3D Vision Inspection for Industrial Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current industrial computer vision systems rely on 2D imaging for inspection, which is inadequate for features characterized by physical size, area, or volume, as it loses critical 3D shape information and can be biased towards visible features, whereas 3D imaging is not widely adopted due to the unreliability of consumer-grade 3D cameras.
Innovation Solution
A method for direct 3D vision inspection using 3D vision systems, involving a training phase to create a 3D alignment model and variance model, and a runtime phase for defect detection by aligning and comparing 3D test images with a reference image, utilizing 3D cameras and image filtering techniques to handle noise and occlusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If 2D imaging is used for inspection, then the system is simple and cost-effective, but critical 3D shape information is lost and inspection results are biased towards visible features
Solution Approach 1:
The patent transitions from 2D imaging to 3D imaging by capturing depth information alongside intensity data. The 3D vision system acquires both 2D intensity images and corresponding depth maps, enabling inspection of features characterized by physical size, area, or volume without losing critical 3D shape information. This dimensional enhancement resolves the information loss problem while maintaining system feasibility.
2Reliability
If consumer-grade 3D cameras are used, then the system becomes more capable of capturing 3D information, but reliability and stability are insufficient for industrial applications
Solution Approach 1:
The patent changes the key parameter from consumer-grade 3D cameras to industrial-grade 3D cameras, which provide the necessary reliability and stability for industrial applications. Industrial cameras offer improved noise characteristics, better depth accuracy, and enhanced robustness while maintaining the capability to capture comprehensive 3D information including shape, size, and volume features.
3Ease of operation
If 3D images are converted to 2D images for inspection, then the inspection process is simplified, but information about 3D shape or volume is lost and results are biased
Solution Approach 1:
The patent maintains 3D image data throughout the inspection process rather than converting to 2D. The system performs registration, comparison, and defect detection directly on 3D point clouds and depth maps, preserving all 3D shape and volume information. This approach enables accurate measurement of features characterized by physical dimensions while avoiding the bias inherent in 2D projections.
4Measurement precision
If direct 3D vision inspection is implemented, then accurate detection of 3D defects is enabled, but system complexity and processing requirements increase
Solution Approach 1:
The patent segments the inspection process into distinct phases: data acquisition (intensity and depth), registration (alignment to CAD model), comparison (defect detection), and classification. This segmentation manages complexity by organizing processing steps while enabling precise 3D defect detection through specialized algorithms for each phase, including point-to-surface distance calculations and statistical analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method for three dimensional (3D) vision inspection using a 3D vision system. The system and method comprising acquiring at least one 3D image of a 3D object using the 3D vision system, using the 3D vision system; extracting a 3D visible runtime mask of the 3D image; using the 3D vision system, comparing the 3D runtime visible mask to a 3D reference visible mask; and, using the 3D vision system, determining if a difference of pixels exists between the 3D runtime visible mask and the 3D reference visible mask.