Development of an automatic weld defect identification system for radiographic testing

1996 ◽  
Vol 29 (3) ◽  
pp. 185
Author(s):  
Dilip Kumar ◽  
Luis Ganhao

On a recent project, four high pressure steam separator vessels were received from overseas after fabrication. There was suspicion on the quality of fabrication when non destructive examination (NDE) reports were reviewed. There were major concerns with the quality of radiographic films as they did not meet the ASMe Section VIII Div. 1 Code requirements as well as client specifications. Subsequent examination of welds using radiographic testing (RT) revealed crack-like features around nozzles in the region adjoining (but outside) the weld metal. Macro etching at the surface around nozzles showed that the weld area was extended beyond the apparent weld/base metal interface. Further examination of a cross section cut out from one vessel nozzle confirmed the initial doubts that weld repairs had been performed that were not reported. Metallography of the cross section indicated evidence of significant cracking associated with carbon contamination and very high hardness (up to 365 HV; in one particular case 609 HV) in affected areas. This was believed to be due to improper and incomplete cleaning by grinding after performing carbon arc or, flame gouging to remove a weld defect. Further detailed NDE was carried out using advanced ultrasonic testing (UT), i.e. phased array UT and time of flight diffraction (TOFD) and all defects (many new that were undetected by RT) were repaired per ASME Section VIII Div. 1 Code and client specification. This experience was a lesson for the design office and helped make a decision to be much more vigilant and to ask for greater quality surveillance on overseas fabrication of critical equipment for all future projects. The paper discusses the detailed investigation as well as findings.


2021 ◽  
Vol 12 (5) ◽  
pp. 390-394
Author(s):  
Distun Stephen ◽  
Dr.Lalu P.P

Weld defect identification from radiographic images is a crucial task in the industry which requires trained human experts and enough specialists for performing timely inspections. This paper proposes a deep learning based approach to identify different weld defects automatically from radiographic images. To employ this a dataset containing 200 radiographic images labelled for four types of welding defect- gas pore, cluster porosity, crack and tungsten inclusion is developed. Then a Convolutional Neural Network model is designed and trained using this database.


2016 ◽  
Vol 87 (3) ◽  
pp. 035110 ◽  
Author(s):  
Hongquan Jiang ◽  
Zeming Liang ◽  
Jianmin Gao ◽  
Changying Dang

Sign in / Sign up

Export Citation Format

Share Document