Automated deep learning for defect detection in carbon fibre reinforced plastic composites

Tools

Tunukovic, Vedran and McKnight, Shaun and Mohseni, Ehsan and Pierce, S. Gareth and Pyle, Richard and Duernberger, Euan and Loukas, Charalampos and Vithanage, Randika K.W. and Lines, David and Dobie, Gordon and MacLeod, Charles N. and Cochran, Sandy and O'Hare, Tom (2023) Automated deep learning for defect detection in carbon fibre reinforced plastic composites. In: 50th Annual Review of Progress in Quantitative Nondestructive Evaluation, 2023-07-24 - 2023-07-27, Sheraton Austin Hotel at the Capitol,.

[thumbnail of Tunukovic-etal-QNDE-2023-Automated-deep-learning-for-defect-detection]
Preview
 Text. Filename: Tunukovic-etal-QNDE-2023-Automated-deep-learning-for-defect-detection.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (1MB)| Preview

Abstract

Carbon Fibre Reinforced Polymers (CFRPs) are used extensively in the aerospace industry because of their unique physical properties and reduced weight that enables lower fuel consumption. This increase was especially rapid in the past decade, with CFRPs accounting for around 50% of the total material weight used in flagship models by Airbus and Boeing [1,2]. Before shipping, Non-Destructive Testing (NDT) methods are used to validate and control the quality of manufactured parts. Commonly used NDT technologies are radiographic testing, eddy current testing, and Ultrasonic Testing (UT). In the aerospace industry, UT is most prominent due to its flexibility and safety. However, when UT is done manually, reliability issues are often observed due to human inspector errors [3]. In addition to this, manufactured parts that need to be inspected are quite large (e.g., wing covers), resulting in slow inspection times. On the other hand, when NDT robotic inspection is deployed, large amounts of data can be captured in a short period of time. While this accelerates the acquisition of information, data interpretation is still done manually thus creating a bottleneck. Therefore, an automated data interpretation system would greatly improve the NDT process. To overcome these challenges, this project proposes a fully automated Deep Learning (DL) approach that leverages current technological advances in Machine Learning (ML) field for defect localization, sizing, and automatic report generation based on ultrasonic amplitude C-scans. Such an approach could decrease the processing time from approximately 6 hours for a 15-meter wing cover to just minutes, significantly benefiting the process throughput. In this research, a manually annotated semi-analytical simulated dataset in form of C-scans was used for training of "You Only Look Once" family of models for the detection and sizing of back-drilled holes and delamination defects in CFRPs. The purpose of using model-based simulations for training was the scarcity of real-world data, and a novel approach of image augmentation was introduced to ensure that the simulated scans closely mimic the experimental data. For NDT inspection, a force-torque-controlled 6-axis industrial robotic arm was used to deliver a phased array ultrasound roller probe to both defect-free and defective CFRP samples of varying thicknesses. The roller-probe array was connected to an array controller and water-coupled to the surface of the CFRPs. Raster scans were performed while the array was excited in linear-scan mode with a sub-aperture of 4 elements and an operating frequency of 5 MHz. Lastly, amplitude C-scan images of 64 x 64 resolution were extracted and used as an object detection validation dataset. These combined methods result in an accurate and precise deep learning network that enables rapid analysis of image data (with the possibility of real-time analysis).

ORCID iDs

Tunukovic, Vedran ORCID logoORCID: https://orcid.org/0000-0002-3102-9098, McKnight, Shaun, Mohseni, Ehsan ORCID logoORCID: https://orcid.org/0000-0002-0819-6592, Pierce, S. Gareth ORCID logoORCID: https://orcid.org/0000-0003-0312-8766, Pyle, Richard ORCID logoORCID: https://orcid.org/0000-0002-5236-7467, Duernberger, Euan, Loukas, Charalampos ORCID logoORCID: https://orcid.org/0000-0002-3465-8076, Vithanage, Randika K.W. ORCID logoORCID: https://orcid.org/0000-0002-1023-2564, Lines, David ORCID logoORCID: https://orcid.org/0000-0001-8538-2914, Dobie, Gordon ORCID logoORCID: https://orcid.org/0000-0003-3972-5917, MacLeod, Charles N. ORCID logoORCID: https://orcid.org/0000-0003-4364-9769, Cochran, Sandy and O'Hare, Tom;
CORE (COnnecting REpositories)