fieldid E-Thesis & Research สถาบันเทคโนโลยีไทย-ญี่ปุ่น
สืบค้น:

เขตข้อมูล ข้อมูล
บทคัดย่อ
STATE-OF-HEALTH ESTIMATION OF LITHIUM-ION BATTERY THROUGH AI-DRIVEN ELECTROCHEMICAL IMPEDANCE SPECTROSCOPY MODELING. : This thesis develops a hybrid State of Health (SOH) estimation framework for lithium-ion batteries by integrating Electrochemical Impedance Spectroscopy (EIS), Equivalent Circuit Models (ECM), and Long Short-Term Memory (LSTM) networks. To test the diagnostic reliability of AI models under interrupted real-world conditions, NMC pouch cells were cycled for 200 cycles, deliberately separated into Phase 1 (cycles 1-100) and Phase 2 (cycles 101-200) by a 40-day rest period. Four ECM-extracted resistance parameters were fed into the LSTM, and its internal feature-weighting logic was cross-checked against a Pearson correlation benchmark. During continuous operation (Phase 1), the LSTM achieved high predictive accuracy (RMSE 0.99%) and correctly matched the physical benchmarks, identifying Warburg diffusion (Rw) as the primary degradation driver. However, following the 40-day rest (Phase 2), the battery physical degradation regime fundamentally shifted. While the LSTM maintained high prediction accuracy (RMSE 0.75%), it entirely lost its diagnostic fidelity. Instead of adapting to the new physical hierarchy of resistance parameters, the model relied heavily on historical capacity trends to sustain its accuracy.
ผู้แต่ง
CHATCHARAN YOMJINDA
ประเภทสิ่งพิมพ์
Thesis (MET) 2025
เลขหน้า
190
หัวเรื่อง
Engineering Technology--Thesis
เอกสารฉบับเต็ม
Download full text