Applying Finite Mixture Models to Quantify Respirable Dust Mass in Coal and Metal-Nonmetal Mines Using Fourier...

Area of Science:

• Occupational Health and Safety
• Analytical Chemistry
• Environmental Science

Background:

• Respirable dust mass is a major occupational hazard in mining.
• Complex and heterogeneous mine dust matrices challenge traditional Fourier transform infrared (FT-IR) spectrometry calibration.
• Direct-on-filter analysis with partial least squares (PLS) is a rapid method but struggles with heterogeneity.

Purpose of the Study:

• To evaluate Mixture of Experts (MoE) finite mixture models as a novel approach for direct-on-filter FT-IR analysis of respirable dust mass.
• To compare the accuracy of MoE models against PLS methods in heterogeneous mining environments.
• To investigate the impact of model complexity and variable types on MoE performance.

Main Methods:

• Three MoE models of varying complexity were developed and applied to 243 field samples from diverse mine types.
• Spectroscopic data from direct-on-filter FT-IR analysis was used.
• MoE models incorporated cluster discovery, regression, and outlier identification, with some using categorical 'gate' variables like mine type.

Main Results:

• All tested MoE models significantly outperformed PLS in accuracy for determining respirable dust mass (p < 0.05).
• MoE models demonstrated improved accuracy across different mine types when not overfitted.
• The effectiveness of MoE was attributed to its ability to identify outliers and potentially its use of cluster modeling.

Conclusions:

• MoE finite mixture models offer a capable and novel solution for direct-on-filter quantitative analysis of heterogeneous dust samples.
• MoE methods provide a significant advancement over PLS for assessing occupational dust exposure in complex mining environments.
• This approach enhances the accuracy and reliability of health hazard monitoring for the mining workforce.