Unsupervised Defect Detection Using Generator Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated defect detection techniques in material manufacturing are labor-intensive and limited by the need for human-labeled datasets and expertise, making them inefficient and costly, especially in high-volume production environments.
Innovation Solution
An unsupervised machine learning approach using a generator neural network that automatically generates defect-free reconstructions of images, allowing for AI-based defect identification without labeled training data by iteratively training on defect-free images with superimposed defect data, enabling efficient detection of defects in real-time production.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human-labeled datasets are used for training defect detection algorithms, then detection accuracy is improved, but labor costs and training time increase significantly
Solution Approach 1:
The system performs self-supervised learning by automatically generating training labels through synthetic defect superimposition on defect-free images. The generator neural network creates augmented training data without human intervention, allowing the system to train itself on unlimited synthetic examples while maintaining detection accuracy.
Solution Approach 2:
The method pre-generates synthetic defect images by superimposing defect patterns onto defect-free images before training begins. This preliminary creation of labeled training data eliminates the need for time-consuming manual annotation during the actual training process, enabling rapid model deployment.
2Reliability
If human expertise is required for defect identification, then detection reliability is improved, but production efficiency decreases due to manual inspection bottlenecks
Solution Approach 1:
The system replaces manual human inspection with an automated generator neural network that processes images at machine speed. The trained model automatically identifies defects in real-time production streams, eliminating manual inspection bottlenecks while maintaining high detection reliability through continuous learning from synthetic data.
Solution Approach 2:
The generator neural network serves as an intermediary between raw images and defect identification, learning to map image features to defect classifications through synthetic training. This intermediary model captures human expertise patterns while operating at automated speeds, bridging the gap between human accuracy and machine efficiency.
3Extent of automation
If hand-crafted algorithms are used for defect detection, then automation is achieved, but detection effectiveness decreases due to engineering complexity
Solution Approach 1:
The system transitions from fixed hand-crafted algorithm parameters to learnable neural network parameters that adapt automatically during training. The generator network learns optimal feature extraction and defect detection parameters from synthetic data, eliminating the need for complex manual algorithm engineering while maintaining full automation.
Solution Approach 2:
Instead of engineering algorithms to detect defects directly, the system copies human inspection patterns through synthetic example generation. By superimposing defects on defect-free images and training the network to recognize these patterns, the system replicates human expertise without requiring manual algorithm design.
Data Source
AI summary
A computer implemented method including acquiring a live image of a subject physical sample of a product or material; inputting the live image to a trained generator neural network to generate a defect-free reconstruction of the live image; comparing the defect-free reconstruction of the live image with the live image to determine a difference; and identifying a defect corresponding to the subject physical sample at a location of the determined difference. An unsupervised training of the generator neural network includes acquiring a set of images of the subject defect-free physical sample; executing a training phase including a plurality of training epochs, in which: training data images are synthesized by superimposing, onto each member of the set of images of subject defect-free physical sample as a respective parent image, defect image data; the synthesized training data images are reconstructed by the generator neural network which is iteratively trained to minimize a loss function between each reconstruction of the reconstructing of the synthesized training data images and the respective parent image of defect-free physical sample and increase an amount of difference between a training data image and the respective parent image caused by the superimposed defect image data from a minimum to a maximum as a function of a training condition.