Difference-Image Defect Classification With Operator Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated defect detection systems in devices face challenges due to extensive training requirements, inter-operator variance, fatigue, and insufficient data for training classifiers, especially in complex devices with limited defect exemplars, leading to errors and inefficiencies.
Innovation Solution
The system isolates defects by comparing acquired images to reference images, generating difference images for analysis, using machine learning classifiers to identify defects based on size, shape, texture, and contrast, and dynamically updating the classifier with operator feedback and simulated training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated defect detection systems are implemented, then productivity is improved, but the system requires extensive training data and complex setup
Solution Approach 1:
The system performs preliminary actions by using operator feedback during actual operation to collect and label defect data in real-time. This preliminary data collection and labeling process occurs before extensive retraining is needed, allowing the system to build its training dataset progressively without requiring large amounts of pre-collected data.
Solution Approach 2:
The system implements feedback mechanisms where operators review and annotate defect detections, and this feedback is used to continuously update and retrain the machine learning classifier. This closed-loop feedback system allows the model to improve progressively with each iteration, reducing the need for extensive initial training data.
2Adaptability or versatility
If machine learning classifiers are trained with limited defect exemplars, then the system can operate with complex devices, but measurement precision deteriorates
Solution Approach 1:
The system implements dynamic adaptation where the classifier is continuously updated and retrained based on operator feedback and newly collected data. This dynamic approach allows the system to adapt to complex device variations while progressively improving detection accuracy, rather than being static with fixed performance limitations.
Solution Approach 2:
The system performs preliminary defect detection using available limited exemplars, then uses operator feedback to identify additional training cases. This preliminary action allows the system to start operating with complex devices immediately, with accuracy improving progressively as more training data is collected and incorporated.
3Measurement precision
If operators manually examine images to detect defects, then measurement precision is maintained, but productivity decreases due to fatigue and inter-operator variance
Solution Approach 1:
The system merges automated machine learning-based defect detection with human operator review and feedback. This combination allows the system to handle large volumes of images rapidly (maintaining productivity) while operators provide precision judgments on uncertain cases and contribute to continuous system improvement, achieving both high throughput and accurate detection.
Data Source
AI summary
In a method and apparatus for automated inspection, an image is acquired of an object under inspection and a difference image is generated showing the difference between the acquired image and a reference image of a defect-free object of the same type. Characteristics of the difference image, or detected isolated regions of the difference image, are passed to an automated defect classifier to classify defects in the object under inspection. The characteristics of the difference image may be pixels of the difference image or features determined therefrom. The features may be extracted using a neural network, for example. The automated defect classifier is trained using difference images and may be further trained, in operation, based on operator classifications and using simulated images of defects identified by an operator.