서강대학교 전자공학과

Battery Usage-Agnostic Multi-Task Diagnostics Using Contrastive Learning and Knowledge-guided Voltage Relaxation

A research team from the Department of Electronic Engineering at Sogang University, including Jihun Jeon, Hojin Cheon, Hyungseok Seo, and Prof. Hongseok Kim, has published their paper in the prestigious energy journal Journal of Energy Storage (JCR IF 9.8, 2026).

The study proposes a usage-agnostic battery diagnostic framework that leverages voltage relaxation signals to simultaneously estimate capacity, state of health (SOH), and cathode materials through a multi-task learning approach.

By incorporating a knowledge-guided equivalent circuit model (KG-ECM) and supervised contrastive learning, the framework learns a shared and discriminative representation for multiple diagnostic tasks.

Experimental results demonstrate significant improvements over conventional methods, achieving an RMSE of 0.0026 (single-task) and 0.0108 (multi-task), along with 94.6% SOH classification accuracy and 99.6% cathode identification accuracy.

This work highlights the potential of integrating multi-task learning and contrastive learning for reliable and practical battery diagnostics in real-world EV and ESS applications.