基于自编码器无监督学习结构损伤量化检测研究  

Research on unsupervised structural damage quantification detection based on autoencoder

作　　者：刘琦 宁立远 戴华林 王家兴 东尧 Liu Qi;Ning Liyuan;Dai Hualin;Wang Jiaxing;Dong Yao(School of Computer and Information Engineering,Tianjin Chengjian University,Tianjin 300384,China)

机构地区：[1]天津城建大学计算机与信息工程学院,天津300384

出　　处：《国外电子测量技术》2024年第11期116-126,共11页Foreign Electronic Measurement Technology

基　　金：天津市科技计划项目(23YDTPJC00350)资助。

摘　　要：结构健康检测指通过实时或周期性监测评估工程结构的健康状态,深度学习方法因能从原始数据中提取高层特征而备受关注。针对实际应用中损伤类别的多样性,缺乏对损伤状态进行定量分析,提出了部分跳跃卷积自编码器损伤判断量化方法。使用卷积自编码器处理结构响应,将高维数据降维至低维特征空间,通过重构误差设定损伤指标,以判断健康状态;基于低维特征构建损伤系数,实现结构损伤量化。利用国际结构控制协会与美国土木工程协会(IASC-ASCE)IASC-ASCI和IASC-ASCEII数据集验证了算法在损伤判断和量化方面的有效性。实验结果表明,损伤指标对大部分损伤状态的判定准确率达到100%,个别损伤状态下的准确率为96%,对不同损伤状态的量化均符合预期。Structural health monitoring refers to the evaluation of the health condition of engineering structures through real-time or periodic monitoring.Deep learning methods have gained attention due to their ability to extract high-level features from raw data.However,the diversity of damage types in practical applications and the lack of quantitative analysis for damage states remain challenging.In this paper,a partial skip-connected convolutional autoencoder-based approach for damage assessment and quantification is proposed.This method utilizes a convolutional autoencoder to process structural responses,reducing high-dimensional data to a low-dimensional feature space.A damage index is defined based on reconstruction error to assess health status,while a damage coefficient constructed from the lowdimensional features enables quantitative damage assessment.The effectiveness of the algorithm in damage detection and quantification is validated using the IASC-ASCE benchmark structures I and II datasets.Experimental results demonstrate that the damage index achieves 100%accuracy in identifying most damage states,with 96%accuracy in certain specific cases,and that the quantification aligns well with expected values across different damage states.

关 键 词：结构健康检测 卷积自编码器 损伤量化 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TN911.72[自动化与计算机技术—控制科学与工程]