Integrated 3D tensor and deep neural network approach for real-time toxic gas monitoring and device integrity assessment

This study presents an advanced multifunctional graphene-based sensing platform integrating deep neural networks (DNN), phase space reconstruction (PSR), and sophisticated feature engineering for real-time toxic gas monitoring and internal device integrity assessment. The sensor leverages a dual-sensor architecture composed of multilayer graphene and Kapton substrates, modeled through an equivalent circuit model (ECM) that captures subtle electrical impedance changes induced by gas adsorption and structural anomalies. Optical absorption spectra of CO2, NO2, and H2S in the terahertz frequency range reveal distinct molecular fingerprints, enabling precise gas discrimination and concentration quantification with response times as low as 5 s for CO2 detection. The PSR technique transforms time-series sensor data into multidimensional phase space representations fed into a 3D convolutional neural network, enhancing classification accuracy of both toxic gases and internal defects. Structural anomalies within the sensor, including internal gaps and delamination, are detected through characteristic shifts in absorption spectra, underscoring the platform’s dual capability for environmental monitoring and device fault diagnosis. Comparative analyses demonstrate enhanced capability in sensitivity, selectivity, and rapid response relative to existing sensor technologies. This integrated approach establishes a robust, scalable foundation for next-generation wearable, portable, and embedded sensing systems applicable to environmental safety, industrial emission control, and healthcare diagnostics.