Research On Plate Structure Damage Identification Method Based On Active Lamb Wav

As a kind of metal sheet commonly used in modern industrial production,aluminum sheet has a wide range of applications in the production and manufacturing of key parts of aircraft.However,due to the production conditions and environmental impact during service,material aging and other factors,the plate will inevitably develop damage and defects,which greatly shorten the product life cycle and cause a series of safety hazards.Therefore,structural health monitoring of panels is needed to detect the location and extent of damage in the early stages of the structure and to assess its remaining life,which is important for performing timely maintenance operations and preventing structural failures.Traditional damage identification methods rely on manual selection of damage features,and there is a possibility that other features may be neglected to influence the effectiveness of damage identification.Therefore,this thesis uses convolutional neural networks for damage identification of plates,relying on the excellent feature extraction ability of convolutional neural networks to make regression prediction of plate damage location;in addition,a siamese network is designed to improve the problem of low accuracy of traditional damage probability imaging methods.In this thesis,the methodological research and experimental demonstration of aluminum plate damage recognition based on active Lamb waves are carried out as follows:(1)To address the problem that current damage localization methods mostly involve complex signal processing and the degree of generalization needs to be improved,two Convolutional Neural Network(CNN)models with the same structure are designed for damage localization of plate structures.The amplitude of each sampling point of the damage scattering signal is mapped to the corresponding grayscale value by linear grayscale conversion to obtain a grayscale image containing the damage points.The damage grayscale images are labeled with the horizontal and vertical coordinates of the damage points,and the labeled grayscale images are used as the input of two CNN models,which are designed to output the horizontal and vertical coordinates of the damage locations.It is verified through experiments that the trained models can effectively realize the damage localization of the plate structure.(2)To address the current damage imaging methods mostly select features from the damage scattering signal,which relies on expert experience and selection criteria,and has a high misclassification rate for whether the structure is damaged at a certain point.An improved damage probability imaging method based on siamese networks is proposed.The difference between the signal before and after the damage of the structure extracted by the siamese network is used as the damage factor to improve the traditional damage probability imaging method.The experimental results show that the extracted damage factor can better distinguish the damaged and healthy regions of the aluminum plate structure,and the accuracy of damage imaging is higher than that of the conventional damage factor.In addition,the influence of the arrangement and number of sensors on the damage imaging accuracy was investigated.