Pseudo Defect Data Generation With Feature Distribution Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pseudo defective product data generators create pseudo defective product data that are similar to each other, leading to insufficient improvement in inspection device determination accuracy due to the scarcity of actual defective product data in manufacturing sites.
Innovation Solution
A pseudo defective product data generator that acquires actual defective product data, converts it into feature quantities, generates predicted and pseudo defective product data using machine learning models, compares feature quantity distributions to calculate errors, and determines data quality, allowing for the generation of diverse pseudo data that improves inspection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional pseudo defective product data creation methods are used, then a large amount of pseudo defective product data can be generated, but the generated data are similar to each other and do not sufficiently improve determination accuracy
Solution Approach 1:
The patent applies parameter changes by transforming the input defective product data through multiple processing stages: extracting feature quantities, generating predicted feature quantities with variation, synthesizing new data with adjusted parameters, and reorganizing feature distributions. This ensures generated pseudo defective product data maintains parameter diversity while matching the statistical characteristics of actual defective data, thereby improving determination accuracy without requiring excessive data quantity
Solution Approach 2:
The patent implements feedback mechanisms by comparing the distribution of feature quantities in generated pseudo defective product data against the distribution in actual defective product data. The system iteratively adjusts generation parameters based on this comparison feedback to ensure the pseudo data closely matches real defective data characteristics, resolving the contradiction between data quantity and accuracy
2Measurement precision
If many pieces of actual defective product data are collected, then determination accuracy can be improved, but it is difficult to collect sufficient defective product data in manufacturing sites
Solution Approach 1:
The patent applies copying by creating pseudo defective product data as synthetic copies of actual defective product data. Instead of collecting numerous actual defective samples from manufacturing sites, the system generates multiple copies with varied characteristics through feature quantity transformation and distribution matching, achieving sufficient training data without extensive physical data collection
Solution Approach 2:
The patent uses feature quantities as an intermediary representation between actual defective product data and pseudo defective product data. By transforming actual data into feature quantities, adding controlled variations, and reconstructing data with matched distributions, the system efficiently generates diverse training samples without direct copying, improving both accuracy and collection efficiency
Data Source
AI summary
Included are a first feature quantity conversion unit that converts a plurality of pieces of acquired actual defective product data respectively into actual feature quantities, a predicted feature quantity generation unit that generates a predicted feature quantity group by learning of a feature quantity generation model, a pseudo defective product data generation unit that generates a pseudo defective product data group by learning of an image generation model, a second feature quantity conversion unit that converts the pseudo defective product data group to acquire as a pseudo feature quantity group, a feature quantity distribution comparison unit that compares distributions between the predicted feature quantity group and the pseudo feature quantity group to calculate a feature quantity error as a residual error, and a pseudo defective product data quality determination unit that determines the quality of the generated pseudo defective product data group, based on the feature quantity error.