| Glass Fiber Reinforced Polymer(GFRP)thin-walled structure refers to thin-walled structural workpieces with a total thickness of less than 10 mm,which are made by using different quantities of glass fiber layers and cured and bonded with epoxy resin.It has the advantages of good mechanical properties and light weight,and is increasingly widely used in lightweight design of aerospace engineering structures.During the manufacturing and use of GFRP thin-walled structures,it is inevitable that defects such as delamination and inclusions will occur in various epoxy resin bonding layers,seriously affecting the reliability and safety of related workpiece service work.Ultrasonic testing technology,due to its non-contact,costeffective,and easy to operate characteristics,is often used in non-destructive testing of composite materials to accurately evaluate the service condition of related workpieces,which is of great significance for ensuring the safe operation of related workpieces.However,for GFRP thin-walled structures,the resin layer between the fiber layers is very thin,with a thickness of about 0.01 mm to 0.05 mm.The resolution of conventional ultrasonic testing techniques is low,and there are difficulties in accurately detecting the quality of epoxy resin bonding between fiber layers.High frequency ultrasonic testing technology can achieve highresolution detection of the internal condition of materials by using high-frequency ultrasonic testing technology,and is usually widely used in thin-walled material structures such as metals or coatings.Based on this,this article applies high-frequency ultrasonic testing technology to the non-destructive testing of GFRP thin-walled structures.Unlike the single medium characteristics of metal and coating materials,the anisotropic structure of composite materials leads to a more complex waveform of high-frequency ultrasonic detection signals.Compared with carbon fiber composite materials,the noise phenomenon in high-frequency ultrasonic testing signals of glass fiber composite materials is more significant,the waveform is more complex,there is more structural noise and scattering noise,the degree of waveform distortion is large,the defect features are easily submerged,and feature extraction is difficult,which has a negative impact on accurately evaluating the internal bonding interface defects of GFRP thinwalled structures.Therefore,this article mainly conducts research on the processing technology of high-frequency ultrasonic testing signals for GFRP thin-walled structures,corrects the detection signals,extracts features,and identifies defect features,in order to improve the ability of using high-frequency ultrasonic testing technology to detect internal defects in GFRP thinwalled structures.The main research content includes:(1)A sound field distribution model for high-frequency ultrasonic testing of GFRP thinwalled structures was established based on CIVA simulation software.The influence of different probe parameters on the sound field distribution of high-frequency ultrasonic testing was studied,and a high-frequency ultrasonic testing echo model was established.The influence of different detection parameters on the echo characteristics of the detection signal was studied,providing theoretical basis and design guidance for practical high-frequency ultrasonic testing technology of GFRP thin-walled structures.Comparing the actual detection signal with the ideal analog signal,it was found that the noise in the actual detection signal has a significant impact,and is affected by texture noise,resulting in a series of feature distortions in the detection signal waveform,which has adverse effects on subsequent signal feature analysis and processing.Therefore,it is necessary to first remove texture noise information from the actual detection signal,correct the waveform distortion of the detection signal,which is crucial for the subsequent correct analysis and processing of the detection signal features.(2)In order to correct the waveform distortion phenomenon in high-frequency ultrasonic testing signals of GFRP thin-walled structures,it is necessary to perform image enhancement and texture information extraction on the imaging results of the interface between each layer of GFRP thin-walled structures.Firstly,to address the issue of low contrast of defect information in the amplitude imaging results of each layer interface,an image enhancement processing technology based on wavelet multiscale product is proposed.The imaging results of each layer interface are subjected to wavelet multiscale product image enhancement processing to improve the contrast of defect information in the imaging results,providing a foundation for further research on image background texture removal.Secondly,on the basis of obtaining high contrast imaging results,a high-frequency ultrasound image texture extraction technique based on relative total variation regularization is proposed to address the difficulty in extracting texture information from high-frequency ultrasound imaging results of various layer interfaces.This technique can accurately extract texture information of different sizes and directions in the imaging results without damaging defect information.Finally,based on the obtained texture image information of each layer interface,the proposed high-frequency ultrasonic detection signal correction method was used to correct the detection signals of defect free and defect free areas.The results showed that the proposed method can accurately and effectively remove the texture information from the high-frequency detection signals of GFRP thin-walled structures,and achieve the restoration and correction of the waveform features of GFRP thin-walled structure high-frequency ultrasonic detection signals.(3)After correction,there is still a large amount of nonlinear variation noise in the highfrequency ultrasonic testing signal.Due to these noises,traditional feature methods are difficult to accurately extract defect features from the high-frequency ultrasonic testing signal,which has a negative impact on evaluating the state of delamination defects in GFRP thin-walled structures.This article proposes a layered defect evaluation technique for GFRP thin-walled structures based on variational mode decomposition and recursive quantitative analysis methods,establishes an recurrence quantification analysis parameter distribution model for layered defects,and achieves accurate evaluation of layered defects.Firstly,the electronic noise in the detection signal was removed using the variational mode decomposition method,and the detection signal was preprocessed for noise reduction.Then,based on the recursive quantitative analysis method,feature extraction was carried out on the high-frequency ultrasound detection signal.By calculating and selecting appropriate recursive analysis parameters,a quantitative relationship between layered defect features and different recursive variables was established,providing a reference model for the depth and size detection and evaluation of layered defects.Finally,quantitative evaluation of the depth and size of layered defects in GFRP thin-walled structures was achieved.(4)A recognition technique based on multi feature fusion and multi kernel learning is proposed to address the difficulty in accurately identifying high-frequency ultrasonic detection signals of different types of defects in GFRP thin-walled structures.By combining various existing feature extraction techniques,a richer feature information of defect detection signals can be obtained,and the feature fusion process can be improved based on multi-core learning methods to achieve classification and recognition of different types of defect detection signals in GFRP thin-walled structures.Firstly,traditional support vector machines are used to train each single feature dataset separately,and the optimal kernel function is selected for different feature datasets.Then,different kernel functions are used to project each feature vector into a high-dimensional space to achieve multi feature fusion,and the Simple MKL multi kernel learning model is used for training,effectively reducing the redundancy of feature vectors and improving computational efficiency.Compared with traditional detection models,the method proposed in this paper has higher recognition accuracy and efficiency. |
PDF Full Text Request