Electronic Nose Under Dynamic Temperature Modulation for Highly Selective Identification of Industrial Toxic Chemicals

The identification of volatile organic compounds (VOCs) using electronic nose (E-nose) presents significant challenges within cross-sensitivity, response time, and cross-scenario adaptability. This study develops an E-nose that integrates a semiconductor sensor array with pattern recognition techniques for the highly selective, real-time, and cross-scene adaptable identification of toxic VOCs in industry. Specifically, dynamic temperature modulation (DTM) measurements are utilized to improve sensor selectivity. The response data undergo data augmentation to expand sample size and enable real-time identification. Leveraging the augmented samples, a multitask identification model named multiscale CNN-LSTM with attention (MCLA) is developed by combining multiscale learning strategy, convolutional neural network (CNN), long short-term memory (LSTM), and attention mechanism, achieving a species classification accuracy of 99.2% and a low mean absolute error (MAE) of 3.99 for concentration. Furthermore, MCLA can serve as a pretrained model that can be fine-tuned on small sample set for mixture gas identification, thereby enabling cross-scenario adaptability in complex environments. Accordingly, the proposed E-nose can be further integrated with embedded devices, cloud computing, the Industrial Internet of Things (IoT), and artificial intelligence, with the potential for broad applications in environmental monitoring, noninvasive medical diagnosis, and industrial production safety.