Automated Identification And Characterization Of Clustered Weld Defects

Automated identification and characterization of clustered weld defects, which comprise a plural of closely distributed individual member defects, is challenging. In this paper, time, Cartesian and Polar domain features in phased array ultrasonic sectorial scanning data are studied to segment clustered weld defect echoes by density based and K-means clustering. Through multiple domain segmentation, all individual member defect echoes are successfully identified. Then size and depth estimation is performed on each individual member defect to describe its patterns and to locate its depth. The developed technique is applied to experimental data collected from scanning on a tubular T weld test sample and is shown to discriminate, locate, and measure all the individual defects inside clustered defects. Based on the geometrical features of each individual member defect, they are classified as either point, linear or planar defects.