Research on algorithm for improving imaging accuracy of CFRP low speed impact damage
Carbon fiber reinforced polymer(CFRP)composites has small and hidden damage after low-speed impact,and the existence of damage significantly reduces the bearing capacity and service life of CFRP materials.C-scan represents a conventional ultrasonic imaging method.To address the issue of low imaging precision in C-scan detection of internal damage caused by low-velocity impact in CFRP,gradient operators were employed to process the original images,and transfer learning methodology was utilized to conduct damage classification training on ResNet18 and ResNet50 architectures.To enhance the classification model's performance,an image reconstruction model(IRM)based on convolutional neural networks was proposed to improve imaging precision.Additionally,a performance metric σEOL,based on the structural similarity index(SSIM),was introduced to validate the level of image quality enhancement.The iterative training results demonstrate that when the iteration count reaches 200,the σEOL of different types of impact damage is greater than 1.To further improve imaging precision,the ResNet residual connection concept is incorporated,leading to the development of the ResIRM network.Compared to IRM,ResIRM exhibits enhanced detection precision for different types of impact damage,with an average σEOL improvement of 0.85%across all impact types.Furthermore,the gradient saliency heat maps of the classification model processed by ResIRM indicate that ResIRM effectively reinforces the features in damaged regions.
国家科技期刊平台
NSTL
万方数据
