Damage Identification of Sandwich Panel with Truss Core by Integrating Deep Learning Technique

Abstract

This paper reports some preliminary work in IMECH, CAS on integrating deep learning technique to identify internal damages in the sandwich panel with truss core. Cell-missing and unbound nodes are two main types of damages to be concerned. Numerical models of panels with various damage features are presented to provide original structural dynamic responses, and a vibration-based damage index is proposed to extract damage features, therefore image datasets of randomly damaged samples are constructed for training and testing. The training and testing processes are conducted by Faster R-CNN (regions with convolutional neural network). Results reveal that the proposed method can identify damage location and extent in reasonable precision.