利用深度学习与数据融合的结构损伤识别方法  

Structural damage identification method using deep learning and data fusion

作　　者：李晨阳 刘浩铭 周博文 LI Chenyang;LIU Haoming;ZHOU Bowen(College of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学机械与动力工程学院,郑州450001

出　　处：《重庆理工大学学报(自然科学)》2025年第1期155-162,共8页Journal of Chongqing University of Technology:Natural Science

基　　金：船舰设备噪声与振动控制技术国防重点学科实验室开放基金课题(VSN201902)。

摘　　要：为实现高准确率、智能化的结构损伤识别,将模态频率应变能基指标以及曲率模态差作为损伤识别评价指标,并利用卷积神经网络将二者融合进行损伤识别。以简支梁为实验对象,通过ANSYS仿真模拟不同损伤工况,提取特征数据进行训练。仿真实验包括单位置损伤、多位置损伤以及多程度损伤的工况。这2种指标均能够对损伤位置以及损伤程度进行有效识别,并且将2种信号进行特征级融合时,损伤识别的准确率有了进一步提高。以模态频率应变能基指标和曲率模态差为损伤识别指标,利用深度学习和数据融合方法为结构健康监测提供了一种有效的新途径。To achieve high-precision and intelligent structural damage identification,the modal frequency strain energy basis index and curvature modal difference are empoloyed as evaluation indicators for damage identification,and convolutional neural networks are used to fuse the two for damage identification.With a simply supported beam as the experimental object,different damage conditions are simulated through ANSYS simulation.Feature data are extracted for training.Simulation experiments include working conditions with unit damage,multi position damage,and multi degree damage.These indicators accurately identify damage.When the two signals are fused at the feature level,the accuracy of damage identification is further improved.Our results show deep learning and data fusion methods are effective for structural health monitoring by using modal frequency strain energy basis index and curvature modal difference as damage identification indicators.

关 键 词：模态频率应变能基指标 曲率模态差 数据融合 仿真模拟 

分 类 号：TH113.1[机械工程—机械设计及理论]