Research On Damage Detection Of CFRP Laminates Based On Electrical Impedance Tomography

Carbon Fiber Reinforced Polymer(CFRP)is widely used in the field of aeronautics and astronautics.It is of great significance to realize its structural health monitoring.Electrical Impedance Tomography(EIT),as a new non-destructive testing method of CFRP,has the advantages of low cost,no radiation,visualization and so on,which has been widely concerned by researchers.Therefore,the application of EIT method in the damage detection of CFRP laminates has become one of the hot spots of scholars at home and abroad.In this project,EIT method is used to detect the damage of CFRP laminats.(1)The EIT inverse problem is ill posed,and the regularization algorithm is usually used to improve the image quality.Modified Residual Norm Steepest Descent(MRNSD)algorithm is used to improve the reconstruction image quality because of the advantages of MRNSD algorithm in reducing image artifacts and maintaining boundary information.In order to solve the problems of semi convergence and poor anti noise effect of mrnsd algorithm,the pretreatment and soft closed value method are used to improve mrnsd algorithm.Through simulation and experiment,the imaging effects of the improved algorithm and several common algorithms are compared.The results show that the algorithm effectively improves the image quality and anti noise ability of EIT,and realizes the automatic update of the optimal number of iterations,which is conducive to promoting the application of EIT method in CFRP damage detection.(2)Based on the electrical sensitivity of CFRP laminates,an improved Deep Separable Convolutional Neural Network(DSCNN)instead of the traditional convolutional network algorithm is proposed to build EIT-DSCNN damage identification model for CFRP laminates,which is verified by simulation and experiment.The results show that the model can effectively identify the damage types of CFRP laminates,which is more accurate than the traditional method.The calculation amount of the model and the calculation speed can be effectively reduced by the depth separable one-dimensional convolution layer instead of the traditional convolution layer.The damage identificationability of the model under noise conditions can also be improved by using the improved resampling extended data training strategy.