Application of convolutional neural network in random structural damage identification. (February 2021)

• Record Type: Journal Article
• Authors: Zhan, Yulin; Lu, Siji; Xiang, Tianyu; Wei, Tao
• Is Part Of: Structures. Volume 29(2021)
• Page Start: 570
• Page End: 576
• Publication Date: 2021-02
• Subjects: Structural damage identification -- Convolutional Neural Network (CNN) -- Randomness of structural parameters -- Noise-resistance ability -- Identification accuracy; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2020.11.056
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: Structural parameters are random when structural damage identification is carried out. The problem of the influence of the randomness of structural parameters on the accuracy of Convolutional Neural Network (CNN) structural damage recognition is addressed. This paper first describes the definition of CNN and its principle in structural damage identification. Then, the influence of randomness of structural parameters on the damage identification accuracy of CNN was studied, and the noise-resistance ability of CNN was studied through the combined effect of randomness of structural parameters and sampling noise. The results show that the randomness of structural parameters will affect the damage identification accuracy of CNN, and this effect cannot be ignored. Meanwhile, the mass and elastic modulus play a leading role. In addition, subjected to the combined effect of the randomness of structural parameters and sampling noise, the CNN network exhibits a notable noise-resistance ability.