Deployment of Machine Vision Platform for Checking Spot Welds on Metal Strap Belts

Abstract

This paper presents a designed platform used to detect the perfection and number of spot welds on the strap belt of metal sheet coils. Three different approaches include image morphology, thresholding and Hough transform are compared. The results from real implementation show that using the adaptive threshold values in image thresholding approach instead of fixed value can increase the system accuracy from 69% to 88%. Also, to find the pad, the comparison of using Haar cascade machine learning and YOLO deep learning is described.