Advanced structural health monitoring in carbon fiber-reinforced plastic using real-time self-sensing data and convolutional neural network architectures

Abstract

In this study, advanced structural health monitoring (SHM) using a non-destructive self-sensing methodology was proposed for large-sized carbon fiber-reinforced plastic (CFRP). Cyclic point bending tests were performed on three types of CFRPs. The damage severity identification and localization were classified and investigated using four different convolutional neural network (CNN) architectures. Electrical resistance images were used to train each CNN architecture for damage analysis. An optimized CNN architecture for the damage analysis of CFRPs using electrical resistance data was proposed and compared with traditional damage analysis CNN architectures. The applicability of the proposed SHM methodology was verified by analyzing unseen damage in the CFRPs. This study addresses the limitations of previous self-sensing methods by reducing the number of electrodes, which reduces data complexity and increases the sensible area of CFRPs. Thus, this study successfully designed an efficient SHM methodology with a high accuracy of over 90 % for analyzing CFRP damage, including the severity and location, regardless of the type of carbon fiber and stacking sequence of composite structures that showed high applicability.