Welding quality monitoring of high frequency straight seam pipe based on image feature.

Weld surface images were collected using a machine vision technique, and the geometry and texture features of the images were extracted by MATLAB software. Welding quality was determined by a weighted weld strength, elongation, impact energy and bending angle. A relationship between the welding quality and the image features was established. Experimental results indicate that the welding quality can be described quantitatively by such image features as the defect perimeter, invariant moment of IM 1 , IM 7 , IM 5 , IM 4 and rectangular degree, and a BP neural network model can be used to monitor the welding quality online. [ABSTRACT FROM AUTHOR]