Development of high-throughput electron tomography for 3D morphological characterisation of soot nanoparticles

In this work a methodology has been developed to permit high-throughput, high-quality morphology characterisation of soot nanoparticles in 3D, using transmission electron microscopy (TEM) and electron tomography (ET). Nanoparticle morphology plays an extremely important role in determining soots contribution to global climate change, its impacts on human health, and its relation to friction and wear in engines. Morphology characterisation is thus fundamental to enabling full understanding of soot-induced phenomena and subsequent strategies for their mitigation. Initial work employed ET for the 3D reconstruction and morphology characterisation of a flame-generated soot nanoparticle. Variability in 2D-derived measurements as a function of nanoparticle orientation was measured at up to 45%, which may be an important source of error in 2D-derived morphology measurements. Significant discrepancies of up to 36% between 2D- and 3D-derived measurements were also observed. This type of particle was particularly complex and 3-dimensional, and acted as a useful case-study for the identification of areas for possible optimisation in the 3D reconstruction process. A sample of soot-in-oil from a gasoline turbocharged direct injection (GTDI) engine was chosen as the main focus of study in this work. This is due to both the novelty of study and potential for useful research outputs, and the difficulty of study via TEM due to the presence of lubricant oil. General characteristics of particulate matter in the GTDI soot-in-oil sample were understood through TEM imaging, and revealed an abundance of localised amorphous carbon structures and patches of crystalline material, in addition to typical soot nanoparticles. Optimisation of ET was carried out by assessment of the speed and relative accuracy of a number of well-established procedures for tilt-series acquisition, tomographic reconstruction, and tomogram segmentation. It was found that the computationally efficient WBP algorithm produced data of equivalent quality to that of the theoretically more accurate SIRT algorithm, though was significantly quicker. Optimal tilt-series were acquired over ±60° ranges, in increments as large as 3°. For the final stage of optimisation, 6 nanoparticles from the GTDI soot-in-oil sample were reconstructed in 3D via ET. Linear interpolation was found to be useful for increasing speed while retaining accuracy in volume segmentation for producing the final nanoparticle models. Efforts were also made to reduce the amount of time and human involvement required for nanoparticle identification and image acquisition, resulting in the development of semi-automated process for TEM imaging. Large areas of the TEM grid are automatically imaged, and then screened for structures of interest via an automated image processing algorithm and a manual review process. Nanoparticles locations are then communicated to the electron microscope for subsequent 3D study. Using this semi-automated process over 4000 μm2 of a TEM grid was imaged, resulting in identification of 271 soot nanoparticles. Additional steps were implemented for high-throughput 2D morphology characterisation, and resulted in the measurement of 523 individual nanoparticles from the GTDI soot-in-oil sample. Analysis of this data revealed similar size of soot primary particles and aggregates compared with previous studies of soot-in-oil samples, but modest aggregate sizes in comparison to exhaust and flame-generated soots. Accuracy of ET was assessed via computational models of soot-like aggregates which acted as ‘ground truth’ (i.e. 3D morphology was known exactly). The soot-like models were subjected to the 3D-TEM procedure, and post-reconstruction morphology was compared to that of the original models. The results of our prior optimisation work were confirmed, and absolute accuracy of 3D-TEM for the soot-like models was shown to be extremely high (within 3.5% of original values for a range of parameters). From the pool of 271 nanoparticles identified using the automated TEM imaging procedure, an additional 28 nanoparticles were chosen for study in 3D via ET. These nanoparticles were chosen to express the general extremes of visible structures, e.g. particularly large and complex, or small and simple nanoparticles. This total sample size of 34 nanoparticles represents the largest 3D study of soot-in-oil to date, and one of the largest 3D studies of any soot sample to date. The 3D aspect ratio of nanoparticles revealed a general tendency towards 2-dimensionality rather than strong 3-dimensionality, and overall size of nanoparticles was generally small in comparison to aerosol soot. Direct comparison of 2D- and 3D-TEM characterisation showed deviations on the order of 20-35% for some important morphological parameters (volume, surface area, circularity, aspect-ratio), though 2D-derived radius of gyrations measurements were generally accurate. Strong variability was observed across the sample of nanoparticles, and no clear correlations were drawn between particle morphology and the accuracy of 2D-derived measurements. The 3D models were used to explored the full extent of variability in the 2D appearance due to nanoparticle orientation, with an average variability in 2D area of 59%. Several nanoparticles were observed with prominent features such as rings, cavities, and arches that were not appreciable via typical 2D-TEM study, and may be important for greater understand of soot formation and impacts of morphology.