Steel Surface Defect Detection Using Learned Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting periodic defects in steel sheets are hindered by sheet meandering, leading to inaccurate detection and over-detection of defects. Current techniques require significant storage and calculation resources, and are prone to errors due to meandering-induced displacement.
Innovation Solution
A learned model generation method using machine learning, where a teacher image with a defect map and assigned periodic defects is used to generate a learned model. This model takes defect maps as input and outputs the presence or absence of periodic defects, with image size conversion ensuring consistent input sizes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional defect detection methods are used to detect periodic defects, then detection capability is provided, but detection accuracy deteriorates due to steel sheet meandering causing positional displacement
Solution Approach 1:
The patent creates a defect map that copies the spatial distribution pattern of defects from the original inspection data. This defect map serves as a standardized representation that can be analyzed independently of the actual steel sheet position, allowing accurate periodic defect detection even when meandering occurs
Solution Approach 2:
The patent transforms the inspection problem by changing from direct positional analysis to pattern-based analysis. By converting defect positions into a defect map with standardized coordinates and analyzing the spatial distribution patterns, the system becomes insensitive to meandering-induced positional variations
2Measurement precision
If autocorrelation calculation is performed on light-receiving signals to detect periodic defects, then periodic defect detection capability is improved, but storage and calculation resources are excessively consumed
Solution Approach 1:
The patent extracts only the essential defect position information and creates a simplified defect map representation, discarding redundant raw inspection data. This extraction approach maintains the ability to detect periodic defects while dramatically reducing storage and computational requirements compared to full autocorrelation analysis
Solution Approach 2:
Instead of analyzing raw signals to detect periodicity, the patent inverts the approach by first creating a defect map and then analyzing the spatial distribution patterns. This inversion allows periodic defect detection with minimal data storage and computation
3Reliability
If allowable width setting method is used to detect periodic defects during meandering, then detection capability is maintained, but over-detection occurs reducing measurement precision
Solution Approach 1:
The patent analyzes the spatial distribution pattern of defects in the defect map and uses this feedback to determine periodicity. By examining whether defects are distributed at regular intervals in the standardized defect map coordinates, the system accurately identifies periodic defects without the over-detection problems of fixed allowable width methods
Data Source
AI summary
A learned model generation method includes: using a teacher image including a defect map that is an image indicating a distribution of a defect portion of a surface of steel and having an equal image size, and presence/absence of periodic defects assigned in advance to the defect map; and generating a learned model by machine learning, the learned model for which: an input value is a defect map that is an image indicating a distribution of a defect portion of a surface of steel and having an image size of the equal image size; and an output value is a value concerning presence/absence of periodic defects in the defect map.