Repeatability and reproducibility of TEM soot primary particle size measurements and comparison of automated methods

Transmission electron microscopy (TEM) is used widely to measure the size and structure of aerosols, however the repeatability and reproducibility of these measurements has never been investigated. In this study, the primary particle size distributions of three soot samples from a propylene/air co-flow diffusion flame are measured from TEM images by four independent operators. The overall uncertainty, accounting for repeatability and variation across samples and operators, is quantified through gage repeatability and reproducibility (Gage R & R) analysis. A randomized measurement procedure coupled with a semi-automated algorithm called center-selected edge scoring, is found to limit operator bias while also taking a fifth as long as manual measurement. Accounting for all sources of uncertainty, the 95% confidence intervals determined from a single measurement of primary particle mean and standard deviation are ± 14% and ± 33%, which is greater than is commonly reported. Since Gage R & R analysis requires more replicate measurements than is often practical, a simplified approach for obtaining confidence intervals is presented which can capture Gage R & R uncertainties while requiring only a few replicate measurements of a given sample. The Gage R & R results are used to evaluate three automated methods, two employing a Circular Hough Transform (CHT) algorithm and one utilizing Euclidian distance mapping (EDM). The CHT methods underestimate the mean primary particle diameter, while the EDM method overestimates it. Both the CHT method with Canny edge detection and the EDM method assess reasonably well the standard deviation of the size distribution.