A quantitative damage diagnosis method for CFRTP/Ti6Al4V hybrid welded/bolted joints based on an attention-enhanced capsule network

View at publisher Summarise with VIRAT
Authors: Zhi Li, Xingzhou Chen, Jun Zeng Jinming Zhang, Haijin Wang, Huiyue Dong, Yunbo Bi
Cited by: 0
References: 41

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract To achieve high-precision quantitative damage identification for CFRTP/Ti6Al4V hybrid welded/bolted joints (HWBJ) after low-velocity impact (LVI), this work developed a damage diagnosis method based on acoustic emission (AE) and a self-attention enhanced capsule network (SA-CapsNet). First, AE waveform signals from HWBJ were collected through LVI tests and converted into time–frequency images using short-time Fourier transform as input features. On this basis, data augmentation techniques were employed to construct the dataset, effectively mitigating the learning challenges under few-shot conditions. The proposed SA-CapsNet was further applied for quantitative damage diagnosis and compared with traditional capsule network (TR-CapsNet) as well as classical models such as residual network (ResNet), convolutional neural networks (CNN), visual geometry group (VGG), and multilayer perceptron (MLP). The results indicate that the SA-CapsNet, which integrates self-attention into the dynamic routing process, achieves a higher recognition accuracy of 99.47%, compared to TR-CapsNet (98.44%), ResNet (96.35%), CNN (95.31%), VGG (94.79%), and MLP (93.23%). Moreover, comprehensive evaluation metrics show that the average precision, recall, and F1-score of SA-CapsNet reached 0.994846, 0.994792, and 0.994792, respectively—all outperforming the corresponding values of other comparative models. Notably, SA-CapsNet not only demonstrates superior statistical performance but also exhibits significantly enhanced misclassification control, highlighting its remarkable potential for high-reliability structural health monitoring applications.

Fields of Study & Subjects

Ultrasonics and Acoustic Wave Propagation Mechanical Behavior of Composites Non-Destructive Testing Techniques

Publication details

Journal / source
Engineering Research Express
Volume
8
Issue
7
Pages
075521–075521
Publication date
30 March 2026
Language
EN