|
| 基于改进Transformer和CNN的建筑结构损伤识别方法研究 |
| Research on Building Structure Damage Identification Method Based on Improved Transformer and CNN |
| 投稿时间：2025-03-03 修订日期：2025-05-02 |
| DOI： |
| 中文关键词: 建筑主体结构 损伤识别 Transformer 卷积神经网络 有限元模型 |
| 英文关键词: building main structure damage identification Transformer convolutional neural network finite element modeling |
| 基金项目:安徽省高校重大科研项目（2024AH040039）；安徽省高校科研项目（2022AH053127） |
|
| 摘要点击次数: 50 |
| 全文下载次数: 0 |
| 中文摘要: |
| 针对建筑结构损伤识别中振动信号噪声干扰显著、多尺度特征提取困难等问题，本文提出了一种基于改进Transformer和卷积神经网络（Improved Transformer and Convolutional Neural Network，ITCN）的结构损伤识别方法。首先，ITCN设计了一种时域增强模块，通过线性变换与多尺度一维卷积神经网络（1D-CNN）的协同作用，在提取全局结构特征的同时强化局部细节的捕获能力；其次，ITCN对Transformer的结构进行了改进，将前馈神经网络与注意力机制从串联改为并联，以提高模型对噪声的鲁棒性。最后，通过系统的对比实验验证，ITCN模型在损伤识别准确率上较传统CNN、长短期记忆网络及标准Transformer等方法均有显著提升，展现出优越的工程应用价值。 |
| 英文摘要: |
| Aiming at the problems of significant vibration signal noise interference and difficult multi-scale feature extraction in building structural damage identification, this paper proposes a structural damage identification method based on Improved Transformer and Convolutional Neural Network (ITCN). First, ITCN designs a time-domain enhancement module that strengthens the ability to capture local details while extracting global structural features through the synergy of linear transformation and multi-scale 1D convolutional neural network (1D-CNN); second, ITCN improves the structure of Transformer by changing the feed-forward neural network and the attention mechanism from series to parallel connection to improve the noise robustness of the model. Finally, through systematic comparison experiments, the ITCN model shows significant improvement in damage identification accuracy compared with traditional CNN, long and short-term memory network, and standard Transformer, demonstrating superior engineering application value. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
