Neural Network Vision Inspection for Defect Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Human visual inspection of machine-manufactured parts is unreliable due to human error and limitations in identifying defects, which can occur anywhere on the product and vary in size, posing a burden in terms of time and accuracy.
Innovation Solution
An automated machine vision-based system using a neural network for defect detection, comprising training with historical datasets, pre-processing images, and segmenting them into input patches for analysis through computational layers to generate probability scores and overall defect scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human visual inspection is used to detect defects, then the system is simple and easy to operate, but the reliability and accuracy of defect detection deteriorates due to human error and limitations
Solution Approach 1:
The patent replaces the mechanical human visual inspection system with an automated machine vision system using cameras, image processing algorithms, and defect detection software. This substitution eliminates human error and limitations while maintaining inspection functionality, thereby improving reliability without requiring complex manual operational procedures
Solution Approach 2:
The patent creates a digital copy of the visual inspection process through machine vision systems that capture images of manufactured parts and use image processing to replicate and enhance human visual capabilities. This copying approach allows automated analysis of defect patterns while eliminating the need for human inspectors, improving reliability while keeping the system relatively simple
2Productivity
If human inspectors manually examine every part for defects, then comprehensive inspection coverage is achieved, but the inspection time and productivity deteriorates due to the heavy burden on inspectors
Solution Approach 1:
The patent replaces slow manual human inspection with automated machine vision systems that can process multiple images simultaneously and analyze defect patterns rapidly. This substitution enables comprehensive inspection coverage of all manufactured parts without the time constraints of human inspectors, significantly improving productivity while reducing total inspection time
Solution Approach 2:
The patent implements preliminary defect detection by capturing images of parts during or immediately after manufacturing processes. This early detection approach allows for rapid identification of defects before parts move through the production line, enabling faster inspection cycles and improved productivity without compromising comprehensive coverage
3Difficulty of detecting and measuring
If human inspectors are trained to identify all types of defects, then detection capability improves, but the difficulty of operation and training burden increases
Solution Approach 1:
The patent replaces the need for human training and expertise with automated machine vision systems pre-programmed with defect detection algorithms. These systems can identify various defect types including scratches, dents, discolorations, and dimensional variations without requiring operators to undergo extensive training, thereby maintaining high detection capability while greatly improving ease of operation
Solution Approach 2:
The patent uses image processing techniques to transform raw visual data into enhanced images with adjusted parameters such as contrast, brightness, and edge enhancement. This parameter transformation makes defects more visually prominent and easier to detect automatically, improving detection capability while keeping the system easy to operate without requiring expert knowledge
Data Source
AI summary
Provided are various mechanisms and processes for automatic computer vision-based defect detection using a neural network. A system is configured for receiving historical datasets that include training images corresponding to one or more known defects. Each training image is converted into a corresponding matrix representation for training the neural network to adjust weighted parameters based on the known defects. Once sufficiently trained, a test image of an object that is not part of the historical dataset is obtained. Portions of the test image are extracted as input patches for input into the neural network as respective matrix representations. A probability score indicating the likelihood that the input patch includes a defect is automatically generated for each input patch using the weighted parameters. An overall defect score for the test image is then generated based on the probability scores to indicate the condition of the object.