Traditional autoregressive (AR) model-based damage identification methods construct structural damage sensitive features by trial and error, which are time-consuming, laborious and may lead to poor recognition effect. This study applies convolutional neural networks (CNNs) to quickly and automatically extract high-dimensional features of autoregressive model coefficients (ARMCs). In this research, AR model was utilized to fit the acceleration time series. The input matrices marked with damage location were produced by ARMCs, and then those matrices were sent to the proposed CNN for training. The trained CNN was employed for damage identification and localization. The effectiveness of the proposed method was verified by the damage identification and localization of a three-storied frame structure. The performance of the proposed CNN was compared with multilayer perception (MLP), random forest, and support vector machine (SVM). Meanwhile, the prediction results from different sample types were also discussed. Furthermore, parametric study in relation to the number of accelerometers and ARMCs used is conducted. These analyses demonstrate that the accuracy of CNN tests results reach 100%, 6.67%, 20%, and 25% higher than that of MLP, random forest, and SVM, respectively. Besides, other metrics calculated in this paper (e.g., precision, recall) further indicate that the proposed CNN performs well. The combination of AR and CNN does show excellent performance in damage identification and localization, which seems to be able to resist external excitation changes and accurately identify the multi-location damage and minor damage using limited accelerometers and ARMCs.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Autoregressive Model-Based Structural Damage Identification and Localization Using Convolutional Neural Networks


    Additional title:

    KSCE J Civ Eng


    Contributors:
    Tang, Qizhi (author) / Zhou, Jianting (author) / Xin, Jingzhou (author) / Zhao, Siyu (author) / Zhou, Yingxin (author)

    Published in:

    Publication date :

    2020-07-01


    Size :

    13 pages




    Type of media :

    Article (Journal)


    Type of material :

    Electronic Resource


    Language :

    English