Structural Damage Identification And Localization Based On Lamb Waves And Deep Learning

Plate structure is the most basic and extensive component of engineering equipment.In order to ensure the safe and healthy service of plate structure in equipment and avoid accidents caused by damage,it is of great significance to carry out structural health detection.Among many detection means,the nondestructive testing technology based on ultrasonic Lamb wave has been widely used in damage detection of plate structures due to its advantages of strong penetration,high sensitivity to damage,and ability to provide more comprehensive damage location and degree after medium reflection.In this paper,based on practical engineering problems,aiming at the problem that the traditional damage identification algorithm requires a large number of sensors and so on,the ultrasonic Lamb wave propagation theory,combined with finite element simulation and experiment,is used to carry out the Lamb wavebased damage identification and positioning research of plate structure,and the convolutional neural network and fusion localization algorithm are designed to realize the identification and positioning of defects.It has the characteristics of intelligence,high precision and high efficiency.This paper firstly introduces the research status of ultrasonic Lamb wave nondestructive testing technology,including the development of traditional ultrasonic Lamb wave damage recognition algorithm and damage recognition technology based on deep learning.By means of theoretical derivation,combining elastic dynamics and particle motion displacement,two propagation modes of Lamb waves in the plate are studied,and the expression formulas of symmetric mode and antisymmetric mode are obtained,according to which the group velocity and phase velocity of Lamb waves are obtained,and the dispersion curve of aluminum alloy plate is drawn by MATLAB programming.In this paper,piezoelectric sensors and ultrasonic Lamb wave excitation and reception are introduced.Finite element simulation is used to show the propagation process of Lamb wave in aluminum plate,and then convolutional neural network in deep learning is briefly introduced.Then,the experimental platform of ultrasonic guided wave data acquisition for fatigue crack growth was built,and the one-dimensional deep convolutional neural network was designed to train the collected data.The results were compared with the traditional classification network.The results show that the modified network model has a very good recognition effect on the guided wave signal of fatigue crack,and the actual detection accuracy can reach 98%.Then,the data of small damage on aluminum plate was collected,and the convolutional neural network was optimized by the introduction of Bayesian theory to enhance the expression ability of the network for uncertainty.The validity of the model was verified by comparing the convolutional neural network with data.The results show that the modified algorithm has advantages in recognition of small damage compared with the convolutional neural network.Finally,the damage location on aluminum plate is studied.In view of the limitations of traditional methods,such as large number of sensors and complex signal processing means,the elliptic positioning and RAPID fusion algorithms are proposed to study the damage on aluminum plate.Through the network layout with different number of sensors,the damage of some sensors in the network and multiple damages in aluminum plate are taken into account.Effective positioning of different locations of single damage and multiple damage is realized,with positioning error of about 3mm.Considering that the sensor part is damaged,positioning and imaging can be effectively realized with positioning error of 5.83 mm,showing good robustness.