a b s t r a c tTwo gasoline turbocharged direct injection (GTDI) and two diesel soot-in-oil samples were compared with one flame-generated soot sample. High resolution transmission electron microscopy imaging was employed for the initial qualitative assessment of the soot morphology. Carbon black and diesel soot both exhibit core-shell structures, comprising an amorphous core surrounded by graphene layers; only diesel soot has particles with multiple cores. In addition to such particles, GTDI soot also exhibits entirely amorphous structures, of which some contain crystalline particles only a few nanometers in diameter. Subsequent quantification of the nanostructure by fringe analysis indicates differences between the samples in terms of length, tortuosity, and separation of the graphitic fringes. The shortest fringes are exhibited by the GTDI samples, whilst the diesel soot and carbon black fringes are 9.7% and 15.1% longer, respectively. Fringe tortuosity is similar across the internal combustion engine samples, but lower for the carbon black sample. In contrast, fringe separation varies continuously among the samples. Raman spectroscopy further confirms the observed differences. The GTDI soot samples contain the highest fraction of amorphous carbon and defective graphitic structures, followed by diesel soot and carbon black respectively. The A D1 :A G ratios correlate linearly with both the fringe length and fringe separation.

No abstract

F ormation of soot is a known phenomenon for diesel engines, however, only recently emerged for gasoline engines with the introduction of direct injection systems. Soot-in-oil samples from a three-cylinder turbocharged gasoline direct injection (GDI) engine have been analysed. The samples were collected from the oil sump after periods of use in predominantly urban driving conditions with start-stop mode activated. Thermogravimetric analysis (TGA) was performed to measure the soot content in the drained oils. Soot deposition rates were similar to previously reported rates for diesel engines, i.e. 1 wt% per 15,000 km, thus indicating a similar importance. Morphology was assessed by transmission electron microscopy (TEM). Images showed fractal agglomerates comprising multiple primary particles with characteristic core-shell nanostructure. Furthermore, large amorphous structures were observed. Primary particle sizes ranged from 12 to 55 nm, with a mean diameter of 30 nm and mode at 31 nm. Particle agglomerates were measured by nanoparticle tracking analysis (NTA). The agglomerates were found to range between 42 and 475 nm, with a mean size of 132 nm and mode at 100 nm. The distribution was shifted towards larger sizes with a minor concentration of very large agglomerates observed around 382 nm. While deposition rate and agglomerate morphology were similar to diesel engines, distinctive amorphous carbon and smaller particles were observed. Hence, existing knowledge for diesel applications might not be directly transferrable.

No abstract

A pre-trained convolution neural network based on residual error functions (ResNet) was applied to the classification of soot and non-soot carbon nanoparticles in TEM images. Two depths of ResNet, one 18 layers deep and the other 50 layers deep, were trained using training-validation sets of increasing size (containing 100, 400 and 1400 images) and were assessed using an independent test set of 200 images. Network training was optimised in terms of mini-batch size, learning rate and training length. In all tests, ResNet18 and ResNet50 had statistically similar performances, though ResNet18 required only 25-35% of the training time of ResNet50. Training using the 100-, 400-and 1400-image trainingvalidation sets led to classification accuracies of 84%, 88% and 95%, respectively.ResNet18 and ResNet50 were also compared for their ability to categorise soot and non-soot nanoparticles via a fivefold cross-validation experiment using the entire set of 800 images of soot and 800 images of non-soot. Cross-validation was repeated 3 times with different training durations. For all cross-validation experiments, classification accuracy exceeded 91%, with no statistical differences between any of the network trainings. The most efficient network was ResNet18 trained for 5 epochs, which reached 91.2% classification after only 84 s of training on 1600 images. Use of ResNet for classification of 1000 images, the amount suggested for reliable characterisation of soot sample, requires <4 s, compared with >30 min for a skilled operator classifying images manually. Use of convolution neural networks for classification of soot and non-soot nanoparticles in TEM images is highly promising, particularly when manually classified data sets have already been established.