Transforming 3D multi-physics simulations into real-time onboard intelligence for energy storage systems thermal management with physics-informed neural networks

Accurately capturing the electro-thermal behavior of batteries and the coupled thermal-fluid dynamics of cooling media during energy storage system operation is critical for ensuring stable and safe performance. Currently, battery thermal management systems typically monitor only a few points of surface temperature, which restricts precise assessment of spatial temperature distribution and cooling fluid flow. We develop a breakthrough methodology that transforms high-fidelity numerical simulations into lightweight neural network surrogates capable of onboard deployment. Physics-informed deep neural networks are trained on comprehensive 3D multi-physics simulation datasets, establishing nonlinear mappings between operational parameters and spatial field distributions. The resulting surrogate model achieves remarkable computational acceleration from hours to sub-seconds (0.72 s) while maintaining high accuracy (maximum temperature error < 0.43 K). Successfully integrated into an Energy Management System operating on standard hardware, the onboard model enables three critical real-time applications: (1) 3D field reconstruction that extends monitoring capabilities beyond sparse sensor networks. (2) Multi-objective thermal management optimization revealing quantitative trade-offs between temperature control and energy consumption through Pareto analysis. (3) Non-invasive abnormal state detection through reverse parameter estimation, achieving 2.19% relative error in identifying internal heat generation anomalies. This work establishes the first practical framework for transitioning from offline multi-physics modeling to real-time onboard intelligence, demonstrating how physics-constrained machine learning can transform complex simulation tools into deployable engineering solutions. The methodology provides a foundation for next-generation battery management systems, enabling predictive thermal control, autonomous optimization, and enhanced safety monitoring in energy storage applications.