联合CNN与LSTM神经网络的斜拉索损伤识别方法
A method for identifying cable damage in cable-stayed bridges by combining CNN and LSTM neural network
朱三凡 1刘世凤 2余印根 3王志俸 2夏樟华2
作者信息
- 1. 健研检测集团有限公司,福建 厦门 361004
- 2. 福州大学土木工程学院,福建 福州 350116
- 3. 福建省永正工程质量检测有限公司,福建 福州 350001
- 折叠
摘要
索结构中,拉索响应与损伤之间处于高度非线性状态,常规数学模型对拉索损伤识别普遍存在精度欠佳问题.针对该问题创建了七丝钢绞线的拉索有限元模型,提出基于组合指标的CNN&LSTM神经网络损伤识别方法.借助该拉索有限元模型模拟 4 类损伤工况,分类提取响应.对比分析总能量变化率、频率、能量比偏差与能量比方差等不同指标对损伤程度的表征,建立能量与频率相结合的组合损伤指标.对比分析联合CNN&LSTM神经网络面对各损伤指标,以及传统中单独的卷积神经网络(CNN)与长短记忆法(LSTM)针对组合损伤指标的识别结果.研究发现,基于组合指标的联合CNN&LSTM深度学习网络的拉索损伤识别准确率最高,达到 96.67%,高于CNN的86.63%及LSTM的82.15%,表明CNN&LSTM在斜拉索损伤识别应用中具有较大潜力.
Abstract
In cable structures,the cable response and damage are in a highly nonlinear state and the accuracy is often poor in conventional mathematical models for cable damage identification.A finite element model of a seven-wire strand cable was created to address this problem,and a CNN&LSTM neural network damage identification method based on the combination of indicators was proposed.The finite element model of the cable was used to simulate four types of damage conditions and extract response features for each.Different indicators such as the rate of total energy change,frequency,energy ratio deviation,and energy ratio variance were compared and analyzed for their representation of damage severity.A composite damage indicator combining energy and frequency was established.The recognition results of the combined damage indicator using the CNN&LSTM neural network were compared and analyzed against individual indicators using the conventional convolutional neural network(CNN)and long and short term memory(LSTM)network.Results reveal that the accuracy of cable damage identification using the combined indicators with the CNN&LSTM deep learning network is the highest,reaching 96.67%,which is higher than the accuracy of CNN alone(86.63%)and LSTM alone(82.15%).The results demonstrate that CNN&LSTM have great potential in cable damage identification in cable-stayed structures.
关键词
拉索有限元/卷积神经网络/长短记忆法/损伤识别/组合指标Key words
cable finite element/convolutional neural network/long and short term memory/damage identification/composite indicators引用本文复制引用
基金项目
福建省社会发展引导性(重点)项目(2022Y0046)
垒知创新基金(202201)
出版年
2024
NETL
NSTL
万方数据