Scholarlink

<p>Optimizing the performance and lifespan of lithium-ion batteries (LIBs) is a key step toward advanced energy storage. Existing multiphysics models often miss important couplings, which limits simulation fidelity, and their intensive computations slow iterative design. This study presents a physics–data fusion framework for multi-objective optimization. A coupled ageing model—covering electrochemical, thermal, mechanical, and side-reaction effects—generates degradation data in COMSOL Multiphysics. These data are used to train machine learning (ML) surrogate models. Electrode thickness, solid phase volume fraction, and initial lithium-ion concentration are chosen by Latin Hypercube Sampling (LHS). SHapley Additive exPlanations (SHAP) interpretation quantifies each variable’s influence on the surrogate outputs. The surrogate is paired with a genetic algorithm to explore the design space, achieving a 28.47% increase in energy density (ED) and an 8.33% reduction in capacity loss (CL). This approach offers valuable insights for LIB structural design and provides potential guidance for improving battery manufacturing processes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-objective optimization of lithium-ion battery design via machine learning surrogate model: balancing energy density and capacity loss

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • Shaoxiao Ju,
  • Peichao Li,
  • Yufu Luo,
  • Keyong Wang,
  • Hengyun Zhang

摘要

Optimizing the performance and lifespan of lithium-ion batteries (LIBs) is a key step toward advanced energy storage. Existing multiphysics models often miss important couplings, which limits simulation fidelity, and their intensive computations slow iterative design. This study presents a physics–data fusion framework for multi-objective optimization. A coupled ageing model—covering electrochemical, thermal, mechanical, and side-reaction effects—generates degradation data in COMSOL Multiphysics. These data are used to train machine learning (ML) surrogate models. Electrode thickness, solid phase volume fraction, and initial lithium-ion concentration are chosen by Latin Hypercube Sampling (LHS). SHapley Additive exPlanations (SHAP) interpretation quantifies each variable’s influence on the surrogate outputs. The surrogate is paired with a genetic algorithm to explore the design space, achieving a 28.47% increase in energy density (ED) and an 8.33% reduction in capacity loss (CL). This approach offers valuable insights for LIB structural design and provides potential guidance for improving battery manufacturing processes.