Adaptive Diffusion Sampling for Synthetic Defect Image Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The challenge in manufacturing electronic devices is the inefficiency and high cost of defect identification and classification due to a lack of balanced defect and defect-free image data for training AI models, particularly in newer products with limited defect samples, leading to potential human error and system inefficiency.

Innovation Solution

Utilizing a generative AI model to learn defect distributions from source products and generate synthetic defect images for target products, enhanced by an adaptive guided sampling method to improve defect classification accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional defect detection methods are used, then manufacturing processes can be maintained, but defect identification and classification become inefficient and costly due to lack of balanced training data

Engineering Contradiction:

Improvedefect identification efficiencyVSAvoidbalanced defect and defect-free image data

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent uses generative AI models to create synthetic defect images that copy and replicate defect characteristics from source products. These synthetic images serve as virtual copies to balance the training data, allowing the system to learn defect patterns without requiring additional real defect samples from every product type.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent introduces synthetic defect images as an intermediary between real defect samples and the AI training process. These synthetic images act as a mediator that bridges the gap between limited real data and the need for comprehensive training data, enabling efficient defect detection without direct reliance on abundant real defect samples.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If more real defect images are collected for training, then classification accuracy improves, but the process becomes more costly and time-consuming

Engineering Contradiction:

Improvedefect classification accuracyVSAvoidtime to collect and process training data

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by generating synthetic defect images before the actual training process. The generative AI model pre-create a balanced dataset of synthetic defect images that can be immediately used for training, eliminating the time-consuming process of collecting and processing real defect images from multiple product types.

Inventive Principle:

Principle #10Preliminary action

3Productivity

If synthetic defect images are generated without adaptive sampling, then generation speed increases, but defect classification accuracy decreases

Engineering Contradiction:

Improvesynthetic image generation speedVSAvoiddefect classification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent applies adaptive sampling that dynamically adjusts the generation process based on real-time feedback from the generative AI model. The system adapts sampling parameters and denoising strength based on the specific defect type and image characteristics, optimizing both generation speed and classification accuracy for each synthetic image.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes key parameters including denoising strength and sampling steps based on the guidance strength calculated from classifier probabilities. By dynamically adjusting these parameters, the system optimizes the balance between generation speed and the quality of synthetic images for accurate defect classification.

Inventive Principle:

Principle #35Parameter changes

4Measurement precision

If human inspection is used for defect detection, then detailed examination is possible, but human error and high cost are introduced

Engineering Contradiction:

Improvedefect detection detailVSAvoidhuman error rate

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent replaces the mechanical human inspection system with an AI-based automated system. The generative AI model creates synthetic defect images that train the AI classifier, which then performs automated defect detection and classification, eliminating human error and reducing costs while maintaining or improving detection precision through consistent algorithmic analysis.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20260017762A1Adaptive sampling methods for diffusion models for synthetic defect image generation

Publication Date: 2026.01.15 SAMSUNG DISPLAY CO LTD

AI summary

A method may include applying noise to a first real image to generate a first noisy image. Then, the method may include generating a first synthetic image corresponding to an estimate of a first class of synthetic image, and computing a guidance strength of the first synthetic image based on probabilities determined from a multi-class classifier, wherein the probabilities may include a first probability of the first class of synthetic image and a second probability of a second class of synthetic image, and denoising an amount of noise determined based on the guidance strength.