Progressive Defect Segmentation for Quantifying Minor Surface Flaws
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting defects in large equipment components are inefficient, labor-intensive, and prone to missing minor defects, lacking quantitative analysis capabilities.
Innovation Solution
A method utilizing a progressive segmentation network for defect analysis, involving overlapping image cropping, feature extraction through a residual unit and cross-scale feature fusion, and classification using a defect segmentation network to quantify defect dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual visual inspection is used for defect detection, then operational simplicity is maintained, but detection accuracy and quantitative measurement capability deteriorate
Solution Approach 1:
The patent replaces manual visual inspection with an automated deep learning system comprising a progressive segmentation network. This system automatically performs defect detection, segmentation, and quantitative measurement without human intervention, thereby improving both detection accuracy and maintaining operational simplicity through automation.
Solution Approach 2:
The patent creates a digital copy of the defect detection process through trained neural network models. The progressive segmentation network learns from training datasets to replicate expert defect detection capabilities, enabling automated quantitative analysis while preserving the simplicity of operation through software-based solutions.
2Device complexity
If traditional defect detection methods are used, then system complexity is kept low, but defect detection efficiency and productivity deteriorate
Solution Approach 1:
The patent divides the defect detection task into multiple segments: defect detection, defect segmentation, and quantitative measurement. The progressive segmentation network processes images through multiple stages, with each stage focusing on specific aspects of defect analysis, thereby improving efficiency while managing complexity through modular architecture.
Solution Approach 2:
The patent performs preliminary actions by pre-training the progressive segmentation network on large datasets before deployment. The network is pre-trained to recognize various defect patterns, enabling rapid and accurate defect detection during actual operation without requiring complex real-time adjustments, thus improving productivity while maintaining reasonable system complexity.
3Device complexity
If simple detection methods are used, then system complexity is reduced, but the ability to detect minor defects deteriorates
Solution Approach 1:
The patent transitions from traditional 2D image analysis to multi-dimensional feature extraction using the progressive segmentation network. The network analyzes images across multiple feature dimensions and scales, enabling detection of subtle minor defects that would be imperceptible in simple 2D analysis, thereby improving reliability without excessive complexity through sophisticated yet systematic processing.
Solution Approach 2:
The patent changes detection parameters by adjusting network architecture, training parameters, and processing thresholds to optimize minor defect detection. The progressive segmentation network uses multiple processing stages with different parameter settings, allowing sensitive detection of minor defects while managing system complexity through parameter optimization rather than structural complexity.
4Measurement precision
If comprehensive defect analysis is performed, then measurement precision is improved, but processing time and operational complexity deteriorate
Solution Approach 1:
The patent implements continuous defect analysis through the progressive segmentation network, which processes images through multiple continuous stages of detection, segmentation, and measurement. This continuous processing approach provides comprehensive quantitative analysis without requiring separate discrete operations, thereby improving measurement precision while minimizing additional processing time through integrated workflows.
Data Source
AI summary
The present disclosure provides a method for analyzing a minor defect based on a progressive segmentation network, including: acquiring an original image for a surface of a component, and cropping the original image into a plurality of patches; inputting each of the patches to a minor defect feature extraction network to extract an image feature; classifying the patch into a defective image or a non-defective background image according to an extracted image feature; inputting an extracted image feature of the defective image to a defect segmentation network to obtain a segmentation mask image of a corresponding defect; and quantitatively analyzing the defect according to the segmentation mask image to obtain information such as an area, a length and a width of the defect.