A deep learning based methodology for artefact identification and suppression with application to ultrasonic images

• Published in: NDT & E international : independent nondestructive testing and evaluation Vol. 126; p. 102575
• Main Authors: Cantero-Chinchilla, Sergio, Wilcox, Paul D., Croxford, Anthony J.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2022
• Subjects: Artefact identification; Autoencoders; Deep learning; Full matrix capture; Non-destructive evaluation; Phased-arrays; Total focusing method; Ultrasound; Deep learning; Total focusing method; Autoencoders; Non-destructive evaluation; Artefact identification; Phased-arrays; Ultrasound; Full matrix capture
• ISSN: 0963-8695; 1879-1174
• DOI: 10.1016/j.ndteint.2021.102575

This paper proposes a deep learning framework for artefact identification and suppression in the context of non-destructive evaluation. The model, based on the concept of autoencoders, is developed for enhancing ultrasound inspection and defect identification through images obtained from full matrix capture data and the total focusing method. An experimental case study is used to prove the effectiveness of the method while exploring its practical limitations. A comparison with a state-of-the-art methodology based on image analysis is addressed for the identification and suppression of artefacts. In general, the proposed method efficiently provides accurate suppression of artefacts in complex scenarios, even when the defect is located below the footprint of the ultrasonic probe, and also yields the physical parameters needed for imaging as a by-product. •A deep learning method for artefact identification and suppression is proposed.•The grey-box model provides physical parameters used for imaging.•The method is trained using defect-free model data and tested in experimental data.•Ultrasonic image interpretation is facilitated by removing artefacts from raw data.