Smart sensors, smart calibration: Applications in machine learning for coal dust monitoring

Nana A. Amoah, Mirza Muhammad Zaid, Xiaosong Du, Yang Wang, Guang Xu,
Smart sensors, smart calibration: Applications in machine learning for coal dust monitoring,
Green and Smart Mining Engineering,
Volume 2, Issue 3,
2025,
Pages 301-312,
ISSN 2950-5550,
https://doi.org/10.1016/j.gsme.2025.09.010.
(https://www.sciencedirect.com/science/article/pii/S2950555025000485)
Abstract: The recent resurgence of pneumoconiosis among coal miners in the United States has been linked to their exposure to excessive levels of coal dust. PDM3700 monitors are used in the mining industry to measure each miner’s coal dust exposure levels and control overexposure. However, the high cost of the PDM3700 hinders its use in measuring the exposure levels of all miners. Plantower PMS5003 low-cost particulate matter (PM) sensors can measure coal dust concentrations with high spatial resolution in real-time owing to their low cost and small size. However, these sensors require extensive calibration to ensure a high accuracy over long deployment periods. Because they have only been calibrated for mining-induced PM monitoring using linear regression models, the objective of this study was to leverage machine learning algorithms for calibration of coal-dust-monitoring sensors. Laboratory collocation tests were performed using the PDM3700 and aerodynamic particle sizer as reference monitors in a wind tunnel at a wide range of concentrations (0–3 mg/m³), temperatures (20–32°C), and relative humidities (23%–43%). The results revealed that nonlinear machine learning techniques significantly outperformed traditional linear regression models for low-cost sensor calibration. With the artificial neural network (ANN) being the strongest calibration model, Pearson’s correlation of the PMS5003 sensors reached 0.98 and 0.97, those of the Airtrek sensors reached of 0.89 and 0.91, and those of the GasLab sensors reached 0.93 and 0.92. This shows a 2%–11% improvement in model performance over the linear regression model using ANN calibration. The success of the machine learning algorithms used in this study demonstrates the feasibility of deploying low-cost PM sensors for coal dust monitoring in mines.
Keywords: Smart sensors; Smart calibration; Machine learning; Coal dust monitoring; Artificial neural network