Physio-Chemical Indoor Air Quality Analysis and CO2 Ventilation Forecasting Using Artificial Neural Networks in Boat Manufacturing

This study presents a comprehensive analysis of indoor air quality within a boat manufacturing facility, focusing on the physio-chemical parameters and forecasting of CO2 levels using artificial neural networks (ANN). The investigation involved measuring key physical, chemical, and ventilation performance factors, including total volatile organic compounds (TVOC), particulate matter (PM10, PM2.5, PM1), formaldehyde (HCHO), carbon monoxide (CO), temperature, relative humidity (RH), and air movement. The ANN model, employing a multilayer perceptron (MLP) architecture optimized with the Levenberg-Marquardt algorithm, was developed to predict CO2 concentrations based on these inputs. The results revealed that indoor activities such as sanding, cutting, painting, and adhesive application significantly elevated levels of TVOC, particulate matter, and formaldehyde, often exceeding acceptable limits. The ANN model demonstrated high predictive accuracy, with correlation coefficients (R) ranging from 0.7556 to 0.8725 during training and 0.6798 to 0.8163 during validation and mean squared error (MSE) values as low as 0.0048 ppm. The optimal model architecture was identified as 8:15:1, providing a reliable forecast of CO2 levels with an accuracy of up to 87.25%. This study underscores the importance of monitoring indoor air quality in industrial environments and highlights the potential of ANN-based models for enhancing ventilation strategies. By enabling real-time prediction of CO2 concentrations, the model offers a practical approach to maintaining healthier indoor conditions and improving worker safety. The findings suggest that such predictive tools could be effectively implemented in similar industrial settings to mitigate air quality issues and ensure compliance with health standards