融合双模式超声成像与深度视觉的电池缺陷程度辨识方法

武予然; 李泽星; 王鹏博; 申梦圆; 牛靖凯; 妙珊

doi:10.19799/j.cnki.2095-4239.2026.0156

您当前的位置：

首页 >

文章列表页 >

融合双模式超声成像与深度视觉的电池缺陷程度辨识方法

XXXX | 更新时间：2026-05-07

- 融合双模式超声成像与深度视觉的电池缺陷程度辨识方法
- Method for Identifying the Degree of Battery Defects by Fusing Dual-Mode Ultrasonic Imaging and Deep Vision
- 储能科学与技术 2026年页码：1-12
- 作者机构：
  
  1.郑州电力高等专科学校，河南郑州450000
  2.郑州大学电气与信息工程学院，河南郑州 450001
- 作者简介：
  
  武予然（1981—），女，讲师，研究方向为锂电池预警技术，E-mail：wuyuran@zepc.edu.cn；
  妙珊（1989—），讲师，研究方向为储能电池管理与控制，E-mail：miaos@zzu.edu.cn
- 基金信息：
  
  国网河南省电力公司科技项目资助(5217S0260005)
- DOI：10.19799/j.cnki.2095-4239.2026.0156
  中图分类号： TM 92
- 收稿：2026-02-12，
  
  修回：2026-04-21，
  
  网络首发：2026-05-06，
- 稿件说明：
移动端阅览
武予然, 李泽星, 王鹏博, 等. 融合双模式超声成像与深度视觉的电池缺陷程度辨识方法[J]. 储能科学与技术, XXXX, XX(XX): 1-12.

WU Yuran, LI Zexing, WANG Pengbo, et al. Method for Identifying the Degree of Battery Defects by Fusing Dual-Mode Ultrasonic Imaging and Deep Vision[J]. Energy Storage Science and Technology, XXXX, XX(XX): 1-12.
武予然, 李泽星, 王鹏博, 等. 融合双模式超声成像与深度视觉的电池缺陷程度辨识方法[J]. 储能科学与技术, XXXX, XX(XX): 1-12. DOI： 10.19799/j.cnki.2095-4239.2026.0156.

WU Yuran, LI Zexing, WANG Pengbo, et al. Method for Identifying the Degree of Battery Defects by Fusing Dual-Mode Ultrasonic Imaging and Deep Vision[J]. Energy Storage Science and Technology, XXXX, XX(XX): 1-12. DOI： 10.19799/j.cnki.2095-4239.2026.0156.

摘要

为解决电池生产、运输及服役全周期的内外部缺陷检测难题，提升缺陷评估的定量化精度与工程实用性，本文提出一种融合双模式超声成像与深度视觉的电池缺陷程度辨识方法。首先，基于超声透射扫描与反射扫描双模式成像技术，结合多工况缺陷电池制备与数据增强方法，以36块样本电池为对象，构建覆盖气泡、磕碰、划痕、附着物、浸润度等典型缺陷的结构化图像数据库。其次，设计基于深度学习与计算机视觉的电池缺陷辨识模型，提出集成残差学习与注意力门控机制的Attention Residual U-Net缺陷分割网络，通过残差块缓解梯度消失问题，借助注意力门控强化缺陷区域特征，实现对电池微小、不规则缺陷的精准分割。通过残差块缓解梯度消失问题，借助注意力门控强化缺陷区域特征，实现对电池微小、不规则缺陷的精准分割。最后，基于Attention Residual U-Net输出的缺陷掩膜，提取区域尺度、连通域聚集状态、线性连续性及空间分布等视觉信息，采用K-means聚类确定各类缺陷的基础分级阈值，并结合层次分析法（AHP）与故障模式及影响分析（FMEA）完成风险修正与权重分配，将综合得分离散化为优质、良好、合格、风险四个质量等级，实现缺陷程度的量化辨识与分级预警。经所构建包含2500余张缺陷图像数据库的实例验证，该方法mIoU为0.8783、Dice系数为0.9328，可精准辨识电池内外部缺陷并量化状态等级，为电池质量管控提供高效可靠的技术支撑。

Abstract

To address the challenges of internal and external defect detection throughout the full life cycle of batteries including production

transportation and service

and to improve the quantitative accuracy and engineering practicability of defect evaluation

this paper proposes a battery defect degree identification method integrating dual-mode ultrasonic imaging and deep vision. Firstly

based on the dual-mode imaging technology of ultrasonic transmission scanning and reflection scanning

combined with the preparation of defective batteries under multiple working conditions and data augmentation methods

a structured image database covering typical defects such as air bubbles

mechanical indentation

scratches

surface attachments

and electrolyte infiltration uniformity was constructed based on 36 sample batteries. Secondly

a battery defect identification model based on deep learning and computer vision is designed

and an Attention Residual U-Net defect segmentation network integrating residual learning and attention gate mechanism is proposed. The residual block is adopted to alleviate the gradient vanishing problem

and the attention gate is utilized to enhance the features of defect regions

so as to achieve accurate segmentation of tiny and irregular defects in batteries. Finally

to realize quantitative identification and graded early warning of defect severity

the defect thresholds

score mapping rules

and weight allocation are quantified based on K-means clustering

Analytic Hierarchy Process (AHP)

and Failure Mode and Effects Analysis (FMEA). Specifically

K-means clustering and normal distribution fitting are used to determine the natural thresholds of defect pixel areas; AHP is applied to quantify the impact weights of different defects; FMEA is adopted to analyze the failure modes

severity

and occurrence probability of each defect

and the score mapping rules are formulated to map defect parameters to quantitative scores. The comprehensive score is then discretized into four quality levels: Excellent

Good

Qualified

and Risk

realizing hierarchical early warning of battery states Verified by the constructed database containing more than 2500 defect images

the proposed method achieves a mIoU of 0.8783 and a Dice coefficient of 0.9328. It can accurately identify internal and external battery defects and quantify state grades

providing efficient and reliable technical support for battery quality control.

关键词

Keywords

references

Xie Yuxi, Xu Xiang, Liu Shiyan. Machine vision-based detection of surface defects in cylindrical battery cases[J]. Journal of Energy Storage, 2024, 101: 113949.

de Lime A B, Lein T, Maletti S, et al. Impact of electrode defects on battery cell performance: a mini review[J]. Batteries & Supercaps, 2022, 5: e202200239.

Phillips V A, Hugo J A. X-ray Hazard from Electron Microscopes[J]. Science, 1964, 143: 1120-1121.

Qu Zhe, Jiang Peng, Zhang Wei. Development and Application of Infrared Thermography Non-Destructive Testing Techniques[J]. Sensors, 2020, 20: 3851.

Nara H, Yokoshima T, Osaka T. Technology of electrochemical impedance spectroscopy for an energy-sustainable society[J]. Current Opinion in Electrochemistry, 2020, 20: 66-77.

Chacon X C A, Laureti S, Ricci M, et al. A review of non-destructive techniques for lithium-ion battery performance analysis[J]. World Electric Vehicle Journal, 2023, 14:305.

郝奕帆, 祝夏雨, 王静, 等. 电池无损检测监测方法分析[J]. 储能科学与技术, 2023, 12(5):1713-1737.

HAO Yifan, ZHU Xiayu, WANG Jing, et al. Analysis of non-destructive testing and monitoring methods for batteries[J]. Energy Storage Science and Technology, 2023, 12(5): 1713-1737.

靳欣, 张建茹, 王其钰, 等. 混合固液锂离子电池的热失控行为研究[J]. 储能科学与技术, 2024, 13(1): 48-56.

JIN Xin, ZHANG Jianru, WANG Qiyu, et al. Study on thermal runaway behavior of hybrid solid-liquid lithium-ion batteries[J]. Energy Storage Science and Technology, 2024, 13(1): 48-56.

Wang Yong, Lai Xin, Chen Qiang, et al. Progress and challenges in ultrasonic technology for state estimation and defect detection of lithium-ion batteries[J]. Energy Storage Materials, 2024, 69: 23.

Jia Xianyi, Zhu Jiangong, Michael Knapp, et al. A review of battery failure: classification, mechanisms, analysis, and management[J]. Renewable and Sustainable Energy Reviews, 2024, 207: 113949.

李龙飞, 郑永泉, 万旺军, 等. 锂离子电池缺陷检测技术及失效机理分析研究进展 [J]. 材料导报,2025.

LI Longfei, ZHENG Yongquan, WAN Wangjun, et al. Research progress on defect detection technology and failure mechanism analysis of lithium-ion batteries[J]. Materials Reports, 2025.

Williams D, et al. Battery age monitoring: Ultrasonic monitoring of ageing and degradation in lithium-ion batteries[J]. Journal of Power Sources, 2025, 596: 234028.

Peruń G. Advances in Non-Destructive Testing Methods[J]. Materials, 2024, 17: 554.

Geng Chen, Wang Anqi, Zhang Yi et al. Operando monitoring of state of health for lithium battery via fiber optic ultrasound imaging system, 2025.

Riaz W, Qi X, Ji J, et al. Multi-Scale Feature Convolutional Modeling for Industrial Weld Defects Detection in Battery Manufacturing[J]. Fractal and Fractional, 2025, 9: 611.

Zhang Jing, Wang Yong, Jiang Bo, et al. Realistic fault detection of li-ion battery via dynamical deep learning[J]. Nature Communications, 2023, 14(1): 5940.

Kong Xin, Plett G L, Trimboli M S, et al. Pseudo-two-dimensional model and impedance diagnosis of micro internal short circuit in lithium-ion cells[J]. Journal of Energy Storage, 2020, 27: 101085.

Jiang Jun, Li Tao, Chang Chao, et al. Fault diagnosis method for lithium-ion batteries in electric vehicles based on isolated forest algorithm[J]. Journal of Energy Storage, 2022, 50: 104177.

Lu Jun, Xiong Rui, Tian Jun, et al. Deep learning to estimate lithium-ion battery state of health without additional degradation experiments[J]. Nature Communications, 2023, 14: 2760.

Yao Qiang, Song Xin, Xie Wei. State of health estimation of lithium-ion battery based on CNN–WNN–WLSTM[J]. Complex & Intelligent Systems, 2024, 10: 2919-2936.

张有兵, 鲍俊挺, 潘红武, 等. 基于热电联合模型和深度学习的锂离子电池温度估计方法[J/OL ] . 储能科学与技术, 2026, 1 - 13 .

ZHANG Youbing, BAO Junting, PAN Hongwu, et al. Lithium-ion battery temperature estimation method based on thermoelectric coupling model and deep learning[J/OL]. Energy Storage Science and Technology, 2026, 1-13.

Jia Yu, et al. Safety issues of defective lithium-ion batteries: identification and risk evaluation[J]. Journal of Materials Chemistry A, 2020, 8(25): 12472-12484.

Zhang Jing, Lv Xin, Zhang Hao, Liu Bo. AResU-Net: Attention residual U-Net for brain tumor segmentation[J]. Symmetry, 2020, 12(5): 721.

Guo Chao, Szemenyei M, Hu Yu, Wang Wei, Zhou Wei, Yi Yu. Channel Attention Residual U-Net for Retinal Vessel Segmentation[C]. ICASSP 2021, Toronto, Canada, 2021: 1185-1189.

贾九汇, 林赟, 叶盛波, 等. 基于注意力残差 U-Net 的探地雷达杂波抑制方法 [C]// 中国电子学会电波传播分会, 西安电子科技大学. 第十九届全国电波传播年会论文集. 北方工业大学; 中国科学院空天信息创新研究院, 2025: 615-618.

JIA Jiuhui, LIN Yun, YE Shengbo, et al. Ground penetrating radar clutter suppression method based on attention residual U-Net [C]//Radio Wave Propagation Branch of Chinese Institute of Electronics, Xidian University. Proceedings of the 19th National Radio Wave Propagation Conference. North China University of Technology; Aerospace Information Research Institute, Chinese Academy of Sciences, 2025: 615-618.

王丽娜, 孙阳, 张红春, 等. 基于注意力的多尺度残差U-Net的海洋中尺度涡检测[J]. 海洋与湖沼, 2025, 56(1): 64-76.

WANG Lina, SUN Yang, ZHANG Hongchun, et al. Ocean mesoscale eddy detection based on attention multi-scale residual U-Net[J]. Oceanologia et Limnologia Sinica, 2025, 56(1): 64-76.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

半固态与液态电解质大容量电池热失控产热产气特征对比

基于声纹与AI分析技术的化学电池极片缺陷识别

AI for Science时代下的电池平台化智能研发

大容量磷酸铁锂电池模组热失控研究

电池无损检测监测方法分析