Damage Identification Methods Based On Structural Response Vectors And Machine Learning Algorithms

Structural health state of infrastructure has significant effects on human society.It is of great social significance to identify structural damage efficiently and accurately.However,presented damage identification methods have their limitations.On one hand,due to the diversity of the structure itself and the complexity of the environment,it is difficult to establish an accurate finite element model.On the other hand,the accuracy of the measuring instrument cannot be guaranteed,resulting in that the measurement data are inaccurate and incomplete,which deteriorate the effectiveness of damage identification.Therefore,it is of great significance to develop new structural damage identification methods and apply them to practical bridges.Structural Response Vector(SRV)includes both static and dynamic responses of the structures,which is more sensitive to the health state of the structure.Avoiding complex calculation,machine learning methods can mine information from input directly and effectively,and obtain more reliable damage identification results on the basis of big data.Therefore,based on SRV and machine learning,this paper presents the following work:A finite element model of cable-stayed bridge is established based on the North Branch Bridge of Huangpu Bridge on the Pearl River.The damage of bridge is simulated by reduction of elastic modulus of cables.The dynamic and static responses of the cable-stayed bridge under single and multiple damage conditions are obtained.The result shows that the location and damage level of the damaged cable can be detected through the use of the vertical displacement of the main beam,and the location of the damaged cable can be estimated by the order of the extreme point of the natural frequency increment.This provides a theoretical basis for determination of damage-sensitive components for SRV.A damage identification method for simple supported beam is proposed based on SRV consisting of two natural frequencies and five vertical displacements and machine learning algorithms.The damage localization and quantification of the simply-supported steel beam model are studied.Results prove that the method has good recognition effectiveness and calculation efficiency.Principal Component Analysis(PCA)is introduced to reduce the noise of the dynamic components in SRV in order to enhance noise robustness of damage identification method based on SRV and Support Vector Machine(SVM).The method effectively solves the problem that the phase plane method data information is over-compressed and the structure response type is not sensitive enough.In order to verify the feasibility of the proposed damage identification method based onSRV and machine learning alogorithm in practical engineering,this paper studies the single damage and multiple damage identification problems of a cable-stayed bridge.Results show that this method has high recognition accuracy and a fair anti-interference ability in single damage identification of cable-stayed bridges.As for multi-damage identification,the damage localization and quantification of the method for cable-stayed bridges are studied by combining SRV and neural networks with multi-output characteristics,and considerable results are obtained in comparison with traditional damage identification method,it not only avoids the complicated calculation,but also obtains a good recognition effect on the basis of more refined response information,which is more engineering and practical.