Scholarlink

<p>Efficient thermal management of solid-state batteries is a critical challenge in high-altitude eVTOL systems, where reduced air density limits convective cooling and increases the risk of thermal imbalance and degradation. Current methods such as traditional machine learning (ML) models like XGBoost and deep learning (DL) models such as CNN-BiLSTM often fail to simultaneously capture spatial thermal variations across individual cells and long-term degradation dynamics, leading to poor predictive performance and delayed thermal regulation. To address these limitations, a novel physics-aware spatiotemporal predictive framework is proposed, integrating a Convolutional Autoencoder (CAE) for compact spatial thermal representation with a Hierarchical Thermal Dynamics Transformer (HTDT) for degradation-aware temporal forecasting. The proposed framework incorporates physics-inspired electro-thermal proxy variables, including estimated current and Joule heat generation, to improve degradation-sensitive learning and interpretability. The framework does not explicitly solve electrochemical governing equations or implement PINN-based constraints. The model incorporates physics-inspired electro-thermal features, including estimated charging current and Joule heat generation, to enhance electro-thermal modeling fidelity while maintaining data-driven adaptability. The framework is implemented in Python using TensorFlow and PyTorch and evaluated on a high-resolution dataset of solid-state battery cycles capturing thermal, voltage, and resistance dynamics. Comparative results demonstrate superior predictive performance, achieving an RMSE of 0.125&#xa0;°C, MAE of 0.092&#xa0;°C, and R² of 0.985, representing an approximate 30–50% improvement over baseline models. Ablation studies further confirm the contribution of latent spatial embeddings and physics-guided attention mechanisms. The predictive thermal regulation layer enables real-time adjustment of discharge current and operational load to prevent hotspot formation. Unlike existing approaches that treat thermal prediction and degradation modeling independently, the proposed framework uniquely integrates physics-informed features with hierarchical spatiotemporal learning and predictive thermal regulation, enabling accurate and proactive thermal management under high-altitude conditions.</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

Spatiotemporal electrothermal modeling and predictive thermal regulation of solid-state batteries under high-altitude eVTOL operating conditions using deep transformer networks

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Natrayan Lakshmaiya

摘要

Efficient thermal management of solid-state batteries is a critical challenge in high-altitude eVTOL systems, where reduced air density limits convective cooling and increases the risk of thermal imbalance and degradation. Current methods such as traditional machine learning (ML) models like XGBoost and deep learning (DL) models such as CNN-BiLSTM often fail to simultaneously capture spatial thermal variations across individual cells and long-term degradation dynamics, leading to poor predictive performance and delayed thermal regulation. To address these limitations, a novel physics-aware spatiotemporal predictive framework is proposed, integrating a Convolutional Autoencoder (CAE) for compact spatial thermal representation with a Hierarchical Thermal Dynamics Transformer (HTDT) for degradation-aware temporal forecasting. The proposed framework incorporates physics-inspired electro-thermal proxy variables, including estimated current and Joule heat generation, to improve degradation-sensitive learning and interpretability. The framework does not explicitly solve electrochemical governing equations or implement PINN-based constraints. The model incorporates physics-inspired electro-thermal features, including estimated charging current and Joule heat generation, to enhance electro-thermal modeling fidelity while maintaining data-driven adaptability. The framework is implemented in Python using TensorFlow and PyTorch and evaluated on a high-resolution dataset of solid-state battery cycles capturing thermal, voltage, and resistance dynamics. Comparative results demonstrate superior predictive performance, achieving an RMSE of 0.125 °C, MAE of 0.092 °C, and R² of 0.985, representing an approximate 30–50% improvement over baseline models. Ablation studies further confirm the contribution of latent spatial embeddings and physics-guided attention mechanisms. The predictive thermal regulation layer enables real-time adjustment of discharge current and operational load to prevent hotspot formation. Unlike existing approaches that treat thermal prediction and degradation modeling independently, the proposed framework uniquely integrates physics-informed features with hierarchical spatiotemporal learning and predictive thermal regulation, enabling accurate and proactive thermal management under high-altitude conditions.