True‐2D ResNet Approach on Battery Data Images for Machine Learning Performance Prediction
Year: 2025
Type: article
Abstract: Accurate prediction for performance of rechargeable batteries is critical to both academic research and industry applications. With the availability of big data and deep learning framework, machine learning becomes increasingly important to battery research. The performance of machine learning models is strongly influenced by input methodology. Herein, a true 2D ResNet method to train and predict battery performance based on battery test image data is applied. With theoretical analysis and tests on two large battery datasets, the superiority of 2D image representation compared with traditional 1D representation in principal component analysis method and different models is illustrated. Moreover, a 2D ResNet model is also trained to explore the full potential of deep learning in battery cycle modeling, allowing it to learn directly from raw cycling images. This makes the process a true‐2D ResNet method beyond the previous work. When using deep learning models to predict state of health and remaining useful life of the cell, it is found that the method shows great potential for the state‐of‐the‐art prediction accuracy. This work enhances the understanding of advantages of 2D image representation, which offers new perspectives on how to build efficient machine models for batteries.
Source: Advanced Intelligent Systems
Authors: Yuchuang Cao, Mingyang Wang, Minjeong Byeon, Xi Chen, Jeesoon Choi, Xin Li
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,970
OpenAlex ID: W4412506029