Industrial Vision: Rectifying Millimeter-Level Edge Deviation in Industrial Internet of Things With Camera-Based Edge Device

Abstract

Nowadays, to realize the intelligent manufacturing in Industrial Internet of Things (IIoT) scenarios, novel approaches in computer vision are in great demand to tackle the new challenges in IIoT environment. These approaches, which we call <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Industrial Vision</i> , are expected to offer customized solutions for intelligent manufacturing in an accurate, time efficient and robust manner. In this paper, we propose a novel approach to industrial vision, called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Edge-Eye</i> , to rectify the edge deviation automatically for Irradiated Cross-linked Polyethylene Foam (IXPE) production with millimeter-level accuracy. IXPE has been one of the most commonly used materials in industry. During the production process of IXPE sheets, their edges need keep aligned strictly, otherwise, they could quickly get out of the border of the rolling plate and cause the huge economic loss. We deploy a commercial camera with mobile edge node in front of the IXPE sheet to continuously detect and rectify the edge deviation. Particularly, to handle the complex production environment when extracting the edge of IXPE sheet, we deploy a pair of reference bars with high-contrast colors to efficiently differentiate the sheet edge from the background. Then, we propose a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Bi-direction Edge Tracking method</i> to perform the edge detection from both vertical and horizontal aspects. To realize the rectification using mobile edge nodes with limited computing resources, we reduce the cost of computation by extracting the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Minimized Region of Interest</i> , i.e., the edge area overlapped with the higher contrast reference bar on both sides. We further design a negative feedback control system with multi-stage feedback regulation mechanism, keeping the edge deviation within <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">millimeter-level</i> . We implemented <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Edge-Eye</i> on the ARM64 platform and performed evaluation in the practical IXPE production process. The experimental results show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Edge-Eye</i> achieves the average accuracy of 5 mm for the edge deviation rectification, with the average latency of 200 ms for edge deviation detection. During the process of 20-month real deployment for 36 production lines, 66 manpower per day (90% of the overall manpower) has been saved, and the utilization rate of IXPE material increases from 87% to 94%.