Suppr超能文献

一种基于UNet++的方法,用于使用全波场对碳纤维增强塑料层压板进行分层成像。

A UNet++-Based Approach for Delamination Imaging in CFRP Laminates Using Full Wavefield.

作者信息

Yan Yitian, Yang Kang, Gou Yaxun, Tang Zhifeng, Lv Fuzai, Zeng Zhoumo, Li Jian, Liu Yang

机构信息

State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin 300072, China.

Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32611, USA.

出版信息

Sensors (Basel). 2025 Jul 9;25(14):4292. doi: 10.3390/s25144292.

Abstract

The timely detection of delamination is essential for preventing catastrophic failures and extending the service life of carbon fiber-reinforced polymers (CFRP). Full wavefields in CFRP encapsulate extensive information on the interaction between guided waves and structural damage, making them a widely utilized tool for damage mapping. However, due to the multimodal and dispersive nature of guided waves, interpreting full wavefields remains a significant challenge. This study proposes an end-to-end delamination imaging approach based on UNet++ using 2D frequency domain spectra (FDS) derived from full wavefield data. The proposed method is validated through a self-constructed simulation dataset, experimental data collected using Scanning Laser Doppler Vibrometry, and a publicly available dataset created by Kudela and Ijjeh. The results on the simulated data show that UNet++, trained with multi-frequency FDS, can accurately predict the location, shape, and size of delamination while effectively handling frequency offsets and noise interference in the input FDS. Experimental results further indicate that the model, trained exclusively on simulated data, can be directly applied to real-world scenarios, delivering artifact-free delamination imaging.

摘要

及时检测分层对于防止灾难性故障和延长碳纤维增强聚合物(CFRP)的使用寿命至关重要。CFRP中的全波场包含了关于导波与结构损伤相互作用的丰富信息,使其成为一种广泛用于损伤映射的工具。然而,由于导波的多模态和色散特性,解释全波场仍然是一项重大挑战。本研究提出了一种基于UNet++的端到端分层成像方法,该方法使用从全波场数据导出的二维频域谱(FDS)。所提出的方法通过自建的模拟数据集、使用扫描激光多普勒测振仪收集的实验数据以及由库德拉和伊杰创建的公开可用数据集进行了验证。模拟数据的结果表明,使用多频FDS训练的UNet++能够准确预测分层的位置、形状和大小,同时有效处理输入FDS中的频率偏移和噪声干扰。实验结果进一步表明,仅在模拟数据上训练的模型可以直接应用于实际场景,提供无伪影的分层成像。

https://cdn.ncbi.nlm.nih.gov/pmc/blobs/5ace/12299149/1ef74adde969/sensors-25-04292-g001.jpg

文献AI研究员

20分钟写一篇综述,助力文献阅读效率提升50倍。

立即体验

用中文搜PubMed

大模型驱动的PubMed中文搜索引擎

马上搜索

文档翻译

学术文献翻译模型,支持多种主流文档格式。

立即体验