Super-resolution Imaging Of Debonding Damage In Laminated Plate Structures Via Ultrasonic Guided Wave And Deep Learning

Laminate bonded structures are important component of large construction and transportation vehicles.The double-layered steel-rubber laminated structures are widely used in various engineering fields in society due to their excellent mechanical,thermal properties as well as high sound absorption capacity.The bonded quality is one of the most important factors to maintain the high safety and stability of the laminate bonded structure in service.It is important to regularly check the health condition of laminated structures,accurately identify debonding damage and take timely remedial measures.In the field of nondestructive testing and structural health monitoring,ultrasonic guided wave array imaging is an effective damage detection method.Through imaging results damage can be located and identified,while the multidimensional information such as the shape and extent of the damage can be characterized.However,due to the limit of diffraction,it is difficult to accurately identify damage whose size is smaller than one wavelength scale.In recent years,semantic segmentation techniques via deep learning have developed rapidly,and multiple kinds of deep neural networks have been used to solve the problem of complex image feature recognition and achieve super-resolution imaging of tiny features in the region of interest.To address the need for fast and accurate detection of damage location,size and shape in laminated structures by ultrasonic guided wave,this paper combines the total focusing method with a deep learning model to develop a super-resolution imaging method for subwavelength debonding damage via ultrasonic guided wave and deep learning.The main research contents and innovation points of the paper are as follows.Firstly,the boundary conditions and dispersion equations of the double-layered laminate structure in an isotropic medium are analyzed and derived by combining partial wave theory with the global matrix method.The dispersion curves of the steel-rubber laminate are also plotted in this way.On this basis,the dispersion curves of the stiffened plate and the single-layer plate are plotted,while the excitation frequencies for the simulation and the corresponding experiments are determined.Further,based on the comparative analysis of damage imaging algorithms,the total focusing method is selected as the basic imaging algorithm in this paper and its principles are briefly outlined and derived.Secondly,in order to achieve super-resolution imaging of subwavelength damage through deep learning algorithm models,a "code-decode layer" is built as the main framework,embedded with "multi-scale scattering signal capture and recovery","nonlinear enhancement" and "nonlinear enhancement".Network parameters such as loss functions and corresponding network performance evaluation metrics are investigated and determined in order to improve network training accuracy and reduce the risk of errors.Further,numerical simulations are carried out for laminated,stiffened and single-layer plates,while the corresponding simulation models and damage scenarios for the three structures are analyzed and presented.On this basis,the signal data of the guided wave array is acquired and imaged by a full-focus imaging algorithm.Experiments on stiffened and single-layered plates are carried out under existing experimental conditions,and the corresponding experimental data are acquired and imaged.The imaging data are expanded and a database is created by appropriate image enhancement.Finally,the analysis of network training results and the feasibility of the method are investigated for the proposed super-resolution imaging work via deep learning.The data are divided into different datasets by type while corresponding training and detection scenarios are developed.The training is carried out for the different training scenarios.The training results and performance indicators show that the network structure built in this paper has a high training accuracy and a high detection accuracy for damage/debonding damage of different materials and structures,which initially verifies the excellence and suitability of the network model.Then,the array signals were added with different decibel signal-to-noise ratios.With that,the imaging results of total focusing method,time-reversal multi-signal classification technique and network test imaging are analyzed and compared.The results show that the imaging accuracy of the method is better than that of conventional imaging methods.It is less affected by noise,which has high imaging stability.Further,the network detection process is visualized.The results clearly present the process of capturing and recovering the detailed features of the damage.The feasibility of the proposed method is also demonstrated objectively.Last but not least,a simple extended performance test of the proposed method is carried out.The damage super-resolution imaging network structure via ultrasonic guided wave and deep learning proposed in this paper can achieve super-resolution imaging of sub-wavelength damage/debonding damage in laminates and other structures.The method has high detection efficiency and accuracy,high robustness to noise and high suitability.It has high potential and value in the field of non-destructive testing and structural health monitoring via ultrasonic guided wave.