Decoding Capacity Regeneration in Zinc-Silver Batteries: Gas-Guided Multi-Physics Modeling and Hybrid AI Prediction

Accurately predicting battery life is vital for next-generation energy storage. Zinc-silver batteries (ZSBs) offer high power and energy density but suffer from short cycle life due to parasitic oxygen and hydrogen evolution reactions (OER/HER) that cause irreversible swelling. Here, we uncover a counterintuitive capacity regeneration phenomenon in ZSBs under a 20°C scenario: a regime where conventional degradation models fail. By establishing a monitoring system, we capture synchronized electrical, strain, gas concentration, pressure, and temperature signals. We provide evidence that OER and HER act as early gas-phase precursors to regeneration, preceding electrical and mechanical responses. Their coupled electro-chemo-mechanical effects induce distinct surface strain transitions from peak to valley spikes. Leveraging these physical insights, we propose a hybrid mechanism-data driven framework that integrates a chemistry-informed neural network (CINN) with a TCN-Mamba backbone and a spiking neural network for dynamic residual calibration. This architecture embeds electrochemical degradation equations as soft constraints, ensuring interpretability while achieving high accuracy (e.g., RMSE ≈ 0.29 cycles) remaining useful life prediction. Experimental results show superior accuracy and stability over purely data-driven benchmarks. Our findings establish gas-guided multi-physics monitoring as a transformative paradigm for battery health management in complex, safety-critical applications.