Real-time monitoring and control of froth flotation

Many coal handling and preparation plants experience large daily variations in flotation recovery yet lack effective real time monitoring tools for flotation performance measurement, process control and optimisation. This report presents the development and site validation of a low-cost, physics-based drag sensor as a robust alternative for real-time mass pull diagnostics. Two site campaigns conducted in 2024 and 2025 benchmarked the drag sensor against existing froth cameras installed at the plant.

Results demonstrate that the drag sensor consistently achieves stronger correlations with flotation mass pull (R2 up to 0.77) and substantially higher signal-to-noise ratios (up to 8.6) than velocity-based measurements across a wide operating range. In addition, the drag sensor exhibited a linear, stable, and repeatable response across the full operating range throughout the two-year project duration.

The monitoring capability was further enhanced by integration with a Convolutional Neural Network (CNN) for real-time product ash prediction. Using an image dataset collected during a four-day site trial, the CNN model achieved a mean absolute error below 1.0% and the processing time was only 25 minutes. This enabled a high frequency quality feedback loop that is not attainable with conventional laboratory assays, which typically require 3-6 hours for sample preparation and ash analysis. By combining drag-sensor-based mass pull monitoring with AI-driven machine vision for product ash content, this project establishes a practical framework for integrated mass pull and product ash monitoring and control, enabling coal handling and preparation plans to improve process efficiency, operational stability, and the economic value of existing plant assets.