The application of low-cost particulate matter sensors within the realm of occupational health represents a significant advancement in monitoring and improving workplace air quality, crucial for safeguarding employee health and enhancing workplace safety. In occupational settings, where the prevalence of particulate matter can vary significantly depending on the nature of industrial processes and environmental conditions, the deployment of these sensors enables a continuous and dynamic assessment of air quality. By leveraging the data collected from low-cost sensors, occupational health professionals can develop more effective health and safety protocols, including the optimization of ventilation systems, the scheduling of outdoor work to minimize exposure, and the implementation of personal protective equipment (PPE) programs.
The long-term performance assessment and calibration of low-cost sensors are pivotal for ensuring their efficacy and reliability in environmental and occupational health monitoring.  These processes are instrumental in addressing the inherent limitations associated with low-cost sensors, such as variability in sensor performance, potential for systematic bias, and sensitivity to environmental conditions. Long-term assessment facilitates a deeper understanding of the sensors' durability and operational stability, crucial for their application in continuous monitoring of particulate matter concentrations.  This insight is particularly valuable in occupational health settings, where ensuring the long-term reliability of monitoring equipment is essential for protecting worker health. Calibration, on the other hand, involves adjusting the sensor data based on known reference values, compensating for any identified biases, and improving the precision of measurements.  This process is vital for aligning the sensors' readings with those from regulatory-grade instruments, thereby ensuring the validity of the collected data. Moreover, the integration of advanced calibration models, such as machine learning algorithms, allows for the dynamic adjustment of sensors based on ongoing performance data.  This adaptive calibration not only accounts for the temporal variations in sensor sensitivity but also enhances the sensors' ability to accurately reflect changes in environmental pollutant levels.  Consequently, regular performance assessment and rigorous calibration protocols are indispensable for maintaining the integrity of sensor networks over time.
In this study, we evaluated the long-term performance of field monitoring data from two identical new low-cost sensors using Beta Attenuation Equipment inside the monitoring station. The objective was to assess the feasibility of leveraging these sensors for long-term environmental and occupational health monitoring by scrutinizing their accuracy, consistency, and calibration requirements in real-world conditions. Our analysis illuminated that the median, mean, and interquartile range (IQR) values of PM2.5 and PM10 concentrations recorded by the low-cost sensors invariably surpassed those observed by the standard detection station. This persistent overestimation highlights not only a broader data distribution but also underscores a potential systematic bias ingrained within the sensors' measurements over an extended timeframe.  The linearity observed between the data from low-cost sensors and the reports from the standard monitoring station further affirmed the systematic nature of this overestimation. When assessed against the EPA's short-term standards for PM2.5 and PM10, the sensors demonstrated commendable efficacy in PM2.5 monitoring, characterized by a minimal occurrence of false positives and negatives. However, a noticeable overestimation was evident in PM10 measurements, indicating a need for nuanced calibration strategies specifically for PM10 sensors.  The comparative analysis of both sensors revealed a commendable consistency in long-term field monitoring, as evidenced by an R² value exceeding 0.9 and an NRMSE below 5%. Stability analysis further indicated that, while generally stable, the sensor data could be occasionally influenced by varying environmental conditions. Our exploration into calibration methodologies entailed testing four machine learning models: Multiple Linear Regression (MLR), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Among these, RF and XGBoost models stood out, showcasing superior performance, with RF in particular demonstrating enhanced effectiveness on the test set. SHAP analysis for the RF model, identified as the most effective calibration model, revealed sensor readings as the most influential variable, highlighting their indispensable role in predictive modeling. Furthermore, Relative Humidity (RH) emerged as a consistently significant factor, surpassing Dew Point (DT) and Temperature (T) in importance, with higher RH levels often exerting a positive impact on model outputs.
With the implementation of appropriate calibration techniques, low-cost sensors possess the potential to complement the sparse networks of regulatory-grade instruments. By doing so, they enable dense, neighborhood-scale monitoring, thereby enriching our understanding of temporal air quality trends.  Such advancements not only have profound implications for environmental monitoring but also offer a scalable and cost-effective approach to occupational health surveillance.  This approach could facilitate the proactive identification and mitigation of air quality-related health risks in workplace settings, thereby contributing to the safeguarding of employee well-being and productivity.
 

Particulate Matter,Low-cost sensors,Air quality,Field evaluation,Calibration models