Performance of the Sievers 500 RL On-Line TOC Analyzer

Transcription

1 GE Water & Process Technologies Analytical Instruments Performance of the Sievers 500 RL On-Line TOC Analyzer imagination at work

2 Page 2 of 10 Introduction To determine the performance characteristics of the Sievers* 500 RL On-Line TOC Analyzer, a series of measurements of certified TOC standards, conductivity standards and on-line measurements on a low TOC water system were performed. The TOC measurements were performed on six analyzers from the beta build, five of which were equipped with the Super ios. * The on-line measurements were performed to estimate the limit of detection (LOD) using three base model analyzers, one with the standard ios and one with a Super ios. The conductivity measurements were performed on six different analyzers from the pilot production since the tubing in these instruments was passivated using the procedures that will be employed in production units. The conductivity linearity test was performed during the project's validation stage using two of the pre-production alpha units. The TOC testing involved running replicates of the analyzers system protocols which are listed in Table 1. All the standards used for these tests were prepared by our standards laboratory except for the high level standards of 2 and 2.5 ppm potassium hydrogen phthalate (KHP), which were prepared by one of the test participants. The TOC standards for the Model 500 RL are stored in glass vials and acidified to ph 3 with phosphoric acid to minimize absorption of atmospheric CO 2. To determine if there is any difference in the system suitability test using acidified standards, standards for the Sievers 400 TOC Analyzers, which are not acidified and stored in plastic vials, were also measured on the Sievers 500 RL. Table 1 Summary of TOC system protocols used for evaluation of TOC measurement performance of the Sievers 500 RL TOC Analyzer Standard Set Composition Replicates Reagent Water 4 Verification 500 ppb NIST Sucrose 12 for pilot production units 25 μs/cm HCl Acidified System Suitability Reagent Water 500 ppb Sucrose 500 ppb Benzoquinone Rinse Non-acidified Same (no rinse) 3 System Suits Linearity High Level Standards 750 ppb KHP The six 500 RL analyzers were connected to a small, online water system and the analyzers built-in rinse mode initiated. The analyzers were then allowed to operate on-line for two days prior to this study. The analyzers were calibrated using either the recommended singlepoint calibration at 1.5 ppm TOC using KHP, or the optional multi-point calibration which employs calibration with KHP at 0.5, 1.0 and 1.5 ppm. After calibration, a verification was performed, then analyzers returned to on-line measurements of the water system overnight. Over a two week period, the tests in Table 1 were performed on the analyzers, with a daily system suitability test and typically two other system protocols per day. Other than the daily system suitability test and the high levels tests which were performed at the end of the two-week period, the other tests were performed in random order. For the instruments with the Super ppb KHP 6 5 for S/N ppb KHP Reagent Water Calibration Blank 1000 ppb KHP 1500 ppb KHP 2 2 measurements of Blank 2500 ppb on S/N ppb KHP 2500 ppb KHP

3 Page 3 of 10 ios, the auto-restart options were turned on so that the instrument would return to on-line monitoring of the water system once the protocols were completed. The analyzer with the standard ios was not always returned to on-line monitoring of the water system overnight. To determine the limit of detection of the Sievers 500 RL, five analyzers were connected to a low TOC water system which was operated with the dissolved oxygen concentration at ~4 ppb. The analyzers and three Sievers PPT Analyzers were operated for two weeks and the measurement of the water system inspected to identify periods of stable TOC levels. Seven such periods encompassing 5 22 hours of measurements were identified and used for the LOD calculations. When the pilot production units became available, these analyzers were calibrated, then 2 3 verification protocols were run each day over the course of a week. Only the conductivity data from these measurements are included in this report. In a separate study prior to the TOC and conductivity measurements in this report, tests of the linearity of conductivity measurements using a larger water system and a standard addition device were performed. The standard addition device consists of a syringe pump and mixing system that allows on-line production of conductivity standards. By varying the concentration of the solution used to fill the syringe and the syringe flow rate, accurate standards over a range of concentrations can be produced. For the conductivity linearity tests, stock solutions of NaCl were prepared and the standard addition device used to generate conductivity standards from to 35 µs/cm. On-line measurements from two Sievers 500 RL analyzers were obtained during these tests, along with a conventional on-line conductivity sensor (Thornton T-1000). Outliers Several outliers were identified in the accuracy and precision data set. In eight of the standards sets (four system suitability tests, one linearity, one verification and two high level standards sets), unexpectedly high TOC (> 100 ppb) was measured in the reagent water blank. The average TOC from 156 measurements of reagent water was 29 ppb, suggesting that there was some form of TOC contamination in those reagent water samples that gave high TOC values. Since accuracy and precision are calculated from blank-corrected values, these samples with high reagent water TOC were excluded from the data set. Two of the non-acidified system suitability measurements gave unexpectedly low values for benzoquinone (~ 20 ppb), possibly due to mislabeling of reagent water blank vials. A Grubb's test was applied to the remaining 340 TOC standards, and seven outliers for accuracy (due to higher than expected TOC) and one outlier for poor precision were identified and removed from the data set. Figure 1 shows a histogram and box plot for accuracy measurements obtained from the six Sievers 500 RL TOC Analyzers. For the box plot, the center line is the median, the ends of the box are the second and third quartiles. The width of the box is the interquartile range. The means diamond shows the mean and 95% confidence interval of the mean. The bracket along the edge of the plot shows the most dense 50% of the data. The whiskers extend from the box to the outermost data points within ±1.5 times the interquartile distance. Possible outliers are shown outside the whiskers. TOC Accuracy and Precision For each of the TOC standards analyzed in this study (356 total measurements), the accuracy of the measurement was calculated from the blank-corrected average of the three replicate measurement as (expected-measured)/expected. The precision was calculated from the standard deviation of the three replicate measurements divided by the blankcorrected average. Figure 1. Histogram of Accuracy Measurement using six Sievers 500 RL Analyzers along with ± 5% Specification Limits

4 Page 4 of 10 Table 2 Statistics from Accuracy Measurements Statistic Value Mean -0.3% Std Dev 2.9% Std Err Mean 0.16% N 333 Skewness 0.25 Kurtosis 0.77 Statistics for the data set are shown in Table 2. The data are not normally distributed (Shapiro-Wilk W Test) and the kurtosis indicates the distribution is peaked, with more values in the center than a normal distribution. A total of 9.9% of the measurements were outside the ± 5% accuracy specification and a capability analysis indicates that in the long run, one would expect 9% of the measurements to be outside the ± 5% limits. A plot of the absolute value of accuracy versus concentration along with a least-squares regression line is shown in Figure 2. While slightly poorer accuracy was observed at the lower concentrations, the correlation is not statistically significant (p > 0.05). The regression is also strongly influenced by the large number of measurements at 500 ppb (200) versus the 16 to 34 measurements at the other concentrations A histogram and box plot for the precision of replicate measurements of the TOC standards along with the 1% specification is shown in Figure 3. A total of 1.8% of the measurements were outside the 1% precision specification, with a mean precision of 0.2% and a standard deviation of 0.24%. A plot of precision versus the concentration of the TOC standards is shown in Figure 4, with poorer precision observed at the lower concentrations (p < 0.05). Figure 5 shows box plots of the accuracy and precision of the TOC measurements by serial number of the Sievers 500 RL TOC analyzers. Serial number 26 showed the poorest accuracy with the largest variation in accuracy. A means comparison of accuracy using the Tukey-Kramer HSD test indicates that serial number 26 is significantly different from the other Sievers 500 RL used in this study. The poorest precision was observed for serial numbers 18 and 23, although only serial number 18 had a significantly different mean precision. Figure 2. Absolute value of Accuracy measurements versus concentration of TOC standard. Figure 3. Histogram of Precision Measurements using six Sievers 500 RL TOC Analyzers along with 1% specification limit Figure 4. Precision versus concentration of the TOC standards

5 Page 5 of 10 Figure 5. Box plots of Accuracy and Precision by Analyzer Figure 6. Box plots of Accuracy and Precision by Calibration Serial numbers 23, 26, and 27 were the analyzers that were calibrated using the optional multi-point calibration. As shown in Figure 6, there was a slightly poorer mean accuracy for the analyzers calibrated using multi-point versus single-point due to serial number 26. There was no difference in the mean precision between single-point versus multi-point calibration. System Suitability The system suitability test is part of the US Pharmacopoeia Method <643> for TOC measurement of Purified Water and Water for Injection. The test involves analysis of a reagent water blank, a 500 ppb solution of sucrose (viewed as a easy to oxidize compound) and a 500 ppb solution of 1,4-benzoquinone (viewed as a more difficult to oxidize compound ). To determine the suitability of a TOC analyzer for this method, a response efficiency is calculated from the ratio of the blank corrected TOC value for the benzoquinone solution divided by the blank corrected TOC of the sucrose solution. An analyzer is deemed suitable if the response efficiency is between 85% and 115%. A histogram and box plot of 54 measurements of the response efficiency on the six Sievers 500 RL s is shown in Figure 7 along with the USP specification limits. The response efficiencies range from 98% to 107% with a mean of 101% and a standard deviation of 1.8%. A capability analysis indicates that in the long run, no response efficiencies outside the USP limits would be expected using the Sievers 500 RL. The ability of the Sievers 500 RL to pass the system suitability test can also be expressed in terms of the process capability (C p ) or the Process Capability Index (C pk ). C p is

6 Page 6 of 10 simply the difference between the upper and lower specification limits divided by six times the standard deviation of the measurements. Since the C p does not not account for processes that are not centered between the limits, the process capability index is often used and is defined as the the minimum distance between the specification limits and the process mean, divided by three times the standard deviation. For a capable measurements system, values of C p and C pk should be > 1 and a value of 1.33 is often used an a minimum acceptable value. For these system suitability measurements on the Sievers 500 RL analyzers, the C p was 2.8 and the C pk was 2.5, indicating the analyzer is capable of passing the system suitability test. For the Sievers 500 RL, standards solutions are stored in glass vials and acidified to ph 3 with phosphoric acid to minimize absorption of atmospheric CO 2. Acidification may also help to stabilize the photochemically reactive 1,4-benzoquinone. While a stability test of the acidified standards was not part of this study, a comparison was made of the 500 RL standards and non-acidified standards stored in plastic vials. Figure 8 shows box plots of the response efficiency calculated from measurement of 37 acidified sucrose and benzoquinone standards and 16 non-acidified solutions. There is no significant difference in the means or in the standard deviations of the response efficiency using acidified versus non-acidified standards. TOC Linearity The linearity portion of this study consisted of the analysis of two sets of standards. The normal linearity system protocol on the Sievers 500 RL consists of a blank, a 250 ppb standard, a 500 ppb standard, and a 750 ppb standard (sucrose). From the measurement of these standards, the analyzer will calculate the square of the Pearson's Correlation Coefficient along with a limit of detection (LOD) based on the intercept of a plot of the standard deviation of the measurements versus concentration. For this study we also prepared standards at higher concentrations (1,000 ppb, 1,500 ppb, 2,000 ppb, and 2,500 ppb). Combined with the lower concentration standards, these higher concentration standards cover the upper range of the TOC analyzer. Figure 9 shows a plot of measured versus expected concentration for the 324 linearity standards analyzed on the six Sievers 500 RL analyzers. The R 2 for this combined data was Figure 10 shows the results from two individual analyzers (best and worst based on R 2 ) Figure 7. Histogram and box plot from 54 measurements of response efficiency using the Sievers 500 RL TOC Analyzer Figure 8. Box Plot of response efficiency from 37 measurements of acidified standards and 16 measurements with non-acidified standards Figure 9. Measured versus Expected TOC for Linearity Standards on Sievers 500 RL Analyzers

7 Page 7 of 10 along with least-squares regression lines for each of the five linearity sets. For S/N 21, the slopes ranged from 0.96 to 1.01 with R 2 for all five regression lines equal to 1.0, while for S/N 18, the slopes ranged from 0.93 to 0.99 with an R 2 for one regression line equal to The linearity protocol also estimates the limit of detection from the standard deviation of the three replicate measurements of each TOC standard. Figure 11 shows the histogram and box plot for the 35 LODs calculation in this study. The average LOD was 2.5 ppb, with a range from 0.12 to 6 ppb. Detection limits calculated from these TOC standards is always higher than LOD's determined from the standard deviation of on-line measurements of low TOC water. To determine if the TOC response was linear over the entire range of the Sievers 500 RL Analyzer, higher concentration standards of KHP were analyzed. Figure 12 shows a plot of measured versus expected for the complete linearity data set (315 measurements of low level standards plus 213 measurements of higher level standards). The response of the six Sievers 500 RL's was linear over the range from 250 to 2500 ppb with an R 2 of Figure 10. Linearity Data for two individual Sievers 500 RL TOC Analyzers. Regression lines for each of the 5 sets of linearity standards are shown. Figure 11. Histogram and Box plot of Limit of Detection calculated from Linearity Protocol for six Sievers 500 RL TOC Analyzers Figure 12. Plot of Measured versus Expected for 528 TOC standards on the Sievers 500 RL Analyzers.

8 Page 8 of 10 Detection Limit of the Sievers 500 RL The limit of detection for TOC measurements using the Sievers 500 RL was estimated from on-line measurements of a low TOC (< 1.5 ppb), low dissolved oxygen water system. Measurements were performed for two weeks and seven periods of stable TOC were identified. To compensate for water system variability during these periods (primarily due to operation of the reverse osmosis system), the average TOC from the five Sievers 500 RL analyzers was calculated for each measurement. This average was then subtracted from the measured TOC from each analyzer and the standard deviation of the adjusted measurements determined. The LOD is calculated as three times the standard deviation of the adjusted TOC measurements. Figure 13 shows a histogram and box plot of the calculated LOD from 20 measurements of the Sievers 500 RL base model. The average LOD was ppb with a standard deviation of 0.01 ppb. Figure 14 shows a box plot of the LOD data by serial number (along with the results from three PPT analyzers). The three 500 RL base model analyzers (with the sample inlet block) showed lower LOD's compared to the standard and Super ios instruments (S/N 24 and 25). There was no difference in the mean LOD and the standard deviation of the LOD measurements between the three base model instruments and the three PPT analyzers. On the basis of these measurements, we estimate the LOD for the Sievers 500 RL to be 0.03 ppb. Conductivity The results from the measurement of the 25 µs/cm HCl standards on the six Sievers 500 RL analyzers are shown in Figure 15. This variability plot shows the range, measurement means, and individual measurements, along with a separate standard deviation plot. Inspection of the data suggests the presence of several outliers. A Grubb's test was performed and indicated that the circled measurements in Figure 15 are outliers and these measurements were removed from the data set. Statistics for the data set are shown in Table 3, and a plot of the distribution of the remaining data is shown in Figure 16, along with limits corresponding to ± 1% accuracy. The data are not normally distributed (Shapiro-Wilk W Test), with the kurtosis indicating the distribution is peaked with a positive skew. Six of the measurements were less than and one measurement was greater than A capability analysis Figure 13. Histogram and box plot of Limit of Detection (3* standard deviation) from adjusted TOC measurements of a low TOC, low-dissolved oxygen water system. Figure 14. Box Plot of Limit of Detection Data by serial number. There is no difference in the mean LOD between the PPT's and the Sievers 500's except for S/N 24. Table 3 Statistics from Accuracy Measurements Statistic Value Mean Std Dev Std Err Mean N 207 Skewness Kurtosis 1.96

9 Page 9 of 10 Figure 15. Variability chart of measurements of 25 µs/cm conductivity standards on six Sievers 500 RL TOC Analyzers. Circled measurements are statistical outliers (Grubb's Test) and were removed from the data set before accuracy and precision calculations. Figure 16. Histogram (with normal distribution) and box plot for 207 measurements of a 25 µs/cm conductivity standard with specification limits at ± 1% Accuracy Figure 17. Histogram and box plot of precision of conductivity measurements (3 replicates per measurement, 207 measurements with upper specification limit at 0.25% using the ± 1% conductivity accuracy indicates that one could expect 3% of measurements of this 25 µs/cm standard would be outside the limits (1.97% below, 1.04% above). The precision was calculated from the mean and standard deviation of the three measurements, and plot of the distribution of the precision is shown in Figure 17 along with a 0.25% precision specification. The mean precision was 0.04% with a range from 0% to 0.1%. A capability analysis from these data indicates that the precision of replicate measurements of this 25 µs/cm standard should always be within this specification. The results from conductivity measurements of deionized water and NaCl conductivity standards from 0.1 to 35 µs/cm are shown in Figure 18. The response of the two 500 RL's was linear across this conductivity range.

10 Page 10 of 10 Plots of the conductivity measurements' accuracy and precision from this data set, along with the results from an in-line conductivity sensor, are shown in Figure 19. Both the accuracy and precision were independent of the concentration of the NaCl standards (R 2 = p=0.11 for S/N 14, R 2 = p=0.43 for S/N 15). The analyzer shows excellent performance for the online measurement of conductivity, with an accuracy of ±1%, a precision for replicate conductivity measurements of < 0.25% and a linear response from µs/cm to 35 µs/cm. Conclusions The results from these tests demonstrate that the Sievers 500 RL TOC Analyzer is capable of measuring total organic carbon in water with an accuracy of ± 5%, with a precision for replicate measurements of < 1% and a linear response over the concentration range from ppb. The analyzers also demonstrated excellent capability for the USP system suitability test. On-line measurements of low-toc, deionized water indicate the detection limit of the analyzers is 0.03 ppb. Figure 18. Linearity of Sievers 500 RL Conductivity Measurements Figure 19. Accuracy and precision for two Sievers 500 RL's and an in-line conductivity sensor over a range of conductivity concentrations (x = Thornton, + = 500 RL S/N 14, = 500 RL S/N 15) * Trademark of General Electric Company; may be registered in one or more countries. For more information, visit Find a sales partner near you through the Contact Us Section. USA GE Analytical Instruments 6060 Spine Road Boulder, CO USA T T F geai@ge.com , General Electric Company. All rights reserved. Europe Unit 3 Mercury Way Urmston, Manchester, M41 7LY United Kingdom T +44 (0) F +44 (0) generaluk.instruments@ge.com Rev A MC06-019