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

储能测试与评价 | 更新时间：2026-07-28

- 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15卷第7期页码：2777-2788
- 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，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(5217S0260005)
- DOI：10.19799/j.cnki.2095-4239.2026.0156
  中图分类号： TM 92
- 收稿：2026-02-24，
  
  修回：2026-04-21，
  
  网络首发：2026-07-28，
  
  纸质出版：2026-07-28
- 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, 2026, 15(7): 2777-2788.

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, 2026, 15(7): 2777-2788.
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, 2026, 15(7): 2777-2788. 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, 2026, 15(7): 2777-2788. 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，本研究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，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。经构建的包含2500余张缺陷图像数据库实例验证，本方法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

This paper proposes a battery defect severity identification method

integrating dual-mode ultrasonic imaging and deep learning-based computer vision to simultaneously address the challenges of detecting internal and external defects throughout the full life cycle of batteries-including production

transportation

and service-and improve the quantitative accuracy and engineering practicability of defect evaluation. First

a structured image database is constructed based on 36 sample batteries using dual-mode (transmission and reflection scanning) ultrasonic imaging

combined with defect fabrication under multiple working conditions and data augmentation techniques. The dataset covers typical defects

including air bubbles

mechanical indentation

scratches

surface attachments

and electrolyte infiltration uniformity. Second

a defect identification model based on deep learning and computer vision is developed. An Attention Residual U-Net defect segmentation network is proposed

incorporating residual learning to mitigate gradient vanishing and attention gate mechanisms to enhance the feature representation in defect regions. This design enables accurate segmentation of small and irregular defects. Finally

to achieve quantitative defect identification and hierarchical early warning

defect thresholds

scoring rules

and weight allocation are systematically established using K-means clustering

analytic hierarchy process (AHP)

and failure mode and effects analysis (FMEA). Specifically

K-means clustering combined with normal distribution fitting is used to determine the natural thresholds for defect pixel areas; AHP quantifies the relative importance of different defect types; and FMEA evaluates failure modes

severity

and occurrence probabilities to construct score mapping rules. The resulting comprehensive score is then classified as excellent

good

qualified

or risk

enabling hierarchical early warning of battery conditions. When verified against a database of more than 2500 defect images

the proposed method achieves a mIoU of 0.8783 and a Dice coefficient of 0.9328

demonstrating that it can accurately identify both internal and external battery defects and provides reliable quantitative evaluation. Thus

the method provides strong technical support for battery quality control.

关键词

Keywords

references

XIE Y X, XU X, LIU S Y. Machine vision-based detection of surface defects in cylindrical battery cases[J]. Journal of Energy Storage, 2024, 101: 113949. DOI:10.1016/j.est.2024.113949.

DU BARET DE LIMÉ A, LEIN T, MALETTI S, et al. Impact of electrode defects on battery cell performance: A review[J]. Batteries & Supercaps, 2022, 5(10): e202200239. DOI:10.1002/batt.202200239.

PHILLIPS V A, HUGO J A. X-ray hazard from electron microscopes[J]. Science, 1964, 143(3611): 1120-1121. DOI:10. 1126/science.143.3611.1120-b.

QU Z, JIANG P, ZHANG W X. Development and application of infrared thermography non-destructive testing techniques[J]. Sensors, 2020, 20(14): 3851. DOI:10.3390/s20143851.

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. DOI:10.1016/j.coelec.2020.02.026.

CHACÓN 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(11): 305. DOI:10.3390/wevj14110305.

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 Y F, ZHU X Y, WANG J, et al. Analysis of battery nondestructive testing and monitoring methods[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. DOI:10.19799/j.cnki.2095-4239.2023.0846.

JIN X, ZHANG J R, WANG Q Y, et al. Study on thermal runaway of hybrid solid-liquid batteries[J]. Energy Storage Science and Technology, 2024, 13(1): 48-56. DOI:10.19799/j.cnki.2095-4239.2023.0846.

WANG Y Y, LAI X, CHEN Q W, et al. Progress and challenges in ultrasonic technology for state estimation and defect detection of lithium-ion batteries[J]. Energy Storage Materials, 2024, 69: 103430. DOI:10.1016/j.ensm.2024.103430.

JIA X Y, ZHU J G, KNAPP M, et al. A review of battery failure: Classification, mechanisms, analysis, and management[J]. Renewable and Sustainable Energy Reviews, 2026, 225: 116 126. DOI:10.1016/j.rser.2025.116126.

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, 39(11): 7-15. DOI:10.11896/cldb.24100016.

LI L F, ZHENG Y Q, WAN W J, et al. Research progress in failure analysis and defect detection of lithium-ion batteries[J]. Materials Reports, 2025, 39(11): 7-15. DOI:10.11896/cldb.24100016.

WILLIAMS D, GREEN J, BUGRYNIEC P, et al. Battery age monitoring: Ultrasonic monitoring of ageing and degradation in lithium-ion batteries[J]. Journal of Power Sources, 2025, 631: 236174. DOI:10.1016/j.jpowsour.2025.236174.

PERUŃ G. Advances in non-destructive testing methods[J]. Materials, 2024, 17(3): 554. DOI:10.3390/ma17030554.

CHEN G, WANG A Q, ZHANG Y, et al. Operando monitoring of state of health for lithium battery via fiber optic ultrasound imaging system[J ] . Opto-Electronic Science, 2025, 4(6): 240036. DOI:10.29026/oes.2025.240036.

RIAZ W, QI X Z, JI J C, et al. Multi-scale feature convolutional modeling for industrial weld defects detection in battery manufacturing[J]. Fractal and Fractional, 2025, 9(9): 611. DOI:10.3390/fractalfract9090611.

ZHANG J Z, WANG Y N, JIANG B B, et al. Realistic fault detection of Li-ion battery via dynamical deep learning[J ] . Nature Communications, 2023, 14: 5940. DOI:10.1038/s41467-023-41226-5.

KONG X D, PLETT G L, SCOTT TRIMBOLI M, 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. DOI:10.1016/j.est.2019.101085.

JIANG J C, LI T Y, CHANG C, 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. DOI:10.1016/j.est.2022.104177.

LU J H, XIONG R, TIAN J P, et al. Deep learning to estimate lithium-ion battery state of health without additional degradation experiments[J]. Nature Communications, 2023, 14: 2760. DOI:10.1038/s41467-023-38458-w.

YAO Q Z, SONG X H, XIE W. State of health estimation of lithium-ion battery based on CNN-WNN-WLSTM[J]. Complex & Intelligent Systems, 2024, 10(2): 2919-2936. DOI:10.1007/s40747-023-01300-3.

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]. 储能科学与技术, 2026, 15(4): 1363-1374.

ZHANG Y B, BAO J T, PAN H W, et al. Lithium-ion battery temperature estimation method based on thermoelectric combined model and deep learning[J]. Energy Storage Science and Technology, 2026, 15(4): 1363-1374.

JIA Y K, LIU B H, HONG Z G, 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 J X, LV X G, ZHANG H B, et al. AResU-Net: Attention residual U-Net for brain tumor segmentation[J]. Symmetry, 2020, 12(5): 721. DOI:10.3390/sym12050721.

GUO C L, SZEMENYEI M, HU Y T, et al. Channel attention residual U-Net for retinal vessel segmentation[EB/OL]. 2020: arXiv: 2004.03702. https://arxiv.org/abs/2004.03702

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. 西安, 2025: 615-618. DOI:10.26914/c.cnkihy.2025.087842.

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 L N, SUN Y, ZHANG H C, et al. A multi-scale attention residual-based U-Net model for detection of ocean mesoscale eddies[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

双液相开式复叠热泵热力电池储能系统的性能