Research on Ultrasonic Quantitative Evaluation Technology of Complex Defects Based on Neural Network. Issue 1 (1st February 2022)

• Record Type: Journal Article
• Authors: Li, Xinglong; Liu, Shengguo; Cheng, Shuo; Lin, Jindi; Liu, Rongchun; Wang, Leyu; Zhou, Zhilin
• Is Part Of: Journal of physics. Volume 2196:Issue 1(2022)
• Publication Date: 2022-02-01
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/2196/1/012022
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: As one of the five non-destructive testing methods, ultrasonic testing is widely used because of its accurate positioning, high sensitivity and simple operation, but the method is still difficult to locate and quantify complex shape defects. The large amount of data required for ultrasonic imaging leads to low detection efficiency. Based on this, the article establishes an inversion system for evaluating complex shape defects, which includes ultrasonic A-scan technology, BP neural network, image processing technology and signal processing technology. The system is verified by simulation and experiment. The results of the defect inversion are as follows: the similarity coefficients are all greater than 0.89, the maximum value can reach 0.95; the area error is less than 11%, the minimum value can reach 1.2%; the centroid x error is less than 12%, the minimum value can reach 1.58%; the centroid y error is less than 11%, the minimum value can reach 2.15%. The result of defect inversion further verifies the accuracy and reliability of the complex defect inversion system.