Ultrasonic guided wave sensor network data inversion for resin front prediction in carbon fibre composite plastics infusion processes

Tools

Calistru, Cristian Adrian and Mohseni, Ehsan and Tunukovic, Vedran and Pierce, S. Gareth and Lines, David and MacLeod, Charles N. and Bomphray, Iain and Weis, Tobias and Munro, Gavin and O'Hare, Tom; (2025) Ultrasonic guided wave sensor network data inversion for resin front prediction in carbon fibre composite plastics infusion processes. In: 2025 IEEE International Ultrasonics Symposium (IUS). IEEE Symposium (IUS) Ultrasonics . IEEE, NLD. ISBN 9798331523329 (https://doi.org/10.1109/IUS62464.2025.11201629)

[thumbnail of Calistru-etal-IEEE-IUS-2025-Ultrasonic-guided-wave-sensor-network-data-inversion]
Preview
 Text. Filename: Calistru-etal-IEEE-IUS-2025-Ultrasonic-guided-wave-sensor-network-data-inversion.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Out-of-Autoclave (OoA) resin infusion and curing of composites offers a sustainable alternative to the traditional autoclave processing predominantly used in the aerospace industry. However, the adoption of such practices is hindered by concerns over incomplete resin impregnation in the absence of the higher infusion pressures typically found in autoclaves. Enabling the safe use of OoA methods in composite manufacturing requires real-time, in-situ monitoring of the resin flow. Ultrasonic Guided Waves (UGW) were identified as highly sensitive to liquids adjacent to the medium through which they propagate. The fundamental antisymmetric mode of the UGW was found to be attenuated by the resin progression, showcasing measurable amplitude reduction related to the progression of the resin. This study investigates UGW generated by a network of three contact piezoelectric transducers integrated into the upper lid of a setup replicating fluid propagation through a conventional infusion mould. Ultrasonic data was correlated with the exact position of the resin within the mould which was extracted through the use of a machine-vision algorithm measuring the coverage of resin relative to the propagation path, i.e., the straight line between two sensors. Following the data labelling and analysis of trends, a convolutional neural network model for front position estimation was applied, halving the error obtained with analogous statistical models to a best mean absolute error of 5.7 mm.

ORCID iDs

Calistru, Cristian Adrian ORCID logoORCID: https://orcid.org/0009-0000-4393-0871, Mohseni, Ehsan ORCID logoORCID: https://orcid.org/0000-0002-0819-6592, Tunukovic, Vedran ORCID logoORCID: https://orcid.org/0000-0002-3102-9098, Pierce, S. Gareth ORCID logoORCID: https://orcid.org/0000-0003-0312-8766, Lines, David ORCID logoORCID: https://orcid.org/0000-0001-8538-2914, MacLeod, Charles N. ORCID logoORCID: https://orcid.org/0000-0003-4364-9769, Bomphray, Iain ORCID logoORCID: https://orcid.org/0000-0001-6969-4379, Weis, Tobias, Munro, Gavin and O'Hare, Tom;
CORE (COnnecting REpositories)