ANN-Driven Optimization of VOC Adsorption on Activated Carbon with Thermal Breakthrough Forecasting and IoT-Based Real-Time Monitoring

Volatile Organic Compounds (VOCs) have become one of the drivers of environmental deterioration and occupational hazards, and the issue requires competent and clever adsorption methods for their elimination. This study proposes a comprehensive experimental, computational, and IoT-based system that maximizes VOCs adsorption by activated carbon. A packed bed adsorption column was constructed and equipped with two MQ-138 and DHT22 sensors, which could be directly tracked in real time using a NodeMCU-ThingSpeak dashboard. During the experiments, the efficiency of VOC removal was lower at higher inlet concentrations (92.3% to 76.1% at 100 ppm to 300 ppm, respectively) and higher at the optimized flow rate (74.5% - 89.8% at 3.0 to 1.5 L.min-1, respectively). The efficiency was lower at high relative humidity because of competitive adsorption, and higher bed temperatures (up to 45°C) slightly prolonged the breakthrough time. The model used was a 4-8-1 ANN (Artificial Neural Network) whose training was carried out using the LevenbergMarquardt algorithm, which had a high predictive accuracy (R2 =0.987, Root Mean Square Error (RMSE) =1.82), and the experimental value was close to the computed values across a range of inputs. The 3D surface mapping of the ANN model exhibited an ideal area of interaction between the VOC concentration and flow rate. In addition, all IoT delays were less than 1.5 s, and the sensor offset was less than ±5 ppm and ±0.5°C, thus confirming the readiness of the system deployment. These outcomes confirm that it is possible to implement intelligent, responsive VOC mitigation tools that are informed by machine learning and integrated with IoT to manage air quality in the industry.