Multiphysics simulation teaching of a vehicle power battery based on mechanism surrogate modeling

Under the background of the New Energy Vehicle Industry Development Plan, higher education faces the critical task of cultivating high-level interdisciplinary engineering talents. Power batteries are the core components of new energy vehicles, which involve complex electrochemical–thermal coupling mechanisms. However, traditional teaching practices encounter several challenges: the abstract nature of electrochemical models makes it difficult for students to understand their internal mechanisms; safety risks and equipment limitations restrict high-rate charging and discharging experiments in the classroom; and the fragmented teaching of modeling, parameter identification, and experimental verification lacks a systematic closed-loop perspective. Therefore, this study aims to develop a reproducible and integrated multiphysics simulation teaching case. By focusing on Blade batteries, the research explores an integrated design of mechanism modeling, cosimulation, and parameter identification, aiming to transform abstract theoretical models into intuitive cognitive experiences and improve students' abilities to solve complex engineering problems.

This study implements a simulation teaching scheme based on the synergy of mechanism models and surrogate acceleration. First, a multiphysics coupling model of a commercial Blade battery is established. The electrochemical behavior is characterized by a pseudo-two-dimensional model based on the Doyle–Fuller–Newman framework, describing lithium-ion distribution and dynamics. This is bidirectionally coupled with a three-dimensional solid heat transfer model in COMSOL Multiphysics, where electrochemical heat sources (including reversible entropy heat, ohmic heat, and polarization heat) drive temperature evolution, while the temperature field simultaneously adjusts the electrochemical parameters such as diffusion coefficients and exchange current densities. Second, a coordinated MATLAB–COMSOL simulation environment is developed for parameter identification. To overcome the high computational cost of high-fidelity simulations, a surrogate-assisted Teaching–Learning-Based Optimization (TLBO) framework is introduced. The identification process is decoupled into two sequential stages: the first stage identifies 21 electrochemical parameters using the terminal voltage response, and the second stage identifies 8 thermal parameters based on multipoint temperature data. A Kriging surrogate model is constructed to approximate the expensive objective function, and a lower confidence bound acquisition function is employed to balance exploration and exploitation. This strategy triggers high-fidelity simulation feedback only for the most promising candidates, significantly reducing the number of model calls while maintaining high identification accuracy.

The simulation teaching case was validated through experimental data collected from a high-precision battery test platform. Under the 1C constant-current discharge condition, the simulation terminal voltage curves were highly consistent with the experimental measurements, accurately reproducing the local fluctuation characteristics of the discharge platform. The identified electrochemical parameters yielded a root mean square error (RMSE) of 0.046699 V. Regarding the temperature field, the model successfully reconstructed the spatial temperature distribution, showing that the high-temperature region was concentrated near the positive tab and diffused toward the center. The average temperature RMSE remained at 1.021926 K. To evaluate the generalization capability, the identified parameters were extrapolated to a 1.5C discharge condition without recalibration. The model accurately captured the downward shift of the voltage platform and the accelerated temperature rise caused by the increased rate, maintaining a high determination coefficient. Crucially, compared to the traditional optimization method without surrogate assistance, the proposed strategy reduced the number of high-fidelity model calls from 9600 to 349, achieving a 96.4% improvement in computational efficiency. Furthermore, sensitivity analysis revealed that voltage prediction is dominated by geometric and mass transfer parameters, while temperature prediction is primarily governed by boundary convection and thermal conductivities.

This research provides a systematic and efficient pathway for the teaching of power battery simulation. By integrating mechanism modeling with surrogate-assisted optimization, the proposed teaching mode effectively resolves the conflict between simulation accuracy and computational time in the classroom setting. The visualization of multiphysics fields helps students bridge the gap between abstract mathematical equations and physical phenomena. The quantitative sensitivity analysis and cross-rate validation further cultivate students' rigorous engineering thinking and abilities to interpret complex system behaviors. This case study serves as a valuable reference for the construction of virtual simulation laboratories and the reform of courses related to new energy vehicle engineering.