A Common Knowledge-Driven Generic Vision Inspection Framework for Adaptation to Multiple Scenarios, Tasks, and...

Area of Science:

• Industrial Manufacturing
• Computer Vision
• Artificial Intelligence

Background:

• The shift to customer-centric manufacturing increases product complexity and quality demands, challenging traditional machine vision.
• Existing AI research often overlooks composite tasks, method traceability, and knowledge transfer between different inspection scenarios.
• A gap exists in generic, adaptable vision inspection frameworks for complex industrial products.

Purpose of the Study:

• To propose a common, knowledge-driven, generic vision inspection framework for standardized product inspection.
• To address challenges in composite tasks, method traceability, and knowledge communication in industrial vision systems.
• To enable adaptive metrics and information decoupling for complex product inspection.

Main Methods:

• Developed a framework for progressive alignment of task-related object perception using multi-granularity and multi-pattern approaches.
• Abstracted inspection as a reconfigurable process of multi-sub-pattern space combination mapping and difference metrics.
• Implemented strategies for knowledge improvement and accumulation from historical data for continuous pipeline enhancement.

Main Results:

• Generated a detection pipeline for complex products, demonstrating continuous improvement via failure tracing and knowledge enhancement.
• Achieved superior pose estimation (1.034°, 52.308 mm) and detection rate (0.462 to 0.927) compared to state-of-the-art deep learning methods.
• Verified adaptability across different imaging methods and industrial tasks, highlighting the importance of knowledge commonalities.

Conclusions:

• The proposed knowledge-driven framework offers a standardized and adaptable approach to complex industrial product inspection.
• Adaptability in vision inspection is achieved through mining inherent knowledge commonalities, multi-dimensional accumulation, and reapplication.
• The framework effectively addresses limitations of traditional machine vision and current AI approaches in complex manufacturing environments.