Multi-Task Damage Identification Method for Composite Stiffened Plate Based on Lamb Wave and Multi-Task Deep Learning

Abstract

With the rapid growth of the demand for structural health monitoring in the aerospace field, especially under complex and changeable structural and damage cases, damage detection methods are facing the challenges of insufficient detection accuracy. To solve this problem, this paper proposes a detection method of damage location of composite stiffened plates based on Lamb wave and multi-task deep learning. In this paper, experiments are carried out on a composite stiffened plate with two stiffeners. Firstly, the damage monitoring area is divided into four sub-areas, each sub-area is further subdivided into nine grids, and five damage signals with different damage sizes are collected at the center of each grid. Through the differential operation between the baseline signal and the damaged signal, the damage scattering signal is extracted as the input of the deep learning model. The model training is only based on the data of the first sub-area, and its generalization ability is evaluated through two test sets: one test set contains the random signals of the non-center location of the first sub-area, and the other test set comes from the random signals of the other three sub-areas. Then, the Inception module and BiLSTM is used to extract the multi-level and multi-scale features of the damage signal, which effectively captures the features of the signal, thus significantly improving the accuracy of damage location. Finally, two output branches are designed to realize the multi-task learning target model, which can predict the x and y coordinates of the damage location at the same time. The experimental results show that this method can accurately identify the damage at any location in the monitoring area. In the tests of different sub-areas, the model still shows good robustness and generalization ability, showing a certain application potential.