PCA-Res2Net Model-Based Method for Damage Detection of CFRP Using Electrical Impedance Tomography

Electrical impedance tomography (EIT) has become a new inspection tool for damage detection in carbon fiber reinforced polymer (CFRP) composites. However, the inverse problem of EIT is severely nonlinear, ill-posed, and underdetermined, limiting the resolution and accuracy of images reconstructed with EIT. To solve the above problems, a PCA-Res2Net algorithm for CFRP damage detection based on Res2Net structure is presented, which consists of three modules: initial imaging, data dimension reduction, and deep feature extraction. First, L1 regularization algorithm is used to map the voltage measurement value to the conductivity distribution for preliminary damage imaging; then, principal component analysis (PCA) is used to perform feature analysis on the conductivity distribution data for removing redundant background information and achieving data dimensionality reduction; finally, the simplified image data are inputted into Res2Net for deep feature extraction. Three traditional EIT image reconstruction algorithms (Tikhonov, Conjugate Gradient, L1) and a deep learning algorithm [Invertible Neural Networks (INNs)] are compared and analyzed with the PCA-Res2Net algorithm. Simulation results demonstrated that the PCA-Res2Net algorithm yields more satisfying reconstructions, and both its correlation coefficient (CC) and structure similarity index measure (SSIM) exceed 97%. Compared with INN, the SSIM and CC of PCA-Res2Net achieved maximum improvements of 8.87% and 19.76%, respectively. To further verify the feasibility of the proposed method, a 16-electrode EIT experimental platform is built to detect the damage samples of CFRP laminates. Experimental results demonstrated that the PCA-Res2Net model can effectively reduce the artifacts of the reconstructed images and improve the damage recognition accuracy and edge clarity.