Procedural Defect Modeling for Virtual Surface Inspection Environments.

Area of Science:

• Manufacturing Engineering
• Computer Vision
• Data Science

Background:

• Automated visual surface inspection systems require extensive defect data for hardware configuration and model training.
• Acquiring sufficient, representative, and accurately annotated defect samples is a significant challenge in industrial settings.

Purpose of the Study:

• To present a method for simulating geometrical defects in virtual environments to address data scarcity.
• To enable inspection planning experts to assess defect visibility with different hardware setups.
• To facilitate the creation of training-ready datasets with pixel-precise annotations.

Main Methods:

• Development of parameterized models for adaptable simulation of geometrical defects using procedural methods.
• Integration of these models into virtual surface inspection planning environments.
• Generation of synthetic images with corresponding pixel-precise annotations.

Main Results:

• Successful simulation of diverse geometrical defects in virtual product models.
• Demonstrated utility for assessing defect visibility across various acquisition hardware configurations.
• Creation of synthetic datasets suitable for training defect detection models.

Conclusions:

• Parameterized procedural models offer an effective solution for generating synthetic defect data.
• Virtual simulation aids in both inspection system planning and the development of robust AI models for surface inspection.