Deep learning paradigm shift in composite materials performance prediction: A review of state-of-the-art applications

Abstract

Given the extensive application of composite materials in aerospace and a diversity of other fields, an accurate prediction of composite properties has become increasingly important. However, traditional experimental methods are time-consuming and costly. Deep learning (DL) has emerged as a transformative tool in composite materials research due to its powerful data processing capabilities. This paper reviews the application of DL models in predicting composite materials properties, providing a comparative analysis of four mainstream DL architectures: convolutional neural network (CNN), recurrent neural network (RNN), autoencoder (AE), and generative adversarial network (GAN). The associated fundamental principles, applications, and recent advancements are addressed, summarizing DL model evaluation methodologies in classification, regression, and image-based tasks. Moreover, this review considers current challenges and future research directions, offering valuable insights to inform further investigations. This study aims to serve as a significant reference for researchers engaged in this field of research.