Work

Graph Neural Network–Based State‐of‐Health Estimation Applying Interpretable Distribution of Relaxation Times Features Extracted From Electrochemical Impedance Spectroscopy

Year: 2026

Type: article

Abstract: Multiple methods have been used for estimating the state‐of‐health (SOH) of a lithium‐ion battery (LIB), but rarely do we see methods that compute the SOH and provide a physical interpretation of the battery. Therefore, we introduce in this study a framework that integrates graph‐based and conventional learning techniques to assess the battery degradation by extracting the features from the distribution of relaxation times (DRTs). We first changed the electrochemical impedance spectroscopy (EIS) data into non‐negative DRT features, which were obtained through the adaptive Tikhonov regularization with the regularized least square inversion. From this, a graph‐structured representation is represented, and then we use a graph neural network (GNN) architecture, such as graph sample and aggregate (GraphSAGE), graph convolutional network (GCN), and graph attention network (GAT), for SOH regression. As a nongraph benchmark, we used the support vector regression (SVR) and long short‐term memory (LSTM) networks to evaluate the performance of graph‐based models. GraphSAGE achieved R 2 = 0.838 on B0005 and R 2 = 0.851 on B0006 in the optimal configuration, demonstrating competitiveness with SVR. The pipeline robustness was confirmed through a three‐level bootstrap analysis with DRT peak uncertainty below 5% under realistic noise. The analysis also identified the solid electrolyte interface (SEI) and the charge‐transfer band (3–10 ms) as the dominant SOH indicator in our case. Four additional battery groups, spanning from 4°C to 44°C, confirmed cross‐condition transferability. Since the DRT‐GNN method is also a computationally lightweight approach that provides physically interpretable features reflecting ongoing electrochemical degradation processes, it can potentially be extended to an online battery management system (BMS).

Source: International Transactions on Electrical Energy Systems

DOI: 10.1155/etep/2012898

Authors: Barancira Ted Doral, Oluwaseun I. Ekuewa, Femi E. Okemakinde, Sein Oh, Seok-Teak Yun, Jonghoon Kim

Institutions: Chungnam National University, Korea Transportation Safety Authority, Gyeongsang National University

Language: English

Topic: Battery Modeling and Management

Subfield: Automotive Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Affordable and clean energy

Open Access status: gold

Estimated APC paid: $2,350