Damage Identification on Aircraft Wing Using Convolutional Neural Network Based Pattern Recognition

Abstract

Damages to the aircraft body must be identified in order to avoid unforeseen events that could result in catastrophes if not examined properly. As most of these damages are provoked under normal working conditions, such as in this case during aircraft flight operations, rectifying these damages is nearly impractical. This work veils through a compiled report of Structural Health Monitoring (SHM) of an aircraft wing aiming at the deterioration caused throughout the entire structure of the wing which is subjected to numerous aerodynamics and mechanical constraints. The SHM report provides data which is used to study intramural Damages. Based on the investigation, the Convolutional Neural network (CNN), which is a typical algorithm of deep learning along with Recurrence graph, was chosen as the appropriate Methodology. The structural response recurrence graph can reveal details about the damage, similarity, and underlying structure. Wavelet packets are used to filter and recreate the initial structural response signal, considering the vibrational coupling between the aircraft and wing. Then, as a novel kind of damage feature, the recurrence graph of distinct damages is taken as input picture of the CNN. The findings show that this method provides added information about damage than traditional statistical pattern recognition techniques. These outcomes confirmed the proposed CNN-based model’s suitability for use with additional aircraft components.