Board Defect Filtering Using Circuit Layout and CNN Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In industrial manufacturing, automated optical inspection (AOI) of circuit boards often results in a high number of false alarms, leading to a significant workload for manual double-checking, as the criteria for defect detection are stringent and prone to misjudgments, with up to 97% of defects being false alarms such as dander, dirt, or glue shadows.
Innovation Solution
A board defect filtering method and device that receives a defect list and circuit layout image, analyzes defect locations, crops defect images, and uses a defect classifying model to determine whether images are qualified or unqualified products, reducing misjudgments by cross-comparing circuit layouts and defect images, and employing techniques like region of interest cropping and convolutional neural networks for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated optical inspection (AOI) is used to detect defects with strict criteria, then defect detection capability is improved, but the number of false alarms increases leading to higher manual double-checking workload
Solution Approach 1:
A defect classifying model serves as an intermediary between the AOI inspection results and manual verification. The model automatically classifies detected defects as qualified or unqualified, filtering out false alarms before they reach manual checking, thus reducing manual workload while maintaining detection precision
Solution Approach 2:
The patent replaces the manual mechanical verification process with an automated defect classifying model that uses machine learning algorithms. This substitution eliminates the need for manual inspection of all detected defects, significantly improving productivity while maintaining or enhancing detection accuracy through intelligent classification
2Reliability
If strict defect inspection criteria are applied, then product quality is improved, but the number of misjudgments increases requiring manual intervention
Solution Approach 1:
The defect classifying model incorporates feedback mechanisms where the classification results are validated and refined through continuous learning from actual inspection data. This feedback loop enables the model to improve its classification accuracy over time, reducing misjudgments while maintaining strict quality criteria
Solution Approach 2:
The patent changes the parameter of classification accuracy from a static metric to a dynamic one that improves through model training and feedback. By continuously refining the classification algorithm based on actual inspection outcomes, the system maintains high reliability while improving measurement precision to reduce misjudgments
Data Source
AI summary
A board defect filtering method is provided. The method includes: receiving a defect list; obtaining a plurality of defect images of a plurality of defect records on the defect list; receiving a circuit layout image; analyzing a defect location of a first defect image of the plurality of defect images according to the circuit layout image; cropping the first defect image to obtain a first cropped defect image according to the defect location; inputting the first cropping defect image to a defect classifying model; and determining whether the first defect image is a qualified product image or not according to an output result of the defect classifying model.