Research On Structural Damage Detection Method Under Finite Space Samples

It’s important for preventing serious structural deterioration that determining whether damage occurs,damage location and degree.However,in the actual damage detection process,the following problems exist: First,in the process of damage detection,the test data may be insufficient due to insufficient number of sensors,difficult measurement methods,or complicated environment,that is,a small sample problem may occur;Second,many structural damage detection methods usually require in-depth analysis by measuring external excitation information,but in many cases,the excitation information is difficult to be measured accurately;Third,many methods can only solve a simple binary preliminary problem,that is,whether there is damage and preliminary damage location.However there is little research on the evaluation of the damage degree and the multi-classification judgment of the damage type.In order to solve the small sample problem existing in structural damage detection,this paper combines the transmissibility function and finite element simulation to expand spatial samples.Firstly,the actual structure is simulated.Then information of the measured point in the actual situation is reversed by calculating the transmissibility function of the relevant measuring point in the simulation condition to expand the vibration response data that cannot be obtained in the actual process.Finally,experiment validated the correctness of the method.In structural damage detection,many methods can’t get rid of the dependence on excitation measurement,this paper combines the transmissibility function and principal component analysis to discriminate the damage and evaluate the degree of damage.Firstly,the transmissibility function between the measuring points is obtained,and then the principal component analysis method can be used to reduce the data of the transmissibility function.Finally the scatter plot is obtained to reflect the structural damage,and the degree of the damage is assessed by the damage indicator PCAC.In this paper,the designed experiment validated the correctness of this idea.The current structural damage detection method can only solve the binary classification problem.Based on the classical convolutional neural network models such as LeNet-5,AlexNet and VGGNet,this paper proposes a method of convolutional neural network model suitable for damage type detection.By simulating different thicknesses and different elastic modulus as multiple damage types and using the transmissibility function under different damage types as input,the network was trained,and the recognition accuracy of the model was verified by simulation data.