Machine learning for Lamb wavebased delamination detection in composite structures

Abstract

Composite materials are increasingly used in the aerospace industry due to their high strength-to-weight ratio and favourable mechanical properties. However, composite structures are susceptible to complex failure mechanisms such as delamination, while conventional inspection and maintenance procedures are often expensive and time consuming. Structural health monitoring based on guided Lamb waves has therefore emerged as a promising approach for automated damage detection. This thesis investigates a methodology for delamination detection and localisation in composite plates by combining finite element simulations of guided Lamb waves with machine learning. Finite element simulations were performed to model wave propagation in undamaged and delaminated plates, generating a dataset used to train and evaluate a one-dimensional convolutional neural network (CNN). Three plate sizes were considered, and the influence of noise level and training dataset size was examined. For delamination detection, the proposed CNN achieved classification accuracies of (96.4%±7.0%), (99.4%±0.5%), and (78.0%±23.0%) on the 100mm, 250mm, and 500mm plates, respectively, at a signal-to-noise ratio of 20 dB. For localisation, the corresponding mean squared errors were (0.00191 ± 0.00092), (0.00232 ± 0.00043), and (0.01068±0.00240). The results demonstrate that finite element simulations can be used to generate datasets suitable for machine learning-based structural health monitoring, and that the proposed CNN can accurately detect and localise delaminations under favourable noise and training data conditions