Development and Experimental Evaluation of a 3D Vision System for Grinding Robot

Area of Science:

• Robotics
• Computer Vision
• Manufacturing Technology

Background:

• Current grinding robots lack automatic workpiece positioning and measurement capabilities.
• Economic and precision limitations hinder the integration of intelligent features in existing systems.
• Automated target detection is crucial for improving grinding robot efficiency and intelligence.

Purpose of the Study:

• To develop and validate a 3D vision system for automated machining target detection in grinding robots.
• To detail the hardware architecture and data processing methods for the proposed vision system.
• To assess the system's accuracy and potential for integration into intelligent grinding systems.

Main Methods:

• A 3D vision system was mounted on the robot's fourth joint.
• Point cloud preprocessing using voxel grid filters and feature descriptor extraction.
• Approximate Nearest Neighbors (FLANN) for difference point cloud identification and segmentation for path point extraction.
• Detection error compensation model for system calibration and transformation of machining information.

Main Results:

• The 3D vision system demonstrated an absolute average error of 0.154 mm in repeated measurements.
• The system's absolute measurement error due to compound error was typically under 0.25 mm.
• Successful transformation of machining information into the grinding robot base frame.

Conclusions:

• The proposed 3D vision system effectively detects machining targets for grinding robots.
• The system achieves high precision, suitable for industrial grinding applications.
• Easy integration into intelligent grinding systems is feasible, enhancing automation and efficiency.