >65%) were obtained from commercial sources. Only the gas oil used to produce batch 12 already contained SY124. For all others, the SY124 blank value was confirmed by HPLC-DAD. The blank gas oils were subsequently spiked gravimetrically, mixed thoroughly, filled into 3-mL Pyrex ampules, flushed with an Ar/He mixture, and sealed on an automatic ampuling machine. A total of 2 mL of sample was filled in each ampule. Tightness of the sealing was confirmed with a He-leak detector.
10.1021/ef049820d CCC: $27.50 © 2004 American Chemical Society Published on Web 09/21/2004
1852
Energy & Fuels, Vol. 18, No. 6, 2004
Linsinger et al.
Table 1. Gravimetric Addition of Dyes to the Various Samples other dyes added batch no.
gas oil
1 2 3 4 5 6
A A A B C D
7
E
8 9 10
F F G
11 12
H I
substance none none none Solvent Red C Solvent Red C Dyeguard Blue 79R Solvent Red 24 Quinizarin Solvent Red C Solvent Red 161 Dyeguard Green DL(33) Coumarine none
concn [mg L-1]
0.223 4.78 4.98 10.09 1.97 0.199 15.0 4.91 2.04
SY124 added concn [mg L-1]
Ua [mg L-1]
0 6.01 0.12 0.279 5.96 8.97
0.016 0.003 0.003 0.023 0.036
4.77
0.064
0 6.02 7.19
0.000 0.126 0.093
6.02 unknown
check
a U is the expanded uncertainty resulting from the error propagation of weighing on several balances and preparation of stock solutions from the pure compounds (k ) 2) for a 95% confidence limit comprising uncertainties from preparation and homogeneity.
The homogeneity of each batch was checked by analyzing 10 units of each batch and determining the SY124 content in duplicate. The individual ampules were selected using a random stratified sampling scheme, and the measurements were done in one series, ensuring repeatability conditions. The results were plotted against the filling sequence to check for any significant trends, and standard deviations within and between units were calculated using analysis of variance (ANOVA). Furthermore, u*bb, the maximum heterogeneity that could be hidden by method repeatability, was calculated as described by Linsinger et al.5 A small trend of the SY124 content over the filling sequence was found for three batches. However, even in the worst case, the maximum difference was less than 1.2% and was thus negligible compared to the method variation. The stability of the samples was assessed a priori using test batches of 0.12 and 6 mg L-1 stored at room temperature over a period of 8 months. In a previous report,2 no changes were found for SY124 in gas oil at a temperature of 40 °C, thus making dispatch at ambient temperature possible. Additional confirmation of the stability of the sample was derived from samples analyzed after the closing date for the interlaboratory study. The results obtained have confirmed the stability of the material over the time of the study. The final composition of the 12 batches is shown in Table 1. A solid calibration standard of SY124 (92% pure) was kindly provided by BASF, Ludwigshafen, Germany. This calibration standard was completely independent from the samples of the validation study and was dispatched to the participants for method calibration. 2.2. Method. Samples are filtered before use if necessary. Calibration solutions of 1, 5, and 10 mg L-1 SY124 in xylene are prepared. The samples are analyzed by HPLC (normal phase silica, 5 µm; length, 200-250 mm; inside diameter, 3.0-5 mm) using isocratic conditions [eluent: ethyl acetate: toluene ) 2:98 (v/v)] and photometric detection in the visible (vis) range. The community reference method will be published in the Series C of the Official Journal of the European Union. At the beginning of the study, it was not clear whether the best detection wavelength would be 410 nm (absorbance maximum) or 450 nm (less spectral interferences). Participants were therefore asked to submit results for both wavelengths to allow the selection of the most appropriate wavelength. We will focus on the results at 450 nm in this paper because the study showed that this is the most appropriate wavelength. Findings made at 410 nm will only be included in some cases.
2.3. Organization of the Study and Participants. European customs laboratories and laboratories of producers of SY124 were invited to participate in the validation study. Finally, 26 laboratories from Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Portugal, Spain, Sweden, The Netherlands, and United Kingdom participated in the study. Each participant received two units of each batch, one screwcapped bottle of solid calibration standard, and a results report form. Two replicate determinations per ampule (i.e., four results per batch) were performed. Participants were asked to submit details on their chromatographic system and the correlation coefficient of the calibration function together with the final results. 2.4. Evaluation. A total of 25 of the 26 laboratories submitted results in time. One of these 25 laboratories used a different method than the common reference method and was not included in the evaluation of the reference method. Separate evaluations were performed for results obtained by HPLC separation and vis detection at 410 and 450 nm following ISO 5725 “Accuracy (trueness and precision) of measurement method and results”6 and ISO 4259 “Petroleum productssDetermination and application of precision data in relation to methods of test”.7 The results for each batch were screened for outliers of variance using the Cochran test and for outlying mean values using the Hawkins test. Normal probability plots were prepared to check the distribution of the results for deviation of the normal distribution. Finally, all datasets of all laboratories were scrutinized using Mandel’s h and k statistics for trends of between-laboratory and within-laboratory consistency, respectively. An ANOVA was performed with the retained datasets. Repeatability standard deviation (sr), between-laboratory standard deviation (sL), and reproducibility standard deviation (sR) were calculated as the standard deviation within groups, the standard deviation between groups, and the square root of the squared sum of sr and sL, respectively. Method bias was evaluated by comparing the mean of accepted laboratory means with the target value deduced from the gravimetric preparation. Using this target value with its associated small uncertainty allowed a very accurate assessment of a potential method bias.
3. Results and Discussion 3.1. Outliers. More outlying variances than outlying mean values were detected. The Cochran test resulted frequently in a high number of outliers (more than six). The number of outliers excluded on statistical grounds after these two tests was restricted to three (about 10% of the number of results) because this “snowballing” is well-known for this test.6 The general policy was rather to keep data than to discard them. If several results would be outliers and visual inspection showed no great difference between variances, all datasets were kept. A plot of all data and standard deviations and the corresponding normal probability plot for batch 2 are shown in Figure 1. Consistently high or low laboratory averages and consistently high variances can be detected using Mandel’s h and k statistics. As can be seen in Figure 2 from (5) Linsinger, T. P. J.; Pauwels, J.; van der Veen, A. M. H.; Schimmel, H.; Lamberty, A. Accredit. Qual. Assur. 2001, 6, 20-25. (6) ISO 5725 Accuracy (Trueness and Precision of Measurement Method and Results, 3rd revision; ISO: Geneva, Switzerland, 1996). (7) ISO 4259 Petroleum products, Determination and application of precision data in relation to methods of test, Committee draft TC 28 N2168 2002.
Determination of Solvent Yellow 124
Energy & Fuels, Vol. 18, No. 6, 2004 1853
Figure 1. Data of results for batch 2. The left side shows laboratory averages. The error bars correspond to (1 standard deviation. The right side shows the normal probability plot after outlier elimination (laboratory 25, Cochran outlier).
Figure 2. Mandel’s h and k statistics for the determination at 450 nm. Dashed lines show 95% confidence limits; bold lines show 99% confidence limits. Table 2. Mean of Means, Repeatability Standard Deviation (sr), Between-Laboratory Standard Deviation (sL), and Reproducibility Standard Deviation (sR) for the Various Batches for the Determination of SY124 at 410 nma target mean of value means batch [mg L-1] [mg L-1] 2 3 4 5 6 7 9 10 11 12
6.01 0.12 0.279 5.96 8.97 4.77 6.02 7.19 5.95 unknown
6.07 0.12 0.279 6.02 9.05 4.78 6.12 7.16 5.87 6.10
sr [mg L-1] ([%])
sL [mg L-1] ([%])
sR [mg L-1] ([%])
0.035 (0.58) 0.009 (7.00) 0.025 (8.82) 0.040 (0.67) 0.061 (0.68) 0.045 (0.93) 0.051 (0.83) 0.070 (0.98) 0.044 (0.76) 0.092 (1.5)
0.22 (3.5) 0.011 (8.9) 0.023 (8.2) 0.19 (3.1) 0.25 (2.7) 0.14 (2.9) 0.25 (4.1) 0.19 (2.7) 0.17 (2.9) 0.36 (5.9)
0.22 (3.6) 0.014 (11.3) 0.034 (12.0) 0.19 (3.2) 0.25 (2.8) 0.15 (3.0) 0.25 (4.1) 0.21 (2.9) 0.17 (3.0) 0.37 (6.1)
Table 3. Mean of Means, Repeatability Standard Deviation (sr), Between-Laboratory Standard Deviation (sL), and Reproducibility Standard Deviation (sR) for the Various Batches for the Determination of SY124 at 450 nma target mean of value means batch [mg L-1] [mg L-1] 2 3 4 5 6 7 9 10 11 12
6.01 0.12 0.279 5.96 8.97 4.77 6.02 7.19 6.02 unknown
6.04 0.12 0.27 5.99 9.05 4.78 6.10 7.13 5.87 6.01
sr [mg L-1] ([%])
sL [mg L-1] ([%])
sR [mg L-1] ([%])
0.041 (0.68) 0.007 (5.44) 0.014 (5.00) 0.033 (0.55) 0.064 (0.71) 0.049 (1.03) 0.079 (1.29) 0.070 (0.98) 0.061 (1.05) 0.032 (0.54)
0.23 (3.8) 0.015 (12.4) 0.014 (5.2) 0.22 (3.7) 0.27 (3.0) 0.15 (3.0) 0.27 (4.4) 0.20 (2.8) 0.18 (3.0) 0.19 (3.2)
0.23 (3.8) 0.016 (13.5) 0.020 (7.2) 0.23 (3.7) 0.28 (3.1) 0.15 (3.2) 0.28 (4.6) 0.21 (2.9) 0.19 (3.2) 0.20 (3.3)
a Batches 1 and 8 do not contain SY124, so no variability data can be calculated.
a Batches 1 and 8 do not contain SY124, so no variability data can be calculated.
the plot of Mandel’s h statistics, most laboratories reported results above and below the mean value, showing that constant laboratory bias was rare. The right side of Figure 2 shows the outliers of variance. However, no laboratory consistently reported results with exceptionally high variances. The datasets were therefore suitable for further evaluation. 3.2. Variability. Repeatability, between-laboratory standard deviation, and reproducibility calculated from the accepted datasets for 410 and 450 nm are shown in Tables 2 and 3. The repeatability standard deviation is significantly smaller (on average a factor 4) than the reproducibility standard deviation at the level of 6 mg L-1, which is normal for chemical analyses. Probably the difference between reproducibility and repeatability is mainly caused by the difficulty of precisely diluting the stan-
dards for the calibration curve. The repeatability standard deviation and the between-laboratory standard deviation are on the same order at a level of 0.12 mg L-1. This means that variation seen is largely governed by the intrinsic method repeatability. Variability was similar for the determination at 410 and 450 nm. This is remarkable considering that the signal at 410 nm is about twice as high as that at 450 nm because of the absorbance spectrum of SY124. The advantage of a higher signal was apparently lost because of more interferences: Most participants remarked that more disturbing peaks were visible at 410 nm than at 450 nm. 3.3. Trueness. Target values for all batches except batch 12 were known very precisely from the gravimetrical preparation. The trueness of the method was checked by comparing the mean of accepted laboratory
1854
Energy & Fuels, Vol. 18, No. 6, 2004
Linsinger et al.
means with the target value. Only minor, statistically insignificant differences between the mean of means and the target values were observed for most samples, as can be seen in Table 3. The only exception was batch 11, which was also the only batch for which the difference was significant on a 95% significance level. This bias is, however, small enough to be negligible for a measurement in only one laboratory. It was therefore concluded that the method did not show any significant bias. 3.4. Limit of Detection and Quantification. Limits of detection and quantification were calculated as 3 and 10 times the repeatability standard deviation of a sample with a SY124 content close to the limit of detection. The data from batch 3 were chosen for this purpose, and a limit of detection of 0.020 mg L-1 and a limit of quantification of 0.065 mg L-1 were obtained. These limits were checked for appropriateness by an investigation of the results for batches 1 and 8, which were known not to contain SY124: Only six results above zero were reported for both samples, together of which only one result was higher than 0.020 mg L-1, but a closer investigation by the laboratory showed that the retention time of the peak in question was 2% longer than that of SY124. The peak was therefore not SY124, and the finding was refuted. This investigation proved that the calculated limits are achievable and realistic and are sufficiently low to allow quantification at a concentration level of 0.12 mg L-1 in the matrices of interest. The situation was less advantageous for detection at 410 nm: One result for batches 1 and 8 was above 0.020 mg L-1, with the results ranging up to 0.75 mg L-1. 3.5. Limits for Specification Testing. Specification testing consists of checking whether the specified value (here between 6 and 9 mg L-1) is confirmed using a single analytical test result. The results for the reproducibility of the evaluated method allow one to decide to calculate the testing margins for the determination at 450 nm according to ISO 4259. For this, an average reproducibility standard deviation was calculated using the relative sR of batches 2, 5-7, and 9-12 according to the equation
sjR )
x
∑sR,i2 ) 3.52% 8
(1)
This quadratic addition was chosen because variances are additive but standard deviations are not. Upper and lower limits of the specification c were then calculated as
limit ) c ( with
0.84R x2
(2)
R ) 2x2csjR The resulting limits were 9.5 and 5.6 mg
(3) L-1.
4. Conclusions and Outlook The method was found sufficiently accurate to be used for the determination of SY124 in gas oil at a detection wavelength of 450 nm. A detection wavelength of 450 nm is preferred over one of 410 nm because of the reduced danger of interferences, which led to a higher number of false positives when using 410 nm. The limits of detection and quantification at the detection wavelength of 450 nm were low enough to guarantee also detection of dilution down to 2%, or 50 times below the target marker concentration of 6 mg L-1. No significant method bias was observed, ensuring comparability of the results from various laboratories. The method is therefore suitable as a community reference method and will be published in the C Series of the Official Journal of the European Union. Two certified reference materials containing 6.0 and 0.12 mg L-1 SY124 in gas oil are in preparation and will presumably be available from IRMM early in 2005. Acknowledgment. This work was supported by the Directorate-General for Taxation and the Customs Union (DG-TAXUD) of the European Commission. The authors thank John Hogg Technical Solutions Ltd. and BASF for providing pure SY124 and the following laboratories for participation in the validation study: Agenzia delle Dogane (Rome, Italy), BASF Aktiengesellschaft (Ludwigshafen, Germany), Belastingsdienst/ Douane Laboratorium (Amsterdam, The Netherlands), Customs Laboratory of National Customs Board (Riga, Latvia), Customs Technical Laboratory (Praha, Czech Republic), Finnish Customs Laboratory (Espoo, Finland), Force Technology (Brondby, Denmark), General Chemical State Laboratory (Athens, Greece), John Hogg Technical Solutions (Manchester, U.K.), Laboratoire des Douanes de Paris (Paris, France), Laboratoire National de Sante´ (Luxemburg, Luxembourg), Laboratorio Central de Aduanas (Madrid, Spain), Laborato´rio da Direca˜o-Geral das Alfaˆndegas (Lisboa, Portugal), Laboratorio de la Aduana (Barcelona, Spain), Laboratorium van Douane en Accijnzen (Leuven, Belgium), LGC (Teddington, U.K.), Lithuanian Customs Laboratory (Vilnius, Lithuania), National Laboratory of Forensic Science (Linko¨ping, Sweden), Orgachim (Oissel, France), State Laboratory (Dublin, Ireland), Technische Untersuchungsanstalt (Wien, Austria), Zolltechnische Pru¨fungs- und Lehranstalt (Berlin, Germany), Zolltechnische Pru¨fungs- und Lehranstalt (Frankfurt, Germany), Zolltechnische Pru¨fungs- und Lehranstalt (Hamburg, Germany), Zolltechnische Pru¨fungs- und Lehranstalt (Ko¨ln, Germany), and Zolltechnische Prufungs- und Lehranstalt (Mu¨nchen, Germany). EF049820D