Cross-resolution Operando Thermal Reconstruction for Lithium-Ion Batteries via Physics-Informed Generative Diffusion Models

The lithium-ion battery (LIB) is a complex multi-physics system in which thermal effects play a pivotal role in coupling various underlying physical phenomena, such as electrochemical re- actions, degradation mechanisms, and mechanical deformations. With the development of the battery technologies, conventional thermal models face a fundamental limitation: achieving high- resolution simulations typically requires increased model complexity, which in turn compromises their suitability for online thermal field reconstruction. Here, we incorporate machine intelligence into the physics-based reconstruction of thermal fields by proposing a generative thermal reconstruction (GenTR) method. GenTR is an innovative, intelligent, remote server-oriented frame- work that directly generates dynamic, high-resolution thermal field details (128×128 pixels) from a localized low-resolution model (8×8 nodes). To achieve this, GenTR for the first time incorporates solid physics guidance into generative artificial intelligence (GAI). The integration of physics-based information with GAI ensures real-time visualization of the internal temperature field, enabling image-based reconstruction within only 150 iteration steps. The mean absolute proportional errors (MAPEs) of GenTR under various conditions remain below 1%. Furthermore, in subsequent discussions, we demonstrate that GenTR effectively circumvents the exponential increase in model complexity that typically accompanies the pursuit of high temporal and spatial fidelity in large-scale systems.