Damage quantification of composite laminate via variational auto-encoder Lamb wave data expansion and spatio-temporal fusion transfer model

Composite materials are widely used in aerospace industry, and its structural health monitoring (SHM) is one of the key research directions in this field. With the development of artificial intelligence, it has turned into a widely discussed research subject to analyze and process Lamb wave response signals by using deep learning method to determine the structural integrity. This study presents a method for quantifying damage in composite laminate via variational auto-encoder (VAE) and spatio-temporal fusion transfer model is proposed. First, VAE model is established to realize the reconstruction and expansion of Lamb wave obtained through experiment, which solves the problem of scarce data in SHM methods based on Lamb wave and machine learning. Second, 1DCNN and multibranch ConvBiLSTM (MB-ConvBiLSTM) spatio-temporal fusion networks are designed to quantify the size and location of damage, respectively. Finally, the monitoring area is divided into four quadrants. The trained model in the first quadrant is transferred to other quadrants, respectively, and compared the learning ability of the models with insufficient samples. The results demonstrate that the model after transfer learning (TL) can obtain damage information faster and can realize high-precision damage quantification on 32 small-scale samples, which prove the validity and reliability of TL theory in SHM.