To ensure the validity of the background measurements, the researchers followed a well-designed analytical protocol that included:
The performance of a statistically valid number of injections, typically three to four injections per test mixture.
Testing of the instrument with the same test mixture twice per day to monitor the instrument drift.
Calculation of the mean response of several injections from statistically valid injection data.
Relative standard deviation was important to the test results ," the tighter the results, the greater the accuracy of the background measurement on the ethylene. Both detectors were known to be linear in response characteristics. The data were not force-fitted to other equations to maximize the coefficient of variation.
The CO tests
The PDID data were evaluated. The regression analysis showed the data had a 0.9997 correlation coefficient and a non-zero intercept. Because the higher concentration point has a tendency to skew the data, the data were re-evaluated with only the lower part of the curve. However, the data on the lower part of the curve were not as precise as those on the higher part of the curve. The correlation coefficient was only 0.9985 with a zero intercept of -115 ppb.
Evaluating CO data
To test their hypothesis that the ethylene used to make the mixture contained 115 ppb of CO, the researchers increased the concentration of the samples by 115 ppb and re-plotted the data. The curve showed the correlation coefficient did not change, and the zero intercept was 0.06 ppb.
Test One with the methanizer FID system showed a curve with a correlation coefficient similar to the PDID, but the intercept was +29 ppb. This inferred that the system had no response at 29 ppb.
In Test Two with the FID, the correlation coefficient was a little better (0.9996) than both Test One and the PDID, but the intercept was -18 ppb.
That the PDID shows a negative intercept infers the existence of a 115-ppb CO background in the ethylene used to make the test mixtures. In the case of the methanizer FID Test One, the results indicate the signal was suppressed by some 29 ppb. This typically is seen in cases in which the support gases have some level equal to or greater than the sample. In the methanizer FID Test Two, the levels of both CO and CO2 in both He and H2 were reduced to sub-ppb levels. In this case, the system shows the CO background in the ethylene was 18 ppb.
The PDID provided relatively good results all the way down to the 50-ppb level. The methanizer FID still provided more precise measurements over a wider range of concentration ," except for one excursion at 125 ppb. This detector still provided precise information.
Table 1 shows the CO test results.
The CO2 tests
The PDID regression analysis on the bottom part of the curve showed a correlation coefficient of 0.998 and a zero intercept of -178 ppb. Methanizer FID Test One showed a correlation coefficient of 0.9994 with an intercept of -195 ppb.
Test Two showed a correlation coefficient of 0.995 with an intercept of -181 ppb.
Unlike the CO testing, the CO2 tests showed little deviation between the PDID and the methanizer FID. The impact of further purification is not unexpected because the CO2 in the H2 and He is very low, typically in the ~10-ppb range.
The PDID provided more precise measurements for CO2 than for the CO measurements. The precision of both systems, however, is still good enough to enable confidence in the results obtained.
Table 2 shows the CO2 test results.
Overall test results
The PDID data are of high enough quality that the researchers could not ignore the 115-ppb background in the base ethylene used to make the mixtures. The fact that the methanizer FID is susceptible to support-gas bias is troublesome because much of the petrochemical industry uses this detector for making CO/CO2 measurements.
Are other variables associated with the methanizer FID suppressing the CO response? This possibility requires further investigation.
Based on the results of this rigorous testing process, the company presented the following conclusions to the ASTM D2 committee:
The present methods used to measure both CO (methanizer FID) and C2 H2 (FID) at <500 ppb are inadequate.
The use of a common high-accuracy standard significantly improves both the precision and accuracy of the GC methods.
The use of PDID and He free from CO/CO2 eliminates bias.
The use of support gases (He and H2) free from CO/CO2 on a methanizer FID reduces the bias, but does not totally eliminate it.
The agreement between the two methods is excellent when both systems use support gases of the same quality. The CO2 work also validates the quality of the test mixture generated for this study because the minor components were introduced from the same blending mixtures.
Researchers must perform additional work to ascertain the reason for the bias between the two analytical methods. The extensive use of the methanizer FID in both petrochemical applications and the standards manufacturing process could continue to perpetuate the bias in the measurements.
The use of support gases (He and H2) free from both CO and CO2 reduces the bias that is observed on the methanizer FID. High-quality standards at the ppb level can be manufactured effectively in an ethylene matrix. The process of establishing the background in the base ethylene can be biased by the analytical method used.
The manufacture of polyethylene with undetected impurities can have significant adverse financial impacts. Therefore, the use of high-purity methanizer FID support gases is a wise first step for the petrochemical industry in addressing this measurement bias issue. CP
Denyszyn is a manager with Praxair Distribution Inc., Houston., which is part of Praxair Inc., a global supplier of atmospheric, process and specialty gases, high-performance coatings, and related services and technologies. He can be reached at email@example.com.