An online visual inspection method for aluminum strip welding defects

Abstract

The present application relates to the field of visual detection, specifically to an online visual detection method for aluminum strip welding defects, which synchronously collects deformed fringe images and infrared thermal images through joint calibration, then processes the fringe data to reconstruct the three-dimensional morphology of the weld through phase unwrapping and coordinate calibration; at the same time, a dynamic perception model of the molten pool is constructed by using the optical flow method to convert the infrared thermal image into a motion vector field, capturing the dynamic changes of the molten pool; then, the system fuses the three-dimensional morphology, surface temperature and motion vector field data to form a multi-modal feature image; through molten pool feature distribution verification, abnormal data is removed to ensure the reliability of the analysis. Finally, a three-branch detection head is used for deep analysis of the image, which not only can identify the types of defects such as pores, cracks, undercut, etc., but also can quantify the severity and accurately locate the geometric position, shape and contour, so as to realize the comprehensive online monitoring of the welding quality.

Images