Generative Image Inspection for Irregular Internal Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect inspection methods, particularly for irregularly positioned, sized, or shaped defects, are labor-intensive and subject to varying quality based on inspector skill, making automated detection challenging.
Innovation Solution
An inspection device utilizing a generative model constructed through machine learning to automatically determine defect presence or absence by generating a restored image from an inspection image, using training data from defect-free inspection targets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual inspection is used to determine defect presence, then detection accuracy for irregular defects is improved, but inspection time and labor requirements increase significantly
Solution Approach 1:
The patent replaces the mechanical visual inspection system with an automated image processing system using deep learning models. The system processes ultrasonic inspection images through trained neural networks that automatically identify defects, eliminating the need for manual visual inspection while maintaining high detection accuracy for irregular defects.
Solution Approach 2:
The patent transforms the inspection approach by changing from manual parameter assessment to automated feature extraction. The deep learning model automatically extracts relevant features from images and makes determination decisions based on learned patterns, fundamentally changing how inspection parameters are evaluated.
2Measurement precision
If visual inspection is used to determine defect presence, then detection accuracy for irregular defects is improved, but labor requirements increase significantly
Solution Approach 1:
The patent replaces the mechanical visual inspection system with an automated image processing system using deep learning models. The system processes ultrasonic inspection images through trained neural networks that automatically identify defects, eliminating the need for manual visual inspection while maintaining high detection accuracy for irregular defects.
Solution Approach 2:
The inspection system performs self-service through automated defect detection. The deep learning model independently processes images, extracts features, and makes determination decisions without human intervention, making the system self-sufficient and highly productive.
3Productivity
If automated detection methods are used, then inspection speed is improved, but detection accuracy for irregular defects deteriorates
Solution Approach 1:
The patent introduces deep learning models as intermediaries between the raw inspection images and the final defect determination. These models serve as intelligent mediators that automatically extract meaningful features from images and make accurate defect detection decisions, bridging the gap between automated processing and high detection accuracy.
Solution Approach 2:
The patent performs preliminary training of deep learning models using labeled defect data before actual inspection. This preliminary action enables the model to learn defect patterns and characteristics in advance, so that during actual inspection, the model can quickly and accurately detect irregular defects without manual intervention.
Data Source
AI summary
Determination of presence or absence of a defect having irregular position, size, shape, and/or the like in an image are made automatically. An inspection device includes: an inspection image obtaining section that obtains an inspection image used to determine presence or absence of an internal defect in an inspection target; and a defect presence/absence determining section that determines presence or absence of a defect with use of a restored image generated by inputting the inspection image into a generative model constructed by machine learning that uses, as training data, an image of an inspection target in which a defect is absent, the generative model being constructed so as to generate a new image having a similar feature to that of an image input into the generative model.