GAN-Based Defect Detection for Injection Molding Quality Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning/AI-based product surface inspection methods face challenges in accuracy and reliability due to complex manufacturing environments, low defect rates, and variability in defect characteristics, making it difficult to collect and apply effective AI models for defect detection across different facilities.
Innovation Solution
A deep learning-based quality inspection method and system that uses only non-defective manufactured product images for training, extracting attributes like objectness, brightness, and contrast to calculate quality scores, and generating fake defective features for learning, allowing for defect detection without requiring extensive labeling or prior defect knowledge, and can be applied across various manufacturing fields.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional AI classification methods are used for defect detection, then the system can identify defective products, but the AI accuracy becomes unstable due to incorrect application of classification criteria at the manufacturing site
Solution Approach 1:
The patent creates a virtual copy of the normal product appearance through GAN-generated images. Instead of copying defect patterns which vary across facilities, the system learns the normal state and generates synthetic defective images by modifying this normal template. This approach stabilizes AI accuracy by using a consistent reference (the normal product model) rather than relying on facility-specific defect classification criteria that are difficult to apply correctly at manufacturing sites.
2Measurement precision
If AI models are trained using actual defective product data, then the detection accuracy improves, but it becomes difficult to collect sufficient defective data due to low defect rates in actual production
Solution Approach 1:
The system uses GANs to generate synthetic defective product images that copy the characteristics of real defects without requiring actual defective samples. The generator creates realistic defect patterns by learning from normal product images and injecting synthetic defect features, thereby providing sufficient training data volume while maintaining detection accuracy.
Solution Approach 2:
The system performs preliminary learning of the normal product state before defect detection. By first training the GAN on normal products to establish a baseline model of what a defect-free product should look like, the system prepares in advance for defect identification. This preliminary action enables the subsequent generation of synthetic defective images that can be used for comprehensive training without requiring actual defective samples.
3Adaptability or versatility
If the system is designed to handle various defect types with different characteristics (size, length, color), then it can detect diverse defects, but it becomes difficult to set a single criterion for all facilities and train a unified network
Solution Approach 1:
Instead of creating separate models for each defect type and facility, the system uses a single GAN model that learns the normal product appearance once and can generate synthetic images for various defect types. The GAN copies the normal product structure and selectively introduces different defect characteristics during image generation, allowing one unified network to handle diverse defects across all facilities without requiring facility-specific training criteria.
4Measurement precision
If extensive defect labeling and prior defect knowledge are required for AI training, then the model can be trained accurately, but the deployment time and complexity increase significantly
Solution Approach 1:
The system copies normal product images to generate synthetic defective images, eliminating the need for extensive manual labeling of actual defects. By using unsupervised or self-supervised learning on normal products and generating defects synthetically, the system achieves accurate training without time-consuming data annotation processes, significantly reducing deployment time while maintaining model accuracy.
Data Source
AI summary
Embodiments relate to a deep learning/AI-based product surface quality inspection system which is accurate and reliable in product quality inspection which is a core task in an injection process among various manufacturing fields. The system can provide, to a user, better performance than non-defective/defective manufactured product classification methodology which is an existing commonly used method through a method and a system considering characteristics of a factory environment and an actual product production process for all pipelines of product quality inspection by using only a non-defective manufactured product image unlike most injection process surface inspection AIs developed to date.