|
|
| |
| 基于双向LSTM自动编码器的新型无监督深度学习结构损伤识别 |
| A Novel Unsupervised Deep Learning Framework for Structural Damage Identification Based on Bi-Directional LSTM Autoencoder |
| 投稿时间：2024-12-03 修订日期：2025-01-09 |
| DOI： |
| 中文关键词: 结构健康监测 自编码器 双向长短期记忆 损伤阈值 无监督深度学习 |
| 英文关键词:Structural Health Monitoring Autoencoder Bi-Directional Long Short-Term Memory Autoencoder Damage Threshold Unsupervised Deep Learning |
| 基金项目:国家重点研发计划项目(2019YFC1511004-05) |
|
| 摘要点击次数: 45 |
| 全文下载次数: 0 |
| 中文摘要: |
| 工程结构损伤评估已成为土木工程领域的一个关键研究方向,近年来受到高度关注。虽然基于振动的结构健康监测(SHM)已在无监督深度学习方面取得进展,但在从环境振动信号中定位和量化结构损伤方面仍然存在局限性。因此提出了一种无监督深度学习方法,通过双向长短期记忆自动编码器(Bi-LSTM-AE)实现高效的结构损伤诊断。该方法仅通过振动响应数据实现对结构状态的识别、损伤定位及量化分析。以未损状态的加速度数据为训练基础,以提取并编码损伤特征,使模型可以最小误差地还原输入信号。在损伤状态下,模型表现出显著的重构误差,通过损伤敏感特征(DSF)量化误差差异,并计算损伤检测阈值。该方法的有效性通过滚珠轴承故障检测和短跨钢梁桥的振动实验进行验证。在多种损伤状态中,Bi-LSTM-AE 能够区分正常与故障状态并实现损伤定位,尤其在轻微损伤识别方面展现了高敏感性。研究结果表明,Bi-LSTM-AE 在各种结构应用中均表现出优异的诊断性能,将 Bi-LSTM 网络引入无监督 SHM 框架,不仅显著提升了损伤识别的精度与鲁棒性,还为复杂结构的实时监测与诊断提供了有效途径。 |
| 英文摘要: |
| Structural damage evaluation has become a critical research domain within structural health monitoring (SHM), attracting sustained attention over the years. While vibration-based SHM has achieved progress in unsupervised deep learning, challenges remain in localizing and quantifying structural damage from ambient vibration response. This study proposes an unsupervised deep learning method using a bi-directional long short-term memory autoencoder (Bi-LSTM-AE) for efficient structural damage diagnosis. The method identifies structural states, localizes damage, and quantifies it based on acceleration data of the undamaged state to extract and encode damage features. The model minimizes the reconstruction error of input signals, with pronounced errors under damage conditions. These errors are quantified by Damage-sensitive features (DSF) and damage detection thresholds are established. The approach is validated through ball bearing fault detection and vibration experiments on a short-span steel bridge. Bi-LSTM-AE effectively distinguishes normal and damaged states, achieves damage localization, and demonstrates high sensitivity to minor damage. Results indicate that Bi-LSTM-AE enhances damage identification accuracy and robustness, providing an efficient tool for real-time monitoring and diagnostics in complex structures. |
| 查看/发表评论 下载PDF阅读器 |
|
|
|
|
