Earticle

초록

영어

The electrification of modern energy systems relies heavily on lithium-ion batteries, yet their high energy density exacerbates thermal runaway (TR) safety risks. Traditional TR characterization is expensive, time-consuming, and reliant on post-mortem analysis. To enable design-stage safety screening, this study proposes a physics-informed machine learning framework utilizing the open Battery Failure Databank. We curated 309 experiments and used 11 pre-test variables to predict three complementary TR severity metrics: total energy yield, unrecovered mass, and positive-end ejecta energy fraction. Evaluating four tree-based models via 10-fold stratified cross-validation, Random Forest emerged as the best baseline across all targets. Results revealed a clear predictability gradient: total energy yield showed high determinism (R² = 0.93), unrecovered mass moderate predictability (R² = 0.57), and directional ejecta energy remained stochastic (R² = 0.30). This framework can help reduce reliance on destructive testing.

목차

Abstract
Introduction
Methods
Data source and governance
Feature selection and target definition
Preprocessing and cross-validation
Model training and evaluation
Results
Predictive Performance Benchmark
Discussion
Interpretation of Predictive Patterns
Practical Implications for Battery Safety Design.
Limitations
Conclusions
Acknowledgments
References

키워드

저자

• 원지원 [ Nguyen Chi Nguyen | 경상국립대학교 경영정보학과 ]
• 서종환 [ 경상국립대학교 경영정보학과 ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국경영정보학회 정기 학술대회 [KMIS Conference]

• 간기

  반년간

• 수록기간

  1990~2026

• 십진분류

  KDC 325 DDC 658