A Multidimensional Data Fusion Neural Network for Damage Localization Using Ultrasonic Guided Wave
Hongguang Yun, Ke Feng, Rakiba Rayhana, Shashank Pant, Marc Genest, Zheng Liu
Abstract
Ultrasonic guided wave (UGW) based damage localization on plate-like composite structures plays a vital role in the structural health monitoring (SHM) of aircraft structures. Precisely locating the damage requires full utilization of high-dimensional UGW signals as well as low-dimensional transducer coordinates. However, current deep learning-based methods cannot incorporate transducer coordinates in the neural networks. To address this issue, this paper proposes a novel multidimensional data fusion neural network framework for damage localization on plate-like composite structures using UGW. The proposed framework includes an encoder and a Fourier feature projection head to integrate high-dimensional wave signals and low-dimensional coordinates. A multilayer perceptron is adopted as a decoder to learn features from the encoder and the projection head. Comprehensive experiments demonstrate that the proposed method achieves state-of-the-art results with less than 2 mm absolute distance error. Moreover, a discussion regarding data availability in the training process is performed. The proposed method demonstrates superior robustness over state-of-the-art methods with limited training data.