An Image-Based Data-Driven Model for Texture Inspection of Ground Workpieces.

Area of Science:

• Materials Science
• Manufacturing Engineering
• Computer Vision

Background:

• Manual quality inspection of ground workpieces is time-consuming and lacks consistency.
• Automating inspection using data-driven models offers a more efficient and reliable solution.
• Vision-based datasets are crucial for developing automated inspection systems.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) model for automated quality inspection of ground workpieces.
• To classify abrasive belt grit number, estimate surface roughness, and determine abrasive belt wear.
• To identify the optimal lighting conditions for accurate image-based inspection.

Main Methods:

• Utilized a convolutional neural network (CNN) with transfer learning on a dataset of 750-1000 surface raw images per task.
• Tested three lighting conditions: external coaxial white light, high-angle ring light, and external coaxial red light.
• Developed models for grit number classification, surface roughness estimation, and abrasive belt wear classification.

Main Results:

• The CNN model achieved high accuracy (≥0.9) in classifying textures from abrasive surface images.
• External coaxial white light proved to be the most effective illumination source for the inspection tasks.
• The developed model for abrasive belt wear classification can serve as an effective abrasive belt life estimator.

Conclusions:

• Deep convolutional neural networks are highly effective for automated visual inspection in grinding processes.
• Optimized lighting conditions are critical for achieving high accuracy in vision-based quality control.
• The proposed automated system enhances efficiency and reliability in manufacturing quality assessment.