Research On Laser Ultrasonic Imaging Of Component Damage Based On Deep Learning

With the fast development of modern industrial technology,metal materials are widely used in various fields with their excellent mechanical properties.Due to the influence of high temperature,high pressure,high load and other external environment in the production and service process of metal materials,small defects are easily generated on the surface or inside of metal components.If the damage cannot be identified in time and effective measures are not taken,the failure of metal components during operation will cause major safety accidents in the case of repeated loading.Therefore,the regular detection and evaluation of metal materials are particularly important.As one of the commonly used technologies in the field of non-destructive testing,laser ultrasonic imaging technology has long-distance,non-contact,high-precision and high spatial resolution.It can also overcome the impact of harsh environments such as high temperature and pressure,corrosion,and high radiation.In this paper,the method of experimental verification is used to deeply study the interaction between laser-excited ultrasound and component damage,combined with signal processing and convolutional neural network optimization algorithm to improve the imaging accuracy of the defect edge area.The main research contents of this paper are as follows:(1)Introduced the basic theory of photoacoustic imaging based on deep learning.According to the excitation principle of laser ultrasonic Lamb wave and its propagation characteristics in aluminum plate,several commonly used detection and processing methods of laser ultrasonic Lamb wave are explored,and briefly described the basic theory of deep learning,which provides the basis for the signal analysis and research in the following text.(2)Developed a set of fully automatic laser ultrasound imaging system.Starting from the practical application of industrial engineering,a set of fully automatic laser imaging system integrating software and hardware combination,scanning control,data acquisition and analysis has been developed,which realizes large-area movable excitation and high-efficiency reception of Lamb waves.Finally,the integration test verifies the feasibility of the system.(3)Carried out experimental verification of laser ultrasound imaging technology.Design experiments and use the Acoustic Emission(AE)sensor at a fixed position to obtain the ultrasonic Lamb wave signal in the typical metal aluminum plate,and use the principle of acoustic reciprocity to realize the wave field replay of the Lamb wave in the test piece.By analyzing the Lamb wave signal in the aluminum plate,the A0 mode of the Lamb wave in the signal is successfully separated,and the amplitude Amax of the A0 mode is extracted to complete the visual imaging of the damaged area.(4)Proposed deep learning convolutional neural network algorithm to optimize the boundary of the damaged area.Aiming at the problem of fuzzy edges of defects in the existing time-domain processing methods,a convolutional neural network is used to establish an optimization model for the boundary of damage area.The data set is expanded by adding noise,and the training set,verification set and test set are randomly allocated.The training results show that the proposed damage area optimization edge algorithm is reliable and feasible.