A System for the Real-Time Detection of the U-Shaped Steel Bar Straightness on a Production Line

Abstract

This study develops an algorithm and a system for steel straightness detection, which combines object detection, edge detection, line detection, clustering, stitching, and bending recognition. The algorithm detects the contour of U-shaped steel bars with widths of 100 mm, named U100, or 150 mm, named U150, and lengths of 8, 10, 12 m. The algorithm uses object detection to extract the center point of the U-shaped bottom as a reference point and line detection to extract lines in the contour. The algorithm selects one-stage or two-stage edge detection based on the light source. Two-stage edge detection enhances the contour features when the light source is insufficient. After contour detection, some parts of the contour disappear due to the light source. The algorithm stitches all lines with an angle difference within ∆θ degrees into one straight line L based on the angle of the longest line. If the length of L exceeds the threshold value MLL, the steel bar is straight; otherwise, it is bent. ∆θ and MLL are used to set the acceptable bending degree. The experiment results show that the algorithm detects 123,128 steel bars in 193 h with an average accuracy of 99.64% for straight steel and an average recall of 95.70% for bent steel. The contribution of this study is the development of a real-time algorithm and its corresponding system for steel straightness determination in a steel factory, ensuring accurate and efficient assessment of steel quality in an industrial setting.

Keywords

• steel inspection
• straightness detection
• AI object detection
• image processing
• automated optical inspection