A damage identification method for aviation structure integrating Lamb wave and deep learning with multi-dimensional feature fusion

Abstract

With the development of aerospace industry, a more suitable structural health monitoring (SHM) method is urgently needed to solve the problem of multi-dimensional extraction and effective utilization of damage information involved in complex aviation structural sensing signals. This paper proposes a method for identifying damage in aviation structures integrating Lamb wave and deep learning with multi-dimensional feature fusion, which can effectively locate and quantify the damage in aviation structure. Firstly, the Lamb wave signal is excited and received in the real aircraft cutting section. The collected signal is processed in one-dimensional (1D) by extracting the damage index, and two-dimensional (2D) by transforming it into the Gramian angular field (GAF), respectively. Then a deep learning model with multi-dimensional feature fusion is established, which includes two neural network branches, namely 1D branch network and 2D branch network. Among them, 1D branch network includes multi-scale convolution Inception-v1 module and Bi-directional long short-term memory (BiLSTM) layer, and 2D branch network includes continuous convolution module and BiLSTM layer. The extracted 1D and 2D damage information is fused and learned to further enhance its spatial and temporal representation ability. Finally, the transfer research across geometric sensor arrays is attempted. The test results show that this method effectively locate and quantify the single-source offset damage and multi-source cumulative damage in the aircraft cutting section. Moreover, the model after transfer learning can realize damage identification on different sensor arrays with less training samples and less time, which proves that this method has significant advantages in the accuracy and robustness of damage identification.