Advanced deep learning model-based impact characterization method for composite laminates

The aim of this study was to develop a structural health monitoring system for smart composite structures through the use of signal processing, deep learning algorithms, and optimization theory. Piezoelectric ribbon sensors were implemented in the preparation of smart composite structures to create...

Full description

Saved in:
Bibliographic Details
Published in:Composites science and technology 2021-05, Vol.207, p.108713, Article 108713
Main Authors: Jung, Kyung-Chae, Chang, Seung-Hwan
Format: Article
Language:English
Subjects:
A. Fabrics/textiles
Algorithms
Artificial neural networks
B. Fracture
B. Impact behaviour
Composite materials
Composite structures
D. Acoustic emission
D. Non-destructive testing
Deep learning
Discrete Wavelet Transform
Laminates
Neural networks
Optimization
Piezoelectricity
Signal processing
Structural health monitoring
Studies
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The aim of this study was to develop a structural health monitoring system for smart composite structures through the use of signal processing, deep learning algorithms, and optimization theory. Piezoelectric ribbon sensors were implemented in the preparation of smart composite structures to create a smart composite fabric that can be embedded in composite laminates to enable self-monitoring. A discrete wavelet transform was applied to the impact signals to convert them into input image data for the predictive convolutional neural network-based models. Optimal values of the hyperparameters were derived based on Bayesian optimization theory. Data augmentation was also employed to secure sufficient data for impact characterization model training. Lastly, the performance of each optimized neural network model was investigated by comparing the test errors under each applied condition. [Display omitted]
ISSN:0266-3538
1879-1050
DOI:10.1016/j.compscitech.2021.108713