Automobile part welding quality online detection method and system based on machine vision
By employing multi-directional data acquisition and feature registration techniques, combined with dynamic illumination compensation and deep learning models, the problems of low efficiency and error accumulation in existing welding quality inspections have been solved, achieving accurate and stable online inspection and automatic calibration.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI VIE AUTO PARTS CO LTD
- Filing Date
- 2026-05-22
- Publication Date
- 2026-06-19
AI Technical Summary
Existing methods for inspecting the welding quality of automotive parts are inefficient, subjective, have a high rate of missed detections, lack multi-dimensional indicators and scenario-based thresholds, have poor environmental adaptability, and lack automatic self-calibration functions, leading to the accumulation of inspection errors.
The system employs a multi-directional industrial camera and a laser contour sensor to collaboratively acquire 2D RGB images and 3D point cloud data. Combined with dynamic illumination compensation and noise reduction processing, it achieves accurate registration of 2D and 3D features through Harris corner detection and Euclidean clustering. It uses macroscopic threshold segmentation and YOLOv8 deep learning model to identify defects in a hierarchical manner, constructs a multi-dimensional quality judgment system and performs fuzzy comprehensive evaluation to achieve automatic self-calibration.
It enables precise and efficient online inspection of welding quality of automotive parts, balancing inspection efficiency and accuracy to meet the needs of different application scenarios and ensure inspection stability and data traceability.
Smart Images
Figure CN122244049A_ABST