Algorithms for Vision-Based Quality Control of Circularly Symmetric Components.

Area of Science:

• Industrial manufacturing automation
• Computer vision for quality control
• Artificial intelligence in quality inspection

Background:

• Industrial quality inspection is rapidly advancing with AI and vision techniques.
• Defect identification in circularly symmetric mechanical components with periodic elements presents unique challenges.
• Knurled washers are a specific case study for evaluating automated inspection methods.

Purpose of the Study:

• To compare the performance of a standard image analysis algorithm with a Deep Learning (DL) approach for defect detection in knurled washers.
• To evaluate accuracy and computational efficiency of both methods.
• To explore the potential extension of these methods to other symmetric components.

Main Methods:

• A standard algorithm utilizing pseudo-signals from grey-scale image analysis of concentric annuli was employed.
• A Deep Learning (DL) approach focused inspection on specific, potentially defective areas of the component profile.
• Performance metrics including accuracy and computational time were measured for both techniques.

Main Results:

• The standard algorithm demonstrated superior overall accuracy and faster computational time compared to the DL approach.
• The Deep Learning (DL) method achieved an accuracy exceeding 99% specifically for identifying damaged teeth.
• Both methods showed potential for application to other circularly symmetric components.

Conclusions:

• For general knurled washer inspection, the standard algorithm is more efficient and accurate.
• Deep Learning (DL) offers high precision for specific defect types, such as damaged teeth.
• The study highlights the trade-offs between different AI-driven quality inspection methods.