Defect Detection Using Reference Patch Representations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Defect detection in manufactured items with high surface variability is challenging, as differently looking items may be regarded as okay, making it difficult to differentiate between okay and faulty items.
Innovation Solution
The solution involves using reference manufactured item patches representations (RMIPRs) to search for okay evaluated manufactured item patches, which can be indifferent to location, and are found based on feedback and popularity, allowing for clustering without prior knowledge of the manufacturing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional defect detection methods are used on items with high surface variability, then the detection system can identify defects, but it cannot reliably differentiate between okay and faulty items due to appearance variations
Solution Approach 1:
The manufactured item is divided into multiple patches, and each patch is independently evaluated and compared against reference patches. This segmentation allows the system to handle high surface variability by processing local regions separately, making the overall detection more reliable despite appearance variations across the item surface.
Solution Approach 2:
The system uses feedback from evaluating multiple reference patches and their representations to iteratively improve the detection process. By comparing evaluated patches against reference patches and adjusting the evaluation based on the results, the system achieves reliable differentiation between okay and faulty items even with high surface variability.
2Measurement precision
If the system evaluates all possible reference patches to ensure accurate defect detection, then detection accuracy improves, but the computational complexity and cost increase significantly
Solution Approach 1:
The system extracts and uses only the necessary reference patch representations (RMIPRs) for evaluation, rather than processing all possible reference patches. By selecting and extracting only the essential reference information needed for accurate detection, the system maintains high detection accuracy while significantly reducing computational complexity and cost.
Solution Approach 2:
The system performs partial evaluation by using a strategically selected subset of reference patches and their representations. This partial action approach allows the system to achieve sufficient detection accuracy without the excessive computational cost of evaluating all possible reference patches, thus resolving the contradiction between accuracy and complexity.
3Measurement precision
If the system uses location-specific reference patches to improve detection accuracy, then detection precision improves, but the system cannot generalize to different locations and production variations
Solution Approach 1:
The reference patch representations (RMIPRs) are designed to be universal and location-independent. The same RMIPRs can be used to evaluate patches at different locations on the manufactured item, allowing the system to maintain high detection precision across all locations while adapting to different production variations without requiring location-specific reference data.
Data Source
AI summary
A method for detecting faulty manufactured items, the method includes segmenting an image of the evaluated MI to provide different groups of evaluated MI patches, generating evaluated MI patches representations (EMIPRs) for the evaluated MI patches of the different groups of evaluated MI patches; matching the EMIPRs to reference MI patches representations (RMIPRs) of reference MI patches, to provide comparison results; wherein the reference MI patches are selected from reference MI patches candidates, based on at least one of (a) a popularity of reference MI patch candidates representations or (b) feedback from a person; wherein reference MI patches of the different groups of reference MI patches differ from each other by one or more reference MI patch attribute; and determining a state of the evaluated manufactured item based on the comparison results.