Damage Detection for Conveyor Belt Surface Based on Conditional Cycle Generative Adversarial Network.

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Belt conveyors are critical for coal mining, but their surfaces are susceptible to destructive foreign bodies.
• Traditional inspection methods are insufficient, leading to increased interest in machine learning-based non-destructive testing (NDT).
• Deep learning (DL) and generative adversarial networks (GANs) show promise for image analysis and defect detection.

Purpose of the Study:

• To propose a novel deep learning method for generating and classifying conveyor belt surface damages.
• To enhance classification performance with limited image datasets.
• To improve the stability and efficiency of coal mining operations through better belt inspection.

Main Methods:

• Development of a multi-classification conditional CycleGAN (MCC-CycleGAN) architecture.
• Utilizing a novel, improved CycleGAN design for enhanced classification.
• Training and testing the network on conveyor belt surface images with various defects.

Main Results:

• The MCC-CycleGAN successfully generated realistic images of conveyor belt surface defects.
• The proposed network demonstrated efficient classification of different types of belt damages.
• Experimental results validated the effectiveness of the DL approach for NDT in coal mining.

Conclusions:

• The novel MCC-CycleGAN method offers a powerful tool for generating and classifying conveyor belt damages.
• This DL-based NDT approach can significantly improve the inspection of critical mining equipment.
• The findings contribute to more efficient and reliable coal transportation systems.