IoT-based filter management system using reinforced ANFIS for VOCs reduction in urban industrial facilities

This study proposes a low-cost, IoT-based multi-sensor system for monitoring volatile organic compound (VOC) emissions and predicting activated carbon filter replacement in small-scale industrial settings. Sensor modules composed of low-cost VOC sensors were installed at the exhaust of adsorption towers to enable real-time monitoring. To improve measurement accuracy, a Reinforced Adaptive Neuro-Fuzzy Inference System (RANFIS) was developed for VOC concentration prediction, incorporating dynamic outlier detection and correction. Based on RANFIS outputs, a Decision Tree (DT) model estimates the breakthrough point of activated carbon filters to support timely replacement. The system was deployed in an actual automotive painting facility using eight sensor modules with three sensor types. RANFIS outperformed Deep Neural Network (DNN) and conventional ANFIS models, improving RMSE by up to 82.4%. The DT model also achieved over 80% accuracy in predicting filter replacement under different efficiency thresholds. This integrated approach enables real-time, autonomous filter maintenance using economical sensor hardware, providing a scalable solution for VOC management. The proposed system supports more efficient and sustainable operation of emission control systems in small industrial sites.