Airborne particle classification with a combination of chemical composition and shape index utilizing an adaptive resonance artificial neural network

A transfer learning approach for improved classification of carbon nanomaterials from TEM images

The extensive use of carbon nanomaterials such as carbon nanotubes/nanofibers (CNTs/CNFs) in industrial settings has raised concerns over the potential health risks associated with occupational exposure to these materials. These exposures are commonly in the form of CNT/CNF-containing aerosols, resulting in a need for a reliable structure classification protocol to perform meaningful exposure assessments. However, airborne carbonaceous nanomaterials are very likely to form mixtures of individual nano-sized particles and micron-sized agglomerates with complex structures and irregular shapes, making structure identification and classification extremely difficult. While manual classification from transmission electron microscopy (TEM) images is widely used, it is time-consuming due to the lack of automation tools for structure identification. In the present study, we applied a convolutional neural network (CNN) based machine learning and computer vision method to recognize and classify airborne CNT/CNF particles from TEM images. We introduced a transfer learning approach to represent images by hypercolumn vectors, which were clustered via K-means and processed into a Vector of Locally Aggregated Descriptors (VLAD) representation to train a softmax classifier with the gradient boosting algorithm. This method achieved 90.9% accuracy on the classification of a 4-class dataset and 84.5% accuracy on a more complex 8-class dataset. The developed model established a framework to automatically detect and classify complex carbon nanostructures with potential applications that extend to the automated structural classification for other nanomaterials.

Single particle mass spectral signatures from vehicle exhaust particles and the source apportionment of on-line PM2.5 by single particle aerosol mass spectrometry

In order to accurately apportion the many distinct types of individual particles observed, it is necessary to characterize fingerprints of individual particles emitted directly from known sources. In this study, single particle mass spectral signatures from vehicle exhaust particles in a tunnel were performed. These data were used to evaluate particle signatures in a real-world PM_2.5 apportionment study. The dominant chemical type originating from average positive and negative mass spectra for vehicle exhaust particles are EC species. Four distinct particle types describe the majority of particles emitted by vehicle exhaust particles in this tunnel. Each particle class is labeled according to the most significant chemical features in both average positive and negative mass spectral signatures, including ECOC, NaK, Metal and PAHs species. A single particle aerosol mass spectrometry (SPAMS) was also employed during the winter of 2013 in Guangzhou to determine both the size and chemical composition of individual atmospheric particles, with vacuum aerodynamic diameter (d_va) in the size range of 0.2-2μm. A total of 487,570 particles were chemically analyzed with positive and negative ion mass spectra and a large set of single particle mass spectra was collected and analyzed in order to identify the speciation. According to the typical tracer ions from different source types and classification by the ART-2a algorithm which uses source fingerprints for apportioning ambient particles, the major sources of single particles were simulated. Coal combustion, vehicle exhaust, and secondary ion were the most abundant particle sources, contributing 28.5%, 17.8%, and 18.2%, respectively. The fraction with vehicle exhaust species particles decreased slightly with particle size in the condensation mode particles.

A robust framework to predict mercury speciation in combustion flue gases

Mercury emissions from coal combustion have become a global concern as growing energy demands have increased the consumption of coal. The effective implementation of treatment technologies requires knowledge of mercury speciation in the flue gas, namely concentrations of elemental, oxidized and particulate mercury at the exit of the boiler. A model that can accurately predict mercury species in flue gas would be very useful in that context. Here, a Bayesian regularized artificial neural network (BRANN) that uses five coal properties and combustion temperature was developed to predict mercury speciation in flue gases before treatment technology implementation. The results of the model show that up to 97 percent of the variation in mercury species concentration is captured through the use of BRANNs. The BRANN model was used to conduct a parametric sensitivity which revealed that the coal chlorine content and coal calorific value were the most sensitive parameters, followed by the combustion temperature. The coal sulfur content was the least important parameter. The results demonstrate the applicability of BRANNs for predicting mercury concentration and speciation in combustion flue gas and provide a more efficient and effective technique when compared to other advanced non-mechanistic modeling strategies.

Black soiling of an architectural limestone during two-year term exposure to urban air in the city of Granada (S Spain)

A two-year term aging test was carried out on a building limestone under different urban conditions in the city of Granada (Southern Spain) to assess its Cultural Heritage sustainability. For this purpose stone tablets were placed vertically at four sites with contrasting local pollution micro-environments and exposure conditions (rain-sheltered and unsheltered). The back (rain-sheltered) and the front (rain-unsheltered) faces of the stone tablets were studied for each site. The soiling process (surface blackening) was monitored through lightness (ΔL*) and chroma changes (ΔC*). Additionally atmospheric particles deposited on the stone surfaces and on PM10 filters during the exposure time were studied through a multianalytical approach including scanning electron microscopy (SEM-EDX), transmission electron microscopy (TEM) and micro-Raman spectroscopy. The identified atmospheric particles (responsible for stone soiling) were mainly soot and soil dust particles; also fly ash and aged salt particles were found. The soiling process was related to surface texture, exposure conditions and proximity to dense traffic streets. On the front faces of all stones, black soiling and surface roughness promoted by differential erosion between micritic and sparitic calcite were noticed. Moreover, it was found that surface roughness enhanced a feedback process that triggers further black soiling. The calculated effective area coverage (EAC) by light absorbing dust ranged from 10.2 to 20.4%, exceeding by far the established value of 2% EAC (limit perceptible to the human eye). Soiling coefficients (SC) were estimated based on square-root and bounded exponential fittings. Estimated black carbon (BC) concentration resulted in relatively similar SC for all studied sites and thus predicts the soiling process better than using particulate matter (PM10) concentration.

Heterogeneity of coarse particles in an urban area

The variation in composition and concentration of coarse particles in Rochester, a medium-sized city in western New York, was studied using UNC passive samplers and computer-controlled scanning electron microscopy (CCSEM). The samplers were deployed in a 5 × 5 grid (2 km × 2 km per grid cell) for 2-3 week periods in two seasons (September 2008 and May 2009) at 25 different sites across Rochester. CCSEM analysis yielded size and elemental composition for individual particles and analyzed more than 800 coarse particles per sample. Based on the composition as reflected in the fluoresced X-ray spectrum, the particles were grouped into classes with similar chemical compositions using an adaptive resonance theory (ART) network. The mass fractions of particles in the identified classes were then used to assess the homogeneity of composition and concentration across the measurement domain. These results illustrate how particle sampling using the UNC passive sampler coupled with CCSEM/ART can be used to determine the concentration and source of the coarse particulate matter at multiple sites. The particle compositions were dominated by elements suggesting that the major particle sources are road dust and biological particles. Considerable heterogeneity in both composition and concentration were observed between adjacent sites as indicated by cofficient of divergence analyses.

Study of chemical composition and morphology of airborne particles in Chandigarh, India using EDXRF and SEM techniques

The elemental composition and morphology of aerosols, collected from March 95 to February 96 and March 96 to August 96 respectively in the city of Chandigarh, India is determined using Energy Dispersive X-ray fluorescence and scanning electron microscopic techniques. The elemental concentration levels are found to be higher by a factor of 2-7 in the spring season as compared to the rainy season. The concentration of spherical and non-spherical (i.e. elongated) aerosols is more in the spring season and is reduced drastically in the rainy season due to the prominent wash out effect of rains. More accurate particle classification and source identification is obtained when based on combination of chemical composition and particle morphology. Possible sources identified from this analysis are soil dust, Industrial activity, Agricultural and Garbage burning, Maritime aerosols and Automobile exhaust.

Receptor modeling application framework for particle source apportionment

Receptor models infer contributions from particulate matter (PM) source types using multivariate measurements of particle chemical and physical properties. Receptor models complement source models that estimate concentrations from emissions inventories and transport meteorology. Enrichment factor, chemical mass balance, multiple linear regression, eigenvector. edge detection, neural network, aerosol evolution, and aerosol equilibrium models have all been used to solve particulate air quality problems, and more than 500 citations of their theory and application document these uses. While elements, ions, and carbons were often used to apportion TSP, PM10, and PM2.5 among many source types, many of these components have been reduced in source emissions such that more complex measurements of carbon fractions, specific organic compounds, single particle characteristics, and isotopic abundances now need to be measured in source and receptor samples. Compliance monitoring networks are not usually designed to obtain data for the observables, locations, and time periods that allow receptor models to be applied. Measurements from existing networks can be used to form conceptual models that allow the needed monitoring network to be optimized. The framework for using receptor models to solve air quality problems consists of: (1) formulating a conceptual model; (2) identifying potential sources; (3) characterizing source emissions; (4) obtaining and analyzing ambient PM samples for major components and source markers; (5) confirming source types with multivariate receptor models; (6) quantifying source contributions with the chemical mass balance; (7) estimating profile changes and the limiting precursor gases for secondary aerosols; and (8) reconciling receptor modeling results with source models, emissions inventories, and receptor data analyses.

Automated analysis of submicron particles by computer-controlled scanning electron microscopy

Automated analysis of submicron particles by computer-controlled scanning electron microscopy is generally possible. The minimum diameter of the detectable particles is dependent on the mean atomic number of the particles and the operating parameters of the scanning microscope. The main limitation with regard to particle size is set by the quality of the particle detection system, which generally is the backscatter electron detector. The accuracy of the results of the x-ray analyses is very often strongly affected by specimen damage, omnipresent especially for environmental particles even at low electron energies and probe currents. With the exception for light elements, the detection limit is approximately 1 wt%. Device-related limitations to automated analysis may be specimen drift and an unreliable autofocus function.

Classification of malignant and benign masses based on hybrid ART2LDA approach

A new type of classifier combining an unsupervised and a supervised model was designed and applied to classification of malignant and benign masses on mammograms. The unsupervised model was based on an adaptive resonance theory (ART2) network which clustered the masses into a number of separate classes. The classes were divided into two types: one containing only malignant masses and the other containing a mix of malignant and benign masses. The masses from the malignant classes were classified by ART2. The masses from the mixed classes were input to a supervised linear discriminant classifier (LDA). In this way, some malignant masses were separated and classified by ART2 and the less distinguishable benign and malignant masses were classified by LDA. For the evaluation of classifier performance, 348 regions of interest (ROI's) containing biopsy proven masses (169 benign and 179 malignant) were used. Ten different partitions of training and test groups were randomly generated using an average of 73% of ROI's for training and 27% for testing. Classifier design, including feature selection and weight optimization, was performed with the training group. The test group was kept independent of the training group. The performance of the hybrid classifier was compared to that of an LDA classifier alone and a backpropagation neural network (BPN). Receiver operating characteristics (ROC) analysis was used to evaluate the accuracy of the classifiers. The average area under the ROC curve (A(z)) for the hybrid classifier was 0.81 as compared to 0.78 for the LDA and 0.80 for the BPN. The partial areas above a true positive fraction of 0.9 were 0.34, 0.27 and 0.31 for the hybrid, the LDA and the BPN classifier, respectively. These results indicate that the hybrid classifier is a promising approach for improving the accuracy of classification in CAD applications.