Deep Ensemble Learning-Based Sensor for Flotation Froth Image Recognition.

Area of Science:

• Mineral Processing and Materials Science
• Artificial Intelligence and Machine Learning

Background:

• Froth flotation is crucial for mineral separation, but traditional methods rely on subjective visual assessment for chemical dosage control, demanding significant operator expertise.
• This reliance on visual cues limits optimal control, potentially impacting concentrate grade and mineral recovery efficiency.

Purpose of the Study:

• To design and validate a novel deep ensemble learning-based sensor for automated froth flotation monitoring.
• To enhance mineral separation efficiency by providing objective, data-driven insights for chemical dosage adjustments.

Main Methods:

• Utilized K-fold cross-validation for robust training and validation of deep neural network (DNN) learners on froth flotation images.
• Developed a membership function to enhance DNN learner recognition accuracy based on validation performance.
• Implemented a Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) with F1 scores for optimal deep ensemble learning combination.

Main Results:

• The proposed deep ensemble learning sensor demonstrated high accuracy in recognizing flotation froth working conditions.
• Optimized chemical dosage adjustments were facilitated, leading to improved concentrate grade and mineral recovery.
• The sensor's effectiveness was validated in a real-world industrial gold-antimony froth flotation setting.

Conclusions:

• The developed deep ensemble learning sensor offers a superior, objective method for monitoring froth flotation processes.
• This technology assists operators in precise chemical dosage control, significantly boosting industrial mineral separation efficiency.
• The approach successfully integrates advanced machine learning techniques for practical mineral processing applications.