Research On Damage Detection Method Based On Guided Wave Field Images Intelligent Recognition

Structural damage detection is critical to maintain the normal use of engineering structures and civil facilities.Machine learning algorithms and deep learning algorithms are applied to damage detection.While most methods based on machine learning are manual feature extraction,deep learning technology adopts automatic feature extraction.The nonlinear mapping relationship between the input image or time series data containing damage information and the output damage detection results is established,so as to quickly and accurately identify the damage.The local wave number of steady-state guided wave field of aluminum plate is larger because of the existence of blind holes,in this paper,a deep learning image recognition method is proposed to study the steady-state and transient guided wave fields of structures under high frequency excitation.The problem of damage detection is transformed into the problem of local feature target recognition in the image,and the blind hole damage detection of aluminum plate and the debonding damage detection of carbon fiber reinforced plastic(CFRP)and isotropic stiffened plate are recognized.The main research contents of this paper are summarized as follows:Firstly,the damage detection theory of Lamb wave is introduced and the dispersion equation of Lamb wave propagation in isotropic plate structure is analyzed.Meanwhile,the dispersion curve of Lamb wave propagation in plate structure is drawn.Compared with processing complex guided wave echo signal,it is more convenient to detect damage directly by generating steadystate guided wave field images.On this basis,the image recognition capability of deep learning is used to extract the damage features of the steady-state and transient guided wave fields.The YOLOv5(You Only Look Once v5)network model is selected as the deep learning algorithm in this paper,and the principle and construction of the algorithm are described in detail.Secondly,in order to apply the YOLOv5 network model to the blind hole damage detection of aluminum plate,the finite element aluminum plate model is established to generate the simulation datasets.The damages at different locations are traversed to construct enough sample images.Three sinusoidal signals of different frequencies are used to excite the structure,and the steady-state guided wave field sample images of aluminum plate structures are generated in batches.In order to improve the effect of network training,RGB three-channel fusion is performed on guided wave field images with three frequencies excitation,and data enhancement and programmed annotation are carried out on the images.The appropriate network parameters are determined and the network model was trained.Meanwhile,the non-contact guided wave field measurement experimental platform is built.The trained model is used to simulate and experiment the steady-state and transient guided wave field detection of aluminum plate,and the blind hole damage detection results are analyzed.Finally,guided wave field images of aluminum plate with blind hole damage under single frequency excitation are taken as samples,and some network parameters are changed and trained.The results of network training and evaluation index show that the network model can realize the high-precision detection of the debonding damage of two different laminated CFRP and the debonding damages of isotropic stiffened plates by training the guided wave field images of aluminum plates,which further verifies the effectiveness and feasibility of the proposed method.The structural steady-state guided wave field damage detection method proposed in this paper based on deep learning image recognition can realize the identification of blind hole damage of aluminum plate,debonding damages of stiffened plate and debonding damages of CFRP.Moreover,the method has high detection accuracy and speed,and has high application value and potential in the field of intelligent damage detection and structural health monitoring.