Machine Vision Based System for Automated Sorting System for Wire Bundle Quality Inspection

Machine Vision Based System for Automated Sorting System for Wire Bundle Quality Inspection

Dr. D.B. Hulwan                                                                                                   Atharva Kadam

Dept. of Mechanical Engineering                                                                    Dept. of Mechanical Engineering

Vishwakarma Institute of Technology Pune                                           Vishwakarma Institute of Technology Pune

dattatray.hulwan@vit.edu                                                                                atharva.kadam22@vit.edu

Ramanuj Bajaj                                                                                          Aditya Bihani

Dept. of Mechanical Engineering                                                           Dept. of Mechanical Engineering

Vishwakarma Institute of Technology Pune                                           Vishwakarma Institute of Technology Pune

ramanuj.bajaj22@vit.edu                                                                        bihani.aditya22@vit.edu

Shashank Attal

Dept. of Mechanical Engineering

Vishwakarma Institute of Technology Pune

Shashank.attal22@vit.edu

Abstract—Ensuring reliable electrical continuity in wire- harness assemblies requires rapid detection and remediation of unintended gaps. We present a real-time machine-vision inspection pipeline built on the YOLOv5 object detector to automatically pinpoint physical discontinuities between jumper wires on a conveyor line. Drawing from a bespoke image corpus of annotated cluttered harnesses, our system trains a convolutional network to output tight ”gap” bounding boxes with confidence scores. Benchmarked on held-out frames, it sustains 87% detection precision and 82% recall, amid overlapping cables and variable illumination. Case studies illustrate how early gap flagging prevents glue over-application, reduces material waste, and offloads manual QA. We analyze failure modes—tiny oc- clusions and extreme wire crossings—and propose enhancements including multi-angle 3D capture and closed-loop integration with robotic applicators. Our results demonstrate that a lightweight YOLO-based approach can streamline wire assembly quality control and pave the way for fully automated glue-dispensing workflows.

Index Terms—YOLOv5, machine vision, object detection, wire harness inspection, jumper wire assemblies, gap detection, real- time quality control, convolutional neural networks (CNNs), automated inspection, bounding boxes, conveyor line automation, synthetic data augmentation, manufacturing automation.