Hybrid Series of Carbon-Vacancy Electrodes for Multi Chemical Vapors Diagnosis Using a Residual Multi-Task Model

Detecting individual gases with various sensors is a well-established field in gas sensing. However, substantial challenges and opportunities remain in the simultaneous detection and classification of multiple gases. Artificial intelligence (AI) integrated gas sensor systems effectively enable multi-gas detection using specialized algorithms. Nevertheless, these algorithms are prone to overfitting owing to their high model complexity; this study proposes a sensor array that engineers carbon vacancies in graphene oxide via metal ion doping and high-temperature reduction, enabling high-sensitivity, simultaneous detection of various gases at low temperatures (20 °C). By integrating an advanced artificial intelligence framework, the acquired electrical signals are transformed, and a multi-task learning (MTL) approach is applied to achieve instantaneous identification of four gas types and four-level concentrations. The proposed MTL framework demonstrates superior performance by effectively mitigating overfitting and improving generalization through feature sharing and mutual regularization between gas type classification and concentration estimation tasks. Experimental validation on vehicle exhaust gas fault diagnosis highlights the method's effectiveness and applicability in complex conditions, achieving 98.22% accuracy and 48% faster inference compared to traditional single-task models. This study provides a basis for developing more intelligent and adaptable sensor systems capable.