Non-Cumulative, Size-Specific Calibration of Low-Cost Particulate Matter Sensors Under Simulated Construction Drilling Events

Urban construction activities are recognized as significant contributors to particulate matter (PM) emissions; however, the accurate real-time monitoring of size-resolved PM fractions presents a formidable challenge. Traditional low-cost PM sensors predominantly report cumulative concentrations, which obscures the distinct health and regulatory significance of PM1, PM2.5, and PM10. This study systematically evaluates the performance of two low-cost sensors—PMS5003 and Sniffer4D—utilizing non-cumulative measurements obtained under controlled laboratory conditions designed to simulate construction PM generated from concrete slab drilling. Sensor performance was rigorously analyzed using Pearson correlation coefficients, standard deviation, and mean percentage differences. Six correction models—linear regression, polynomial regression, Random Forest (RF), XGBoost, Artificial Neural Network (ANN), and Kalman filter—were independently developed for each PM size fraction to enhance measurement precision. Findings reveal that RF and ANN consistently provided the most accurate corrections, particularly for PM1 and PM2.5, with RF achieving a coefficient of determination (R2) > 0.89 for PM1 and R2 > 0.87 for PM2.5 at the 50 s duration. This investigation introduces a size-resolved correction framework specifically designed for construction environments, thereby advancing the capability of low-cost sensors to enable accurate particle-specific exposure assessments.