Guided Wave Data-driven Damage Detection Method For Plate-like Structures

Facing the complex structure and large-scale engineering background of high-end equipment components in China,it is essential to develop a rapid,high-precision and intelligent damage detection method based on the existing damage detection technology in order to solve the safety problem of high-end equipment.In this paper,an intelligent damage detection method based on convolution neural network is proposed,which can extract features of guided waves automatically.The damage detection area is divided,and the damage detection turns into the image classification task based in convolution neural network.Convolution neural network is used to extract the feature relationship between different damage locations and signals.On the basis of damage location recognition,the damage degree is detected and classified.The main work of this paper is summarized as follows(1)Based on the Lamb wave theory,the frequency wavenumber relationship of guided waves in isotropic plate structure is obtained,which provides the theoretical foundation for the selection of excitation frequency.The numerical model of ultrasonic Lamb wave excitation,propagation and interaction with damage in isotropic plate is established by finite element method.In this research,different sizes and shapes of through holes are set up to extract ultrasonic guided wave signals.Combined with the experimental ultrasonic guided wave signals,a deep learning training database for damage identification is constructed.(2)The functions of convolution layer,pooling layer and full connection layer in the basic framework of convolution neural network are studied.The advantages and disadvantages of activation function,loss function and optimization function in convolution neural network are compared.Based on this theory,a convolution neural network model is established to train guided wave data for damage detection.Through the difference between the reference signal and the damage signal and visualization as the input of convolutional neural network,the model is trained to realize the damage detection at different positions,and the generalization ability of the model is tested at different noise levels.Besides,circular through holes with different diameters are placed in the same damage area to produce different signals.The model is trained with those signals to realize the classification of damage size.(3)The experimental platform of isotropic aluminium is set up.The detection area and signal scanning point layout are the same as the numerical simulation.After picking up the signal,the experimental database is established by data preprocessing,and the experimental data and the simulated data set are fused to construct the fusion database.The trained convolution neural network model is transferred to the fusion database.The generalization ability of the model is verified on new data,different quadrants and different noise levels.By segmenting the detection area,the damage recognition is transformed into extraction of feature and classification of signal image.The convolution neural network is used to extract the damage features automatically,which greatly shortens the time of damage identification.(4)The damage intelligent detection method proposed in this paper is compared with the traditional damage detection method.The classical Delay and Sum algorithm is used to image in time domain and frequency wave number domain respectively.Through the analysis of detection time and accuracy,the advantages of the damage identification method in this paper are verified.The method of intelligent damage detection based on convolution neural network proposed in this paper can realize the rapid detection and evaluation of damage.The relevant results show that the deep learning method based on convolution neural network is applied to damage identification.It has the potential engineering application in the field of fast and intelligent damage detection.