A NIRS-based recognition of coal and rock using convolution-multiview broad learning system.

Area of Science:

• Mining Engineering
• Spectroscopy
• Artificial Intelligence

Background:

• Efficient coal production from thick seams requires precise control of the caving process.
• Deciding when to stop roof caving is critical for minimizing mining losses and maximizing cost recovery.
• Existing methods struggle with variable spectral data quality and lack integration across feature processing stages.

Purpose of the Study:

• To develop an innovative recognition system for on-site coal-rock identification using Near-Infrared Spectroscopy (NIRS).
• To create a robust coal-rock recognition method that overcomes limitations of existing techniques in spectral data acquisition and processing.
• To improve the decision-making process for stopping roof caving in thick coal seam mining.

Main Methods:

• Development of an on-site recognition system utilizing Near-Infrared Spectroscopy (NIRS) technology.
• Implementation of a coal-rock recognition method incorporating convolution and multi-view features into a Broad Learning System (BLS) model.
• The method is designed to be invariant to acquisition factors like granularity and sensor angles.

Main Results:

• The developed BLS model achieved a remarkable coal-rock recognition accuracy of 99.78%.
• Experimental deployment of the recognition system on a working face demonstrated an effective identification of the entire coal-caving process.
• The system achieved a practical recognition accuracy of 92.3% in real-world mining conditions.

Conclusions:

• The NIRS-based recognition system provides a viable solution for on-site coal-rock identification in longwall mining.
• The developed method effectively handles variations in spectral data quality and integrates features across processing stages.
• Accurate identification of the coal-caving process enables optimal stopping points, crucial for efficient mining operations.