Revolutionary Deep Learning-Based System Paves the Way for Efficient Battery Health Assessment

Abstract
As the electric vehicle market continues to surge, the proper assessment of used batteries has become increasingly important. However, current technologies for assessing used batteries, which involve separately estimating the State-of-Health (SoH) of the pack and its individual modules, require multiple times of cycling tests and lead to time inefficiency and power consumption. The proposed DeepSUGAR, a deep learning-based framework for SoH estimation using a generative algorithm based on graphical representation techniques to reveal individual module health, offers the advantage of estimating the status of internal modules replying on battery pack SoH. The cycling profiles of a simultaneously measured 14S7P pack and its constituent modules were analyzed, and a convolutional neural network (CNN) was trained by spatializing cycling curves to estimate SoH. DeepSUGAR, trained on pack data, showed outstanding performance with an RMSE of 5.31 × 10−3 and its applicability was validated by testing with module data, resulting in an RMSE of 7.38 × 10−3. Furthermore, the generated module cycling profiles from pack SoH using the deep generative model were fed into the trained CNN and showed a remarkable performance with an RMSE of 8.38 × 10−3. DeepSUGAR can significantly reduce power consumption, processing cost, and carbon dioxide emissions by integrating module-level diagnosis within the pack-level assessment process.

As the electric vehicle market continues to surge, the assessment of used batteries has become increasingly crucial. In a remarkable breakthrough, a team of researchers, led by Professor Donghyuk Kim and Professor Yunseok Choi in the School of Energy and Chemical Engineering at UNIST, along with Professor Hankwon Lim of the Graduate School of Carbon Neutrality at UNIST, has developed DeepSUGAR. This advanced deep learning-based framework offers a novel approach to estimating the State-of-Health (SoH) of exhausted batteries, improving efficiency and reducing power consumption.

Current assessment technologies for used batteries involve separate estimation of the SoH of the battery pack and its individual modules, leading to time inefficiency and excessive power consumption. DeepSUGAR addresses these challenges by utilizing a generative algorithm based on graphical representation techniques, enabling the estimation of individual module health based on battery pack SoH.

The research team analyzed the cycling profiles of a 14S7P pack and its constituent modules, training a convolutional neural network (CNN) to estimate SoH by spatializing cycling curves. DeepSUGAR, trained on pack data, exhibited outstanding performance with a Root Mean Square Error (RMSE) of 5.31 × 10−3. Validation testing with module data resulted in an RMSE of 7.38 × 10−3, further confirming its applicability. Additionally, the generated module cycling profiles from pack SoH using the deep generative model demonstrated remarkable performance with an RMSE of 8.38 × 10−3.

DeepSUGAR offers several key advantages, including reduced power consumption, processing costs, and carbon dioxide emissions, by integrating module-level diagnosis within the pack-level assessment process. This breakthrough technology has the potential to significantly impact battery health management, as it can diagnose the health status of exhausted batteries without being limited by the type of device.

“We have established a verification system that can determine whether a used battery is recyclable without disassembling the battery,” explained Professor Donghyuk Kim. “DeepSUGAR images charging and discharging data, enabling the determination of the health condition of the battery.”

DeepSUGAR’s capabilities extend beyond battery recycling. By predicting the health status of internal modules through pack diagnosis, this technology has the potential to optimize battery performance in various applications, contributing to the realization of green energy in the future.

This research was supported by the Technology Innovation Program, funded by the Ministry of Trade, Industry and Energy (MOTIE), the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (MSIT), and the Carbon Neutrality Demonstration and Research Center of UNIST. The study findings have been published ahead of their official publication in the online version of Journal of Materials Chemistry A on October 17, 2023. It has also been selected as the Back Cover in the November issue.

Journal Reference
Seojoung Park, Dongjun Lim, Hyunjun Lee, et al., ‘A deep learning-based framework for battery reusability verification: one-step state-of-health estimation of pack and constituent modules using a generative algorithm and graphical representation,’ J. Mater. Chem. A., (2023).