Indoor air pollution, primarily caused by volatile organic compounds (VOCs) such as acetone, ethanol, and methanol, poses significant health risks and necessitates the development of efficient, real-time monitoring solutions. Conventional gas sensors often operate at high temperatures and are limited to detecting a single gas type, which restricts their applicability for comprehensive air quality management. To address these challenges, this study presents the development and evaluation of an advanced electronic nose (E-nose) system capable of detecting acetone, ethanol, and methanol at low operating temperatures. The proposed E-nose system comprises a gas sensor array integrating three metal oxide sensors (SM10, SE10, and Sp) designed for the simultaneous detection of these VOCs. The sensor array is seamlessly combined with Internet of Things (IoT) and cloud-based platforms to enable real-time gas monitoring and analysis. Hematite nanomaterials, imprinted with ethanol and methanol at varying concentrations of 10%, 15%, and 25%, were synthesized via a hydrothermal method. Comprehensive characterization of the synthesized nanopowders was performed using scanning electron microscopy (SEM), X-ray diffraction (XRD), photoluminescence (PL), and thermogravimetric analysis (TGA). Electrical performance assessments, including response versus temperature, resistance versus gas concentration, and output voltage versus gas concentration, were conducted to evaluate sensor efficiency. To enhance the accuracy and reliability of gas detection, artificial intelligence (AI) algorithms were employed to analyze sensor data, enabling precise gas concentration monitoring and issuing alerts when predefined threshold levels are exceeded. The dataset, obtained from sensor responses to different concentrations of acetone, ethanol, and methanol, was analyzed using multiple hybrid machine learning models, including CNN-LSTM, PCA-CNN-LSTM, PCA-XGBoost, Random Forest (RF), RF-GB, RF-GB-NN, and KNN. Notably, the RF, RF-GB, and RF-GB-NN models achieved 100% accuracy in classifying both gas type and condition. Regression analysis further demonstrated that the RF-GB model achieved an ideal coefficient of determination (R2 = 1) across all three gases, with RF and RF-GB-NN yielding R2 values of 0.999. Among the models, RF-GB exhibited the lowest limit of detection (LOD), while RF presented relatively higher LOD values. The findings validate the effectiveness of the developed sensor array, combined with AI-driven analysis, in accurately classifying and quantifying gas concentrations. Threshold values were established at 500 ppm for acetone and ethanol, and 200 ppm for methanol, with monitoring capabilities tailored to these limits.
Abdelghani et al. (Thu,) studied this question.