Artificial Neural Networks (ANNs) and Partial Least Squares (PLS) Regression in the Quantitative Analysis of Respirable Crystalline Silica by Fourier-Transform Infrared Spectroscopy (FTIR)

Classification of asbestos and their nonasbestiform analogues using FTIR and multivariate data analysis

This study presents a possible application of Fourier transform infrared (FTIR) spectrometry and multivariate data analysis, principal component analysis (PCA), and partial least squares-discriminant analysis (PLS-DA) for classifying asbestos and their nonasbestiform analogues. The objectives of the study are: 1) to classify six regulated asbestos types and 2) to classify between asbestos types and their nonasbestiform analogues. The respirable fraction of six regulated asbestos types and their nonasbestiform analogues were prepared in potassium bromide pellets and collected on polyvinyl chloride membrane filters for FTIR measurement. Both PCA and PLS-DA classified asbestos types and their nonasbestiform analogues on the score plots showed a very distinct clustering of samples between the serpentine (chrysotile) and amphibole groups. The PLS-DA model provided ∼95% correct prediction with a single asbestos type in the sample, although it did not provide all correct predictions for all the challenge samples due to their inherent complexity and the limited sample number. Further studies are necessary for a better prediction level in real samples and standardization of sampling and analysis procedures.

Estimation and sensitivity analysis of fouling resistance in phosphoric acid/steam heat exchanger using artificial neural networks and regression methods

One of the most frequent problem in phosphoric acid concentration plant is the heat exchanger build-up. This problem causes a reduction of the performance of this equipment and an increase of energy losses which lead to damage the apparatus. In this study, estimation of fouling resistance in a cross-flow heat exchanger was solved using a linear [Partial Least Squares (PLS)] and non linear [Artificial Neural Network (ANN)] methods. Principal Component Analysis (PCA) and Step Wise Regression (SWR) were preceded the modeling in order to determine the highest relation between operating parameters with the fouling resistance. The values of correlation coefficient (r2) and predictive ability which are equal to 0.992 and 87%, respectively showed a good prediction of the developed PLS model. In order to improve the results obtained by PLS method, an ANN model was developed. 361 experimental data points was used to design and train the network. A network containing 6 hidden neurons trained with Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm and hyperbolic tangent sigmoid transfer function for the hidden and output layers was selected to be the optimal configuration. The Garson's equation was applied to determine the sensitivity of input parameters on fouling resistance based on ANN results. Results indicated that acid inlet and outlet temperatures were the high relative important parameters on fouling resistance with importance equal to 56% and 15.4%, respectively.

Near-Infrared Spectroscopy Coupled with Chemometrics and Artificial Neural Network Modeling for Prediction of Emulsion Droplet Diameters

There is increased interest in the food industry for emulsions as delivery systems to preserve the stability of sensitive biocompounds with the aim of improving their bioavailability, solubility, and stability; maintaining their texture; and controlling their release. Emulsification in continuously operated microscale devices enables the production of emulsions of controllable droplet sizes and reduces the amount of emulsifier and time consumption, while NIR, as a nondestructive, noninvasive, fast, and efficient technique, represents an interesting aspect for emulsion investigation. The aim of this work was to predict the average Feret droplet diameter of oil-in-water and oil-in-aqueous mint extract emulsions prepared in a continuously operated microfluidic device with different emulsifiers (PEG 1500, PEG 6000, and PEG 20,000) based on the combination of near-infrared (NIR) spectra with chemometrics (principal component analysis (PCA) and partial least-squares (PLS) regression) and artificial neural network (ANN) modeling. PCA score plots for average preprocessed NIR spectra show the specific grouping of the samples into three groups according to the emulsifier used, while the PCA analysis of the emulsion samples with different emulsifiers showed the specific grouping of the samples based on the amount of emulsifier used. The developed PLS models had higher R² values for oil-in-water emulsions, ranging from 0.6863 to 0.9692 for calibration, 0.5617 to 0.8740 for validation, and 0.4618 to 0.8692 for prediction, than oil-in-aqueous mint extract emulsions, with R² values that were in range of 0.8109-0.8934 for calibration, 0.5017-0.6620, for validation and 0.5587-0.7234 for prediction. Better results were obtained for the developed nonlinear ANN models, which showed R² values in the range of 0.9428-0.9917 for training, 0.8515-0.9294 for testing, and 0.7377-0.8533 for the validation of oil-in-water emulsions, while for oil-in-aqueous mint extract emulsions R² values were higher, in the range of 0.9516-0.9996 for training, 0.9311-0.9994 for testing, and 0.8113-0.9995 for validation.

Monitoring Worker Exposure to Respirable Crystalline Silica: Application for Data-driven Predictive Modeling for End-of-Shift Exposure Assessment

In the ever-expanding complexities of the modern-day mining workplace, the continual monitoring of a safe and healthy work environment is a growing challenge. One specific workplace exposure concern is the inhalation of dust containing respirable crystalline silica (RCS) which can lead to silicosis, a potentially fatal lung disease. This is a recognized and regulated health hazard, commonly found in mining. The current methodologies to monitor this type of exposure involve distributed sample collection followed by costly and relatively lengthy follow-up laboratory analysis. To address this concern, we have investigated a data-driven predictive modeling pipeline to predict the amount of silica deposition quickly and accurately on a filter within minutes of sample collection completion. This field-based silica monitoring technique involves the use of small, and easily deployable, Fourier transform infrared (FTIR) spectrometers used for data collection followed by multivariate regression methodologies including Principal Component Analysis (PCA) and Partial Least Squares (PLS). Given the complex nature of respirable dust mixtures, there is an increasing need to account for multiple variables quickly and efficiently during analysis. This analysis consists of several quality control steps including data normalization, PCA and PLS outlier detection, as well as applying correction factors based on the sampler and cassette used for sample collection. While outside the scope of this article to test, these quality control steps will allow for the acceptance of data from many different FTIR instruments and sampling types, thus increasing the overall useability of this method. Additionally, any sample analyzed through the model and validated using a secondary method can be incorporated into the training dataset creating an ever-growing, more robust predictive model. Multivariant predictive modeling has far-reaching implications given its speed, cost, and scalability compared to conventional approaches. This contribution presents the application of PCA and PLS as part of a computational pipeline approach to predict the amount of a deposited mineral of interest using FTIR data. For this specific application, we have developed the model to analyze RCS, although this process can be implemented in the analysis of any IR-active mineral, and this pipeline applied to any FTIR data.

Comparison of the Analysis of Respirable Crystalline Silica in Workplace Air by Direct-on-Filter Methods using X-ray Diffraction and Fourier Transform Infrared Spectroscopy

A comparison of the analysis of respirable crystalline silica direct-on-filter methods using X-ray diffraction (XRD) and Fourier transform infrared (FT-IR) spectroscopy was undertaken using 253 real workplace air samples from road construction and tunnelling, coal mining, and kitchen benchtop manufacturing in Australia. Using pure α-quartz standards, XRD and FT-IR direct-on-filter analyses produced identical test results, however, the real workplace samples showed a clear discrepancy between FT-IR and XRD results with on average a 9% positive bias of the FT-IR results. The cause of the positive bias was due to matrix interferences which was confirmed by using synthetic mixture air samples. Approximately a third of the data by direct-on-filter method using FT-IR was assessed to be invalid based on the peak height ratio criterion due to excessive interferences and weight overload limitations. The XRD method showed better results due to less interference from the common matrices. XRD could handle up to twice the sample loading and at higher loadings up to 7 mg when a correction was applied. It was also able to achieve a lower limit of detection of 2 µg filter⁻¹ when a slower scan condition was utilized.

Complexity of Respirable Dust Found in Mining Operations as Characterized by X-ray Diffraction and FTIR Analysis

The mineralogical complexity of mine dust complicates exposure monitoring methods for occupational, respirable hazards. Improved understanding of the variability in respirable dust characteristics, e.g., mineral phase occurrence and composition, is required to advance on-site monitoring techniques that can be applied across diverse mining sectors. Principal components analysis (PCA) models were applied separately to XRD and FTIR datasets collected on 130 respirable dust samples from seven mining commodities to explore similarities and differences among the samples. Findings from both PCA models classified limestone, iron, and granite mine samples via their analytical responses. However, the results also cautioned that respirable samples from these commodities may not always fit patterns observed within the model. For example, one unique sample collected in a limestone mine contained no carbonate minerals. Future predictive quantification models should account for unique samples. Differences between gold and copper mine dust samples were difficult to observe. Further investigation suggested that the key to their differentiation by FTIR may lie in the characterization of clays. The results presented in this study provide foundational information for guiding the development of quantification models for respirable mineral hazards in the mining industry.