Research On Material Defect Detection Method Based On Laser Thermography

Metal materials and composite materials are widely used in industrial production life.For aircraft blades,the main problems faced are old aircraft metal blades and new aircraft advanced composite materials and other blade breakage repair problems.The accurate nondestructive detection of surface and internal defects of metal and composite materials is of great value for the assessment of the health condition of the workpiece.Laser thermography-based inspection technology with energy concentration and non-contact characteristics has been widely used in defect detection of materials.However,at present,laser thermography in the process of material inspection is mostly carried out in a point-by-point sweeping manner,which is very timeconsuming,and cannot continuously judge the defect information and lose the defect information,which greatly reduces the inspection efficiency and accuracy;in addition,there is no effective means to quantify the material defects to analyze the defect shape characterization and depth identification.Therefore,an efficient defect detection method for typical defects of metals and composites needs to be studied.By comparing the advantages and disadvantages of different material defect detection techniques,the research work is carried out on the construction of a thermal imaging system for the detection of defects on the material surface and internal layering,the material defect shape characterization and the defect depth identification around the defect detection of metals and composites.The main aspects are as follows.(1)For the detection principle of laser thermography,the detection methods based on pointshaped laser pulse excitation and line laser continuous scanning excitation are proposed to analyze the thermographic defect detection of metals and composites.Combined with the finite element analysis method,the heat conduction simulation model of metals and multi-directional layered composites were established by using COMSOL simulation software.Through simulation analysis,the effects of different physical characteristics of defects and scanning speed on the detection effect of laser thermographic defect detection were explored,showing the theoretical feasibility of the method and laying a theoretical foundation for laser thermographic material detection.(2)A transmission and reflection infrared thermal imaging inspection system was built.The design of a linear continuous laser scanning system was realized by shaping the laser spot.The effects of different defect types on thermal wave transmission at different excitation modes were analyzed.System parameters such as different linear laser sweep speed and heat source power are explored optimally to achieve optimal detection of typical defects in metals and composites.The effects of different depths of cracks,impact damage and delamination defects on metal and composite surfaces are analyzed and compared using transmission and reflection laser defect detection systems,while impact damage and surface scratches on real aero-engine turbine blades are detected.(3)To address the problem that the traditional shape characterization algorithm based on first-order derivatives is not effective in defect shape characterization,a characterization method based on principal component analysis is proposed to characterize the shapes of typical defects in several materials.The algorithm can not only accurately characterize the surface damage of metals and multi-layered composites,but also achieve certain effect of internal layered defect characterization.Both the accuracy of shape characterization and the characterization of internal defects are improved.Finally,the shape characterization method based on principal component analysis achieves the shape characterization of turbine blade surface scratches.(4)In order to improve the recognition accuracy of composite material defect depth,the machine learning algorithm is applied to the metal and composite material defect depth recognition,and the hyper-parameters in the machine learning are optimized by using the gray wolf optimization algorithm in combination with the hyper-parameter search technology;the impact damage categories of four different depths are recognized with an accuracy of 97%.Comparing with random search and grid search,this method has the best discrimination accuracy;for the internal layered burial depth recognition problem,the Bayesian optimization method is used The hyperparameters in the support vector machine were optimized and combined with the experimental data of stratified defects to identify the stratified defects of five different burial depths with an accuracy of 98%.(5)For metal crack depth recognition,in feature engineering,robust CTF and TDF defect features are extracted,and the temperature information of the two defect features is used as the input of the machine learning classifier to achieve the depth recognition of metal surface cracks;in order to improve the accuracy of metal surface crack depth recognition,a crack depth recognition method based on improved neural network architecture search is proposed,and by using NASCell cells as the RNN controller to search the multilayer perceptron model structure,combined with metal surface crack experimental data,seven surface crack depth categories are recognized with an accuracy of 100%.