MXene@Au-functionalized Gel-Car-PVA hydrogel electrolyte for leak-free quasi-solid-state Zn-air fuel cells with uncertainty-aware trimodal sensing

Abstract The translation of self-powered electrochemical sensors from laboratory demonstrations to regulatory-compliant field deployment is impeded by two coupled barriers: (1) liquid electrolyte leakage and evaporation cause baseline drift under fluctuating conditions, and (2) black-box machine learning models lack uncertainty quantification and transparency for compliance testing. We report a co-designed quasi-solid-state Zn-air fuel cell platform that addresses both barriers through synergistic integration of materials engineering and algorithmic innovation, using acetamiprid (ACE) as a model pollutant. A physically cross-linked Gel-Car-PVA hydrogel electrolyte eliminates leakage and evaporation, suppressing signal attenuation to below 0.20% over 30 days across -20 to 25 ℃, resolving the baseline instability that undermines field sensor credibility. Parallel to this, an uncertainty-calibrated one-dimensional convolutional neural network (1D-CNN) with Monte Carlo Dropout and SHapley Additive exPlanations (SHAP) delivers prediction intervals with quantified confidence mechanistic feature attribution, meeting regulatory transparency standards. This co-design enables a trimodal transduction strategy wherein aptamer-triggered release of MXene@Au nanohybrids generates electrochemical, photothermal, and colorimetric responses from a single assay, decoded by the interpretable AI framework. Benchmarking against liquid chromatography-tandem mass spectrometry (LC-MS/MS) in tomato, soil, and field water samples, the platform achieves quantitative agreement (R2 = 0.991, slope = 0.83), confirming on-site screening reliability. This paradigm bridges the portability-accuracy gap in field-deployable trace pollutant surveillance, with implications for agricultural runoff and drinking water protection.