A hybrid DVA–machine learning framework for real-time state of health assessment in commercial lithium-ion batteries
Pereira, Eric L. Hossain, Md Zawad Ogun, Damilola Soares, Davi
Pereira, Eric L.
Hossain, Md Zawad
Ogun, Damilola
Soares, Davi
Other Names
Location
Time Period
Advisors
Original Date
Digitization Date
Issue Date
2026-04-23
Type
Article
Genre
Keywords
Differential voltage analysis,Lithium-ion batteries,State of health
Subjects (LCSH)
Citation
E. L. Pereira, M. Z. Hossain, D. Ogun and D. Soares, "A Hybrid DVA–Machine Learning Framework for Real-Time State of Health Assessment in Commercial Lithium-Ion Batteries," in IEEE Transactions on Industry Applications, doi: 10.1109/TIA.2026.3687187.
Abstract
Lithium-ion batteries (LIBs) are an essential and versatile energy source for portable electronics, grid energy storage, and electric vehicles. To meet the increasing demand for energy, it is crucial to understand physical changes to further improve LIBs performance and safety. State of health (SOH) is a metric that indicates how safe LIBs are for operation. This work presents a comprehensive framework for SOH assessment of LIBs. The proposed framework was validated using data from commercial LIBs tested at 1C/1C (charge/discharge), at different temperatures and different depths of discharge (DOD). Reference performance tests (RPTs) were conducted until lithium-ion (Li-ion) cells reached the end of life (EOL). Differential voltage analysis (DVA) was used to acquire relevant physical parameters from RPTs. These parameters were used as engineered features for the machine learning (ML) models, replacing the use of raw dV/dQ curves commonly adopted in DVA–ML studies. The continuous monitoring of DVA provided updates about Li-ion cell degradation based on internal resistance, slippages, active masses, and stoichiometries of both electrodes. To assess Li-ion cells EOL, three machine learning models, namely, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) were applied. To compare and select the best model, all three were integrated into a Raspberry Pi computer. Results indicate that cells cycled at 100% DOD experienced greater cathode material loss compared to those cycled at lower DODs. SVM and XGBoost outperformed RF in both accuracy (93.3%) and recall for failure detection, with SVM achieving the fastest execution time (0.1583 s). This attained accuracy indicates the potential of this affordable ML-based framework to be employed in real-time assessments of SOH in either small or larger LIBs with different chemistries. © 1972-2012 IEEE.
Table of Contents
Description
Click on the DOI link to access this article at the publishers website (may not be free).
Publisher
Institute of Electrical and Electronics Engineers Inc.
Journal
IEEE Transactions on Industry Applications
Book Title
Series
Digital Collection
Finding Aid URL
Use and Reproduction
Archival Collection
PubMed ID
ISSN
00939994