CNN-Seq2Seq Model-Based Method for Damage Detection of CFRP Using Electrical Impedance Tomography

Electrical Impedance Tomography (EIT) has received widespread attention in the field of Carbon Fiber Reinforced Polymer (CFRP) damage detection due to its advantages of low cost, non-invasiveness, and simple structure, but its limited image reconstruction quality has been restricting the further development and application of EIT. Aiming at the nonlinear characteristics of the EIT image reconstruction problem, based on the Convolutional Neural Network (CNN), the Sequence-to-Sequence (Seq2Seq) model is introduced and a CNN-Seq2seq model for CFRP laminate damage detection is proposed. The proposed model incorporates CNN as the encoder and Seq2Seq model as the decoder, it leverages the formidable nonlinear fitting capabilities of CNN to precisely forecast the conductivity distribution within CFRP laminates based on the measured voltage sequence. Simulation results indicate that the CNN-Seq2Seq model is capable of significantly enhancing the quality of reconstructed images for typical damage models. The achieved Correlation Coefficient (CC) and Structure Similarity Index Measure (SSIM) values for these images are both greater than 0.7. To further validate the feasibility of CNN-Seq2seq model, in this study, a prototype experimental study is conducted on CFRP laminate damaged samples using a 16-electrode EIT experimental platform. The experimental results show that compared with CNN, the CNN-Seq2Seq model effectively reduces the impact of artifacts on image quality and reconstructs the image with more complete edge information. By testing practical damage and tiny damage samples, combined with analyzing the performance of the CNN-Seq2Seq model, the superior computational, convergence, and generalization abilities of the CNN-Seq2Seq model compared to CNN are demonstrated. The training time of the CNN-Seq2Seq model has been reduced by 5.43%, and the CPU occupancy rate has been lowered by 10.20%.