Toggle Main Menu Toggle Search

Open Access padlockePrints

Physical perspective forward-inverse learning for ultrasonic sensing diagnosis in small diameter and thin-wall tube

Lookup NU author(s): Professor Gui Yun TianORCiD



This is the authors' accepted manuscript of an article that has been published in its final definitive form by Elsevier B.V., 2020.

For re-use rights please refer to the publisher's terms and conditions.


© 2020 Elsevier B.V.The ultrasonic testing method is a well-known non-destructive testing technique which has been applied to the tube inspection for guarantying the quality of the production. However, there exist several challenges to detect the defects of tubes with small diameter and thin-wall due to the complex of multiple reflections and waveform conversion. Parameters selection of the transducer takes key role to enhance the detection sensitivity such as frequency, size, refraction angle, distance offset, and focal point distance. This selection is generally dependent on human experience as it is highly time-consuming and subjective. In this paper, a novel parameter selection method based on physical perspective linked forward-inverse intelligence strategy has been proposed for ultrasonic immersed testing method. The optimized parameters can be calculated automatically while both testing and calibration repeated experiments can be avoided. The proposed method is computationally affordable and yields a high accuracy objective performance. Both simulation and experiments have been conducted to verify the efficacy of the proposed method.

Publication metadata

Author(s): Xiao X, Gao B, Tian GY, Gang Cai Z, qing Wang K

Publication type: Article

Publication status: Published

Journal: Ultrasonics

Year: 2020

Volume: 105

Print publication date: 01/07/2020

Online publication date: 10/03/2020

Acceptance date: 24/02/2020

Date deposited: 06/05/2020

ISSN (print): 0041-624X

ISSN (electronic): 1874-9968

Publisher: Elsevier B.V.


DOI: 10.1016/j.ultras.2020.106115


Altmetrics provided by Altmetric


Funder referenceFunder name