Flaw sizing using ultrasonic C-scan imaging with dynamic thresholds.

The threshold for binarisation of an ultrasonic C-scan image has a significant effect on flaw sizing. The 6 dB drop method shows poor performance when extracting flaws from an ultrasonic image. A novel method is presented to improve the accuracy of flaw sizing, whereby the flaws are separated from the background using digital image processing techniques including the Otsu method, bilateral filtering and dilation. The optimised thresholds for each flaw are calculated using the enumeration method to binarise the original image and then the flaw size is predicted using the 6 dB drop method. The dynamic threshold model for flaw sizing is developed on the basis of the generalised regression neural network theory, where the flaw peak value, the depth of flaw echo and the pre-judgement size of the flaw are all taken into consideration. Finally, ultrasonic C-scan experiments are conducted on an AZ80 magnesium alloy shell with natural flaws. Compared with the traditional 6 dB drop method, the experimental results show that C-scan imaging using the dynamic threshold method can reduce the average relative error of flaw sizing from 40.6% to 7.4%. [ABSTRACT FROM AUTHOR]