Unsupervised Learning Data Analysis for Manufacturing Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to accurately classify inspection images of defects and faults in manufacturing environments, especially when new defects occur due to changes in manufacturing conditions, as they do not effectively learn features suitable for each defect type.
Innovation Solution
A data analysis apparatus that acquires and processes inspection images using unsupervised learning to generate feature vectors and clustering results for both existing and new data, allowing for accurate comparison and identification of unique features under different manufacturing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If unsupervised learning is used to classify inspection images without manual labeling, then the operation cost and time for data preparation are reduced, but the accuracy of classifying new defect types under changed manufacturing conditions deteriorates
Solution Approach 1:
The system performs preliminary unsupervised learning on existing data to establish baseline feature vectors and clustering structures before new data arrives. This preliminary classification creates a foundation that can be quickly adapted to new manufacturing conditions, reducing the time needed for manual labeling while maintaining classification accuracy through the comparison mechanism.
Solution Approach 2:
The system implements feedback by comparing clustering results between existing data and new data. When manufacturing conditions change, the comparison mechanism provides feedback about deviations from normal patterns, enabling the system to identify and classify new defect types accurately without requiring extensive manual labeling of new data.
2Device complexity
If a single classification model is used for all manufacturing conditions, then the device complexity is reduced, but the ability to detect defect-specific features under different manufacturing conditions deteriorates
Solution Approach 1:
Instead of using multiple static classification models for different manufacturing conditions, the system employs a dynamic approach where a single unsupervised learning model adapts to different conditions by comparing clustering results. The model dynamically adjusts its classification behavior based on the comparison between existing data patterns and new data patterns, maintaining reliability without increasing device complexity.
Solution Approach 2:
The single classification model is designed to be universal and handle multiple manufacturing conditions through the comparison mechanism. By comparing clustering results across different conditions, the model achieves multi-functionality, detecting defect-specific features under various manufacturing conditions without requiring separate specialized models for each condition.
3Productivity
If existing data and new data are classified together using the same features, then the processing time is reduced, but the ability to identify unique features of new defects deteriorates
Solution Approach 1:
The system segments the classification process into two stages: first, unsupervised learning on existing data to establish baseline features; second, unsupervised learning on new data with comparison to existing patterns. This segmentation allows the system to process data efficiently while preserving unique features of new defects through the comparison mechanism, which identifies deviations from established patterns.
Solution Approach 2:
The system adds a comparative dimension to the classification process by comparing clustering results between existing data and new data. This additional dimension enables the system to identify unique features of new defects without sacrificing processing speed, as the comparison operates on extracted feature vectors rather than raw images, maintaining efficiency while capturing novel defect characteristics.
Data Source
AI summary
According to one embodiment, a data analysis apparatus includes processing circuitry. The processing circuitry acquires a plurality of pieces of first data satisfying a first condition, generates a plurality of first feature vectors by unsupervised learning of the plurality of pieces of first data, generates a first clustering result by clustering the plurality of first feature vectors, acquires a plurality of pieces of second data satisfying a second condition different from the first condition, generates a plurality of second feature vectors by unsupervised learning of at least some of the plurality of pieces of first data and the plurality of pieces of second data, generates a second clustering result by clustering the second feature vectors, and generates a comparison result regarding the plurality of pieces of first data and the plurality of pieces of second data by comparing the first clustering result with the second clustering result.