Autoencoder Defect Detection Error Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional defect detection methods in manufacturing are prone to errors in reconstructing product images, leading to missed detection of subtle defects and reduced accuracy.
Innovation Solution
A defect detection method utilizing an autoencoder to encode and decode product images, compare them with positive sample images, and input vectors into a Gaussian mixture model to determine reconstruction errors and probabilities, allowing for the selection of an error threshold to identify defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional defect detection methods are used to reconstruct product images, then the process is simple, but the accuracy of defect detection is reduced due to errors in image reconstruction
Solution Approach 1:
The patent applies preliminary action by pre-training an autoencoder model on a large dataset of product images before actual defect detection. The autoencoder learns normal product variations during the pre-training phase, so when detecting defects, it can accurately reconstruct normal products while identifying deviations as defects. This preliminary learning process resolves the contradiction by establishing a robust reconstruction baseline that improves detection accuracy without adding complexity during the actual detection phase.
Solution Approach 2:
The patent introduces an autoencoder model as an intermediary between the product image and the defect detection decision. The autoencoder acts as a mediator that reconstructs the input image, and the difference between the original and reconstructed images serves as the defect indicator. This intermediary approach resolves the contradiction by providing a sophisticated yet unified detection mechanism that improves accuracy while maintaining a relatively simple overall system architecture.
2Measurement precision
If traditional defect detection methods are used, then the system is easy to operate, but subtle defects are not detected
Solution Approach 1:
The patent replaces traditional mechanical or manual defect detection methods with an automated autoencoder-based system. The autoencoder automatically learns to distinguish normal product variations from defects through training, eliminating the need for manual threshold setting or complex detection rules. This substitution resolves the contradiction by providing high precision for subtle defect detection while maintaining ease of operation through automation.
Solution Approach 2:
The autoencoder model performs self-service by automatically adapting to different product types and defect characteristics through its training process. The system self-calibrates by learning from training data, eliminating the need for extensive manual configuration or expert intervention. This self-service capability resolves the contradiction by enabling subtle defect detection while keeping the system easy to operate and deploy.
3Measurement precision
If autoencoder-based defect detection is implemented, then detection accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the autoencoder model offline on a large dataset before deployment. This pre-training phase, which is computationally intensive, is performed once rather than during each detection operation. After pre-training, the model can perform rapid defect detection with minimal computational overhead. This resolves the contradiction by shifting computational burden to an offline phase, improving online detection accuracy while reducing real-time resource consumption.
Data Source
AI summary
A defect detection method based on an image of products and an electronic device can accurately determine the error threshold by determining the reconstruction error generated during image reconstruction and by determining the estimated probability generated by the Gaussian mixture model. The test error can then be compared with the error, since the test error and the error threshold are compared numerically, the existence of subtle defects are revealed in the product image, thereby improving the accuracy of defect detection.