Scholarlink

<p><i>DegradAI</i> is a chemistry-aware deep learning framework designed to predict long-term lithium-ion battery capacity fade using limited early-cycle data (~5 h), enabling robust diagnostics across diverse chemistries and real-world operating conditions. Its novelty lies in explicit cathode chemistry identification (LiNiCoAlO<sub>2</sub>, LiNiMnCoO<sub>2</sub>, LiFePO<sub>4</sub>, and LiCoO<sub>2</sub>) integrated with a dynamically adaptive architecture. Comprehensive validation across varied temperatures, discharge rates (0.5–3 C), and dynamic cycling profiles demonstrates scalability and transferability. High accuracy is achieved across most chemistries, with a noted variation for LiNiCoAlO<sub>2</sub> due to complex degradation mechanisms. <i>DegradAI</i> can also generate high-fidelity synthetic data; augmenting training sets with an 80:20 synthetic-to-real data mix significantly reduced experimental data needs, improving relative prediction accuracy by up to ~22% while incurring a quantifiable trade-off (~7% increase) in absolute error metrics. This scalable framework is a valuable tool for battery health management in applications like electric vehicles, consumer electronics, and grid storage, adaptable to emerging chemistries.</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

DegradAI: A scalable framework for early battery health diagnosis from limited data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月8日

查看全文
  • Meghana Sudarshan,
  • Jaya Vikeswara Rao Vajja,
  • Vikas Tomar

摘要

DegradAI is a chemistry-aware deep learning framework designed to predict long-term lithium-ion battery capacity fade using limited early-cycle data (~5 h), enabling robust diagnostics across diverse chemistries and real-world operating conditions. Its novelty lies in explicit cathode chemistry identification (LiNiCoAlO2, LiNiMnCoO2, LiFePO4, and LiCoO2) integrated with a dynamically adaptive architecture. Comprehensive validation across varied temperatures, discharge rates (0.5–3 C), and dynamic cycling profiles demonstrates scalability and transferability. High accuracy is achieved across most chemistries, with a noted variation for LiNiCoAlO2 due to complex degradation mechanisms. DegradAI can also generate high-fidelity synthetic data; augmenting training sets with an 80:20 synthetic-to-real data mix significantly reduced experimental data needs, improving relative prediction accuracy by up to ~22% while incurring a quantifiable trade-off (~7% increase) in absolute error metrics. This scalable framework is a valuable tool for battery health management in applications like electric vehicles, consumer electronics, and grid storage, adaptable to emerging chemistries.