High deformation/damage localization accuracy of fibrous composites through deep-learning of single channel data from carbon nanotube sensors

Abstract

A convolutional neural network (CNN) model by deep-learning single channel data from a serpentine carbon nanotube sensor (S-CNT) with gradient distributed CNTs is proposed for locating deformation/damage in carbon fiber reinforced plastic (CFRP). The real-time resistance-time data caused by bending deformation of CFRP embedded with S-CNT are encoded into more discriminative 2D images for training the CNN. The results show that an accurate deformation localization within 1.5 mm for the trained positions can be obtained. Moreover, static-indentation loading reveals that the CNN model also has high localization accuracy for new deformation/damage locations in CFRP, with an error of less than 5.5 mm.