Automatic Air-Coupled Ultrasound Detection of Impact Damages in Fiber-Reinforced Composites Based on One-Dimension Deep Learning Models

Abstract

Impact damage constitutes a major threat to the performance and safety of fiber-reinforced composites. In this regard, transmission air-coupled ultrasound inspection technology has been identified as an ideal method for detection of common structural defects in modern multilayer composites. However, traditional machine learning algorithms and ultrasonic signal analysis methods are limited in terms of efficiency and accuracy. To remedy the situation, four one-dimensional deep learning models based on A-scan signals obtained from air-coupled ultrasound, which can automatically detect the impact damage in fiber-reinforced polymer composites, are constructed in this paper. Remarkably, all four models have attained high accuracy and recall on the testing sets, even though the training data and test data correspond to different materials and even structures. Among the four models, the long short-term memory recurrent neural network outperforms the other three models, which demonstrates its robustness and effectiveness.