Inline Quality Inspection Auditing With ML Defect Alert Review
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated quality inspection systems face challenges in accurately identifying good and bad material, leading to false alarms, inconsistent quality thresholds, and inefficient overrun management due to operator skill variability, resulting in increased costs and time-consuming manual audits.
Innovation Solution
A machine vision system coupled with a neural network machine-learning algorithm that automatically evaluates product images, generates defect alerts, and learns from human inputs to improve defect classification, reducing the need for manual intervention and ensuring consistent quality control across production lines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated inspection systems are configured with high sensitivity to detect all defects, then defect detection capability is improved, but false alarm rate increases
Solution Approach 1:
The system performs preliminary actions by capturing images of all products before final defect classification, storing them in a database for later review. This allows the inspection system to initially flag all potential defects with high sensitivity, then subsequently review them to filter out false alarms, thus resolving the contradiction between high detection sensitivity and low false alarm rate.
Solution Approach 2:
The patent introduces an intermediary review process where captured images are stored in a database and reviewed by operators or additional analysis before final defect classification. This intermediary step acts as a buffer between the high-sensitivity detection system and the final defect determination, allowing false alarms to be filtered while maintaining high initial detection sensitivity.
2Measurement precision
If manual audit processes are implemented to verify defect alerts, then quality control accuracy is improved, but production time and costs increase
Solution Approach 1:
The patent extracts the audit and review process from the main production line by capturing images during production and storing them in a database for separate review. This allows the primary production process to continue uninterrupted while quality verification occurs in parallel, thus maintaining high quality control accuracy without significantly impacting production time.
Solution Approach 2:
The system creates digital copies of products through image capture and stores them in a database. These copies can be reviewed, analyzed, and audited without affecting the physical production process. This allows multiple review operations on the same product data simultaneously, improving quality control accuracy while maintaining production efficiency.
3Adaptability or versatility
If different operators set inspection thresholds based on their expertise, then quality judgment flexibility is improved, but quality consistency deteriorates
Solution Approach 1:
The patent creates a universal database system that stores captured images and defect information that can be accessed and reviewed by multiple operators. While operators can still apply their expertise in the review process, they all work from the same standardized captured images and criteria, ensuring consistency. The system serves multiple functions: capturing, storing, reviewing, and standardizing quality judgments across different operators.
4Measurement precision
If physical ejection of defective parts is implemented, then defect separation is improved, but production line complexity increases
Solution Approach 1:
Instead of physically ejecting defective parts, the system creates and stores digital copies (images) of all products in a database. Defective products are identified through review of these digital copies, and their locations are recorded. This eliminates the need for complex physical ejection mechanisms while maintaining accurate defect separation through digital tracking and marking.
Data Source
AI summary
A method and system for implementation of quality control includes a machine vision system for capturing images of instances of a product, and a computer system including computer memory containing machine-readable instructions executable by a processor. The processor evaluates the quality control images for a plurality of potential quality defects in the product and generates a defect alert associated with a captured image in which at least one potential quality defect is identified. Information about each defect alert is stored in a database log file in the computer memory. A neural network machine learning algorithm processes the database log file by, in a learning phase, receiving human-initiated input accepting or rejecting each defect alert and storing the human-initiated input in the log file, and, in an automated phase, automatically accepting or rejecting at least some defect alerts without performing the step of receiving human initiated input.