Enhancing lithium-ion battery analysis: A machine learning approach to investigate innovative silicon thin film anode parameters and discharge capacity
JOURNAL OF POWER SOURCES, cilt.674, sa.239808, ss.1-12, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 674 Sayı: 239808
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.jpowsour.2026.239808
- Dergi Adı: JOURNAL OF POWER SOURCES
- Derginin Tarandığı İndeksler: Scopus, Science Citation Index Expanded (SCI-EXPANDED), Chemical Abstracts Core, Chimica, Compendex, INSPEC
- Sayfa Sayıları: ss.1-12
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Süleyman Demirel Üniversitesi Adresli: Evet
Özet
Silicon (Si) thin film anodes offer very high theoretical capacity but also undergo severe volume expansion during lithiation, which causes fracture, loss of electrical contact, and rapid capacity fade in lithium-ion batteries. This study develops a machine learning (ML) framework that predicts discharge capacity and derives processing and operating guidelines for Si thin film anodes. A curated literature dataset with missing fabrication and cycling parameters is completed using ML-based imputation, for which Random Forest gives the highest accuracy. A Random Forest regressor then predicts discharge capacity with strong performance (validation R2 = 0.86, mean absolute error = 174.1 mAh g−1). Feature importance analysis identifies working pressure, cycle number, and C-rate as the most influential variables. The trained model also extrapolates learned degradation trends to later cycle numbers, producing hypothetical capacity fade projections that qualitatively align with the reported behaviour of Si thin film anodes, rather than representing statistically validated long-term predictions. Finally, a model-agnostic Accumulated Local Effects analysis quantifies how fabrication parameters and operating conditions shape both absolute capacity and its fading. The results show that ML not only predicts performance reliably but also supports the rational design of next-generation Si thin film anodes.
Silicon (Si) thin film anodes offer very high theoretical capacity but also undergo severe volume expansion during lithiation, which causes fracture, loss of electrical contact, and rapid capacity fade in lithium-ion batteries. This study develops a machine learning (ML) framework that predicts discharge capacity and derives processing and operating guidelines for Si thin film anodes. A curated literature dataset with missing fabrication and cycling parameters is completed using ML-based imputation, for which Random Forest gives the highest accuracy. A Random Forest regressor then predicts discharge capacity with strong performance (validation R2 = 0.86, mean absolute error = 174.1 mAh g−1). Feature importance analysis identifies working pressure, cycle number, and C-rate as the most influential variables. The trained model also extrapolates learned degradation trends to later cycle numbers, producing hypothetical capacity fade projections that qualitatively align with the reported behaviour of Si thin film anodes, rather than representing statistically validated long-term predictions. Finally, a model-agnostic Accumulated Local Effects analysis quantifies how fabrication parameters and operating conditions shape both absolute capacity and its fading. The results show that ML not only predicts performance reliably but also supports the rational design of next-generation Si thin film anodes.