Skip to search boxSkip to navigationSkip to main content

Detection of porosity across CFRP layers using machine learning techniques applied to theoretical and experimental ultrasound data

Research Output:
Chapter in Book/Report/Conference proceeding
Conference contribution
 Peer-review

Abstract

There is an increasing use of Fiber-Reinforced Polymer composites as a replacement of metallic components in the transport applications such as aircraft and automobile. These structures are known to depict interesting and superior mechanical properties. However, these structures are often subjected to defects that alter their efficiency. Porosity such as voids inclusion, is among the most common of these defects. Since, porosity reduces the mechanical performance of such composite structures, it is important to detect and characterise its level and location across the layers. This study deals with the detection of porosity across CFRP layers using Machine Learning (ML) techniques Applied to theoretical and experimental ultrasound data. Different samples of CFRP composites with various levels of porosity are fabricated and tested for this study. The experimental data is acquired using an ultrasound immersion tank. The theoretical study for this work is built around both analytical and numerical approaches accounting for realistic conditions of the composites testing. Both simulated and measured data are used to apply a ML technique, mainly the Convolutional Neural Networks (CNN), to detect and characterise the porosity within the CFRP layers. C-scan and B-scan results are analysed and presented to demonstrate the potentials of the CNN technique to characterise such defects. It is observed that CNN technique has some interesting potentials for extracting defects such as porosity from complex ultrasound data.

Publication Information

Output type

Research Output:
Chapter in Book/Report/Conference proceeding
Conference contribution
 Peer-review

Original language

English

Publication milestones

  • Accepted/In press - 02/12/2024
  • Published - 02/12/2024

Publication status

Published - 02/12/2024

Publication IDs

  • handle.net: 10547/626549

Host publication title

nan

Access to documents

Handle.net

Persistent link