Coal-gangue recognition via multi-branch convolutional neural network based on MFCC in noisy environment

Area of Science:

• Mining Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional coal-gangue recognition methods struggle with equipment noise, limiting accuracy and adaptability.
• The operation of mining machinery like shearers and conveyors generates significant noise during top coal caving.

Purpose of the Study:

• To develop a robust coal-gangue recognition system capable of operating accurately in noisy mining environments.
• To enhance the adaptability and recognition accuracy of coal-gangue identification systems.

Main Methods:

• Utilized Mel Frequency Cepstrum Coefficients (MFCC) smoothing to enhance sound pressure features.
• Developed a multi-branch convolution neural network (MBCNN) model incorporating smoothed MFCC features.
• Acquired extensive sound pressure signal datasets from laboratory and on-site operations of various mining devices.

Main Results:

• The proposed method demonstrated higher correct recognition accuracy compared to traditional approaches.
• Experiments on noiseless, single-noise, and simulated site datasets confirmed the method's superior robustness.
• The system successfully recognized both coal/gangue status and operational states of site equipment.

Conclusions:

• The MBCNN and MFCC smoothing-based approach significantly improves coal-gangue recognition in noisy conditions.
• This method offers a more adaptable and accurate solution for real-time monitoring in mining operations.
• The system's capability extends to recognizing equipment operational states, providing additional operational insights.