Deep Learning Approach at the Edge to Detect Iron Ore Type.

Area of Science:

• Mining Engineering
• Artificial Intelligence
• Geotechnical Engineering

Background:

• Iron ore transfer in beneficiation plants carries a constant risk of material collapse.
• Current monitoring instrumentation is expensive, complex, and difficult to maintain.
• Early detection of potential landslides is crucial for operational safety and efficiency.

Purpose of the Study:

• To propose and evaluate an edge artificial intelligence (AI) system for early landslide risk detection in iron ore beneficiation.
• To develop a cost-effective and maintainable solution compared to existing instrumentation.
• To assess the feasibility of deep learning models deployed at the device edge for real-time monitoring.

Main Methods:

• Defined device edge parameters and a deep neural network (DNN) model for image analysis.
• Developed a prototype system for collecting and training the AI model with iron ore images.
• Compressed the DNN model for efficient deployment on the edge device.
• Integrated a real-time clock for synchronizing image data with plant process information.

Main Results:

• Field tests demonstrated the prototype's effectiveness under operational conditions.
• The AI model achieved a detection accuracy of 91% and a recall of 96%.
• Synchronization with process information ensured accurate image classification by specialists.

Conclusions:

• Edge AI offers a feasible and accurate solution for detecting iron ore landslide risk.
• The developed system provides a cost-effective and low-maintenance alternative to traditional instrumentation.
• This approach enhances safety by enabling early detection and prevention of material avalanches during ore transfer.