Deep learning-based anomaly detection framework for hydraulic support systems in mining safety.

Area of Science:

• Mining Engineering
• Artificial Intelligence
• Data Science

Background:

• Coal mine safety relies heavily on monitoring hydraulic support systems.
• Anomalous pressure variations in these systems can indicate potential failures.
• Existing monitoring methods may lack the sophistication to detect subtle anomalies.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for anomaly detection in coal mine hydraulic support pressure data.
• To improve the accuracy and efficiency of identifying abnormal pressure patterns.
• To enhance the safety and reliability of coal mine operations.

Main Methods:

• A deep learning framework combining bidirectional LSTM and CNN architectures was proposed.
• Gated residual connections and self-attention mechanisms were incorporated to capture temporal and local features.
• Data preprocessing included temporal resampling, missing value imputation, and normalization.
• The model was trained and tested on pressure data from 10 hydraulic supports in a western China mine.

Main Results:

• The proposed framework demonstrated excellent performance in anomaly detection tasks.
• Ablation studies confirmed the significant contributions of CNN layers (472% test loss increase) and gated residual mechanisms (352% test loss increase).
• The model effectively reconstructed samples and identified anomalies, distinguishing normal from abnormal pressure variations.

Conclusions:

• The novel deep learning framework offers a promising approach for anomaly detection in hydraulic support pressure data.
• Key components like CNN layers and gated residual mechanisms are crucial for model performance.
• Limitations include data quality dependency, fixed threshold strategies, and computational complexity, necessitating further research.