A multi-level damage classification technique of aircraft plate structures using Lamb wave-based deep transfer learning network

Lamb wave-based damage detection is one of the most promising structural health monitoring (SHM) technologies for aircraft structures. In this paper, a Lamb wave-based deep transfer learning network is developed for multi-level damage classification of plate-type structures. A one-dimensional convolutional neural network (1D-CNN) is employed to deeply mine the damage characteristics of complex Lamb wave signals with multiple modes and multiple boundary reflections. The concept of multi-level damage classification is carried out to get different results for different engineers, and a multi-task cascaded 1D-CNN architecture is established for three levels of damage classifications, which is corresponding to different SHM levels, i.e. identifying the damage presence, location, and severity, respectively. In the multi-task cascaded architecture, a fine-tune transfer learning concept is adopted to share partial structures and weight values among different classification models, which could greatly improve the efficiency of the model calculation. In the multi-level damage classification model, the one-dimensional Lamb wave scattering signals with different damage locations and sizes are used as the input without any preprocessing steps, while the classifications of the damage presence, location and size are designated as output of different levels. An experiment has been conducted to verify the proposed multi-level damage classification model. The experimental results show that the training time of this model is reduced by 35% and the accuracy of the proposed model is greater than 99%, which verifies the effectiveness and reliability of the proposed technique.