OCR Text |
Show being bound on sorbents . Note , furthermore , that each trace consists of distinct multiple peaks , a phenomenon especially appa rent in the carbon tetrachloride case. Even the toluene trace , showing very high merged peaks, may well have originated from distinct peaks which were d is torted and coale s ced on their way from leaving the charge , through the inlet of the sample line t o the detector. Empirical quadratic models were fitted to the data , using weighted leas t squares, where the weights were equal to the number of replicates div ided by the sample variance between the replicates. Only statistically significant dependencies were retained in the model, and peak areas were e xponentiated in order to avoid negative values . It should be emphasized that these models were purely statistical and based on no mechanistic phenomena whatsoever . From these models it was possible to generate contour plots for peak area, as shown on Figure 3. The saddle, shown on the left hand bottom figure for toluene is attributed to the fact that at very high hydrocarbon loadings exceeding 10,000 ppm, not only is the response of the FID non linear but also, the formation of soot is accelerated, leading to a depletion of volatile, FID measurable hydrocarbon concentrations . Therefore , FID measurements do not represent puff magnitude for FID concentrations far exceeding 10,000 ppm, i . e . for contours far to the left of darkened line shown on Figure 3 . Contours for p.eak height , representing puff intensity , were qualitatively similar , indicating a similarity between temporal measurements of peak height and integrated measurements of peak area. This suggests that other integrated measurements , such as those of particulate matter trapped on filters during a puff , should represent trends in both puff magnitude and puff intensity . Therefore , a statistical analysis was performed on filter weights recorded for each test, 8 |