Разработка программного обеспечения технического зрения роботизированной ячейки

Abstract

This article addresses the design and implementation of a machine vision system to enhance the flexibility and autonomy of an industrial robotic cell. Two neural-network–based approaches were compared: a Python-based YOLO architecture and Hikrobot SC3000 smart-camera training using the VisionTrain platform, both trained on the same dataset for detecting raw and processed compressor wheels. The VisionTrain model proved more reliable in limiting false detections. To further improve safety and reliability, the system was extended to recognize misaligned parts and human hands. Finally, the vision system was integrated with a Reds C20 manipulator via Modbus TCP over Ethernet, using a Lua script to translate camera coordinates into robot commands, and a control program was developed to adapt the robot’s trajectory based on real-time vision data.