Defect Recognition Using Reconstructed Conformal Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing component inspection methods require large quantities of defective samples for training, making it difficult to detect all possible defects, as they can only recognize defects represented in the training data and are labor-intensive and subjective.
Innovation Solution
The method involves using a processor to generate an encoded array from a sample image without defects, performing stochastic data sampling, and comparing it to an input image to produce a residual image that highlights differences, allowing for the identification of defects without needing extensive defective sample data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large amounts of defective sample data are used for training, then the system can recognize defects represented in training data, but it cannot detect defects not represented in training data and requires extensive data collection
Solution Approach 1:
Instead of training the system to recognize defects by providing defective samples, the patent inverts the approach by training the system with only conforming (defect-free) samples. The system learns what normal components should look like, and then identifies defects as deviations from this learned norm. This inversion allows the system to detect any defect type not previously seen in training data, as long as it deviates from the conforming pattern.
Solution Approach 2:
The patent introduces a residual image as an intermediary representation that captures the difference between the input image and the reconstructed conforming image. This residual image serves as a mediator that highlights defect regions without requiring the system to have seen similar defects during training, enabling detection of novel defect types.
2Measurement precision
If manual inspection by human experts is used, then subjective interpretation occurs, but the process is labor-intensive and inefficient
Solution Approach 1:
The system performs self-inspection by automatically comparing input images against learned conforming patterns without requiring human expert intervention. The automated defect recognition system processes images independently, eliminating the need for manual inspection while maintaining consistent objective criteria for defect identification.
Solution Approach 2:
The patent replaces the mechanical process of manual visual inspection with an automated computational system. The image inspection controller uses algorithms to automatically detect defects, substituting human experts with a computational mechanism that operates faster and without subjective bias.
3Reliability
If classifiers are assigned to specific defect types, then the system can identify known defects, but it requires extensive training data for each defect class and cannot detect unseen defects
Solution Approach 1:
The patent creates a universal defect detection system that can identify any type of defect without requiring separate classifiers for each defect class. By learning the pattern of conforming components and detecting deviations from this pattern, the system achieves multi-functional defect detection capability that works across diverse defect types without additional training data requirements.
Data Source
AI summary
An image inspection computing device is provided. The device includes a memory device and at least one processor. The at least one processor is configured to receive at least one sample image of a first component, wherein the at least one sample image of the first component does not include defects, store, in the memory, the at least one sample image, and receive an input image of a second component. The at least one processor is also configured to generate an encoded array based on the input image of the second component, perform a stochastic data sampling process on the encoded array, generate a decoded array, and generate a reconstructed image of the second component, derived from the stochastic data sampling process and the decoded array. The at least one processor is further configured to produce a residual image, and identify defects in the second component.