ACTIVE DEEP LEARNING-BASED CORROSION DAMAGE DETECTION IN AIRCRAFT STRUCTURES

Abstract

Lamb wave-based damage detection has been demonstrated to be an efficacious method for structural health monitoring (SHM) in general, and corrosion in particular, and is thus deployed in this study. Since a large amount of data is needed for the deep learning networks, this study relies heavily on simulations as the data source and the waveforms are thus generated using simulations. The propagation of the Lamb waves is determined by finite element analysis which is carried out using ABAQUS. The signal features are extracted using continuous wavelet transform for amplitude change observation for presence and extent of the damage. One of the key aspects this paper focuses on is the application of the SHM methodology proposed here for realistic dimensions of corrosion pits. Thus, damage sizes are considered which fall in the range of pitting corrosion morphologies. Simulations are carried out with idealized corrosion pits of varying depths. Methods based on Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN) are used for the inverse problem solution to find the damage parameters and are compared with the numerical results. The results show much promise and could be a viable means of detecting corrosion in aircraft structures.