预应力混凝土梁损伤识别的神经网络方法研究

Neural network method for damage identification of prestressed concrete beams

文章通过有限元分析阐明了固有频率变化率和模态置信度在损伤识别研究中的重要意义,提出了以其为损伤指标的反向传播(backpropagation,BP)神经网络和概率神经网络(probabilistic neural network,PNN)的损伤状态识别方法。为验证所提方法的实用性,对11根多级损伤状态的真实预应力混凝土梁进行识别,并与基于固有频率变化率和模态置信度的普通理论方法进行比较。研究表明,普通理论方法实用性较差,很难有效识别各梁损伤状态;而BP神经网络和PNN识别方法均能有效应用于实际中,且具有很高的损伤识别精度,为结构损伤识别方法研究提供了新思路。

In this paper, through simulating modal analysis of damage beams with ABAQUS, it is expounded that both natural frequency change ratio and modal assurance criterion are of great significance for damage identification. Backpropagation（BP） neural network and probabilistic neural net- work（PNN） for damage identification based on these modal parameters are proposed. Through the experimental analysis of 11 prestressed concrete beams under multistage damage status and the comparison between the proposed method and the theoretical method based on natural frequency change ratio and modal assurance criterion, it is shown that the theoretical method based on these damage indexes is hardly practical in engineering application, while the BP and PNN neural networks can be effectively applied to identifying damages with high precision. The study results can offer a new idea for damage identification researches.