基于EWT-CNN 的结构损伤检测方法研究

打开文本图片集
中图分类号: TB9; TU317 文献标志码: A文章编号: 1674–5124(2025)05–0058–10
Abstract: An adaptive structural damage detection method based on empirical wavelet transform and convolutional neural network is proposed to solve the problem that the damage index of engineering structure is difficult to determine and the damage detection accuracy is insufficient when noise is included. This method can decompose the vibration response signal of the structure into modal components with different natural frequencies and extract damage characteristics, which has high damage detection accuracy and is conducive to online real-time detection. Firstly, the acceleration time series data of structural vibration response are measured and the empirical wavelet transform is carried out according to its amplitude frequency diagram. Then, the mode functions are used as the input of the convolutional neural network to extract structural damage features for the identification of different damage cases. Numerical tests show that under different signal-tonoise ratio conditions, the damage detection accuracy of a single convolutional neural network is 80%-90% , but the accuracy of the proposed method is 100% , and it has stronger anti-noise capability. Shaking table test further verifies the feasibility and effectiveness of the method.
Keywords: empirical wavelet transform; convolutional neural network; structural damage detection; natural frequency; anti-noise capability
0 引 言
工程结构在环境侵蚀、材料老化、疲劳效应和自然灾害等因素作用下,不可避免地将产生各种损伤,这些损伤累积至一定程度将导致结构破坏。(剩余11024字)