Internal temperature sensing for lithium-ion battery under ultrahigh-rate discharge conditions using physics-informed neural network

In modern military applications such as directional energy and electromagnetic emission, lithium-ion batteries (LIBs) need to be operated under ultrahigh-rate discharge conditions, where the temperature difference between the core and the surface of cell exceed tens of degrees celsius. Such thermal gradients pose great difficulties for the conventional thermal management techniques that rely on cell surface temperature, leading to severe underestimation of the maximum temperature and non-negligible control delay in thermal regulation, thereby increasing the risk of overheating. To address this critical challenge, this study proposes a self-adaptive multi-magnitude physics-informed neural network (SA-MMPINN) framework for internal temperature sensing. This framework integrates thermal model into neural networks architectures, enabling high-accuracy sense of internal temperature based solely on surface temperature while simultaneously identifying convective heat transfer coefficients to adapt to different heat dissipation conditions. Experimental validation using instrumented cells with embedded thermocouples shows that the framework can achieve root mean square error below 0.76 °C for core temperature sensing under 20C discharge conditions with temperature ranging from −5 °C to 35 °C. Meanwhile, the convective heat transfer coefficient can be identified with over 97% accuracy. The proposed SA-MMPINN framework enables accurate sensing for the internal temperature evolution of LIBs, which is crucial for informing the design of next-generation intelligent thermal management systems.