HoStB-DVGNN: A Flotation Fault Recognition Method Using Higher Order Spatial-Temporal Block and Dual-Stream...

Area of Science:

• Mineral Processing
• Artificial Intelligence
• Chemical Engineering

Background:

• Accurate fault pattern recognition is vital for efficient froth flotation operations.
• Dynamic flotation data present challenges like high dimensionality, nonlinearity, noise, and uncertainty.
• Existing methods struggle with the complexity of flotation process data, impacting fault recognition accuracy.

Purpose of the Study:

• To develop an advanced fault pattern recognition method for froth flotation.
• To address the challenges posed by dynamic, complex flotation process data.
• To improve the accuracy and robustness of fault detection in industrial flotation.

Main Methods:

• Proposed a fault pattern recognition method using higher-order spatial-temporal blocks (HoStB) and dual-stream variational graph neural networks (DVGNNs).
• Constructed HoStB from key frame images to capture comprehensive flotation conditions.
• Developed a DVGNN with apparent feature and HoStB streams for dynamic froth information extraction.
• Implemented a bilateral self-supervision mechanism with a variational autoencoder (VAE) to boost generalization.

Main Results:

• The HoStB-DVGNN method effectively extracts dynamic froth information.
• The bilateral self-supervision mechanism significantly enhances model generalization performance.
• Extensive experiments demonstrated the method's effectiveness and robustness on benchmark and real-world flotation data.

Conclusions:

• The proposed HoStB-DVGNN method offers a robust solution for fault pattern recognition in froth flotation.
• This approach successfully handles the high dimensionality, nonlinearity, and uncertainty of flotation process data.
• The method validates its effectiveness and robustness in real-world industrial applications, reducing production risks.