基于小波散射变换的RC框架结构震后损伤异常智能检测  

Intelligent detection of post earthquake damage anomaly of RC frame structure based on wavelet scattering transform

作　　者：康帅 王自法[2] 周荣环 贺东青[1] 俞叶 李一民 KANG Shuai;WANG Zifa;ZHOU Ronghuan;HE Dongqing;YU Ye;LI Yimin(School of Civil Engineering and Architecture,Henan University,Kaifeng 475004,Henan,China;Institute of Engineering Mechanics,China Earthquake Administration,Harbin 150080,Heilongjiang,China;China Construction Eighth Bureau Development and Construction Co.,Ltd.,Zhengzhou 450000,Henan,China)

机构地区：[1]河南大学建筑工程学院,河南开封475004 [2]中国地震局工程力学研究所,黑龙江哈尔滨150080 [3]中建八局发展建设有限公司,河南郑州450000

出　　处：《建筑科学与工程学报》2025年第2期27-38,共12页Journal of Architecture and Civil Engineering

基　　金：国家自然科学基金项目(51978634);河南省科技公关项目(232102321076)。

摘　　要：基于传统方法的地震损伤评估存在精度和效率低的问题,为实现基于结构地震响应观测数据的震后结构损伤快速评估,并有效解决数据类型不平衡的问题,提出基于时频分析与卷积自编码(CAE)模型的损伤异常数据检测方法。首先应用小波散射变换对原时域信号进行处理,生成时频数据;然后建立相应的卷积自编码网络模型,将时频数据输入CAE模型进行重构训练,根据重构误差确定异常判断阈值;基于该阈值精准区分数据中的异常值,并将计算结果与基于时域输入方法的计算结果进行对比分析,最后验证该方法在噪声环境下的有效性。结果表明:基于CAE模型的异常检测方法可以很好地识别损伤数据集中的异常序列,召回率达到了90%以上,且在小波散射变换作用下的异常检测效果更好;时频分析结合CAE模型的损伤异常数据检测方法极大地提升了震后损伤评估的效率,基于小波散射变换的耗时仅为基于传统时域输入方法的1/3,该方法在噪声作用下也表现出较高的检测精度。Earthquake damage assessment based on traditional methods has the problems of low accuracy and efficiency.In order to achieve rapid assessment of post earthquake structural damage based on structural seismic response observation data and effectively solve the problem of data type imbalance,a damage abnormal data detection method based on time-frequency analysis and convolutional autoencoder(CAE)model was proposed.Firstly,the wavelet scattering transform was applied to process the original time domain signal to generate time-frequency data.Then,the corresponding convolutional autoencoder network model was established,and the time-frequency data was input into the CAE model for reconstruction training.The anomaly judgment threshold was determined according to the reconstruction error.Using this threshold,anomalies in the dataset were accurately distinguished,and the results were compared with those obtained using direct time-domain input method.Finally,the effectiveness of this method under noisy conditions was validated.The results show that the anomaly detection method based on the CAE model can well identify abnormal sequences in the damage data set,achieving a recall rate of over 90%,and the anomaly detection effect under the action of wavelet scattering transform is better.The proposed damage anomaly detection method integrating time-frequency analysis with the CAE model can significantly enhance the efficiency of post earthquake damage assessment.The computation time using wavelet scattering transform is only one-third of that using traditional time-domain input methods,and the method maintains high detection accuracy even in noisy environment.

关 键 词：小波散射变换 异常检测 震后损伤 深度学习 框架结构 

分 类 号：TU375[建筑科学—结构工程]