The assessment of structural reliability of welded structures requires data on the reliability of NDT. To establish reliability of NDT, a lot of data are needed which relate kind, size, location and orientation of defect to test results of specific NDT methods. However, there is few data available. This is considered largely because it is difficult to fabricate a specimen with a defect of which real size and location are precisely known.
The authors have developed specimens containing defects of which sizes and locations are known. The fabrication procedure of a specimen with simulated defect of controlled size and location by diffusion bonding is described. It was demonstrated that a planer defect as thin as 1μm is possible to be reproduced. Using specimens thus manufactured, reliability of ultrasonic testing was studied in terms of detectability and accuracy in defect sizing. It was shown that the detectability is affected more by defect orientation than by defect length or width. As for the accuracy, it was shown that ultrasonic testing in accordance with JIS Z 3060-1983 tends constantly to overestimate defect shorter than about 15mm. It was also demonstrated that indicated lengths fit fairly well to the normal distribution function. Mean values and standard deviations of indicated lengths were obtained for actual lengths ranging from 5 to 20mm where width and orientation of defect were taken as parameters.