Development of a bench system with capacitive sensor, sample compression, and TinyML for iron ore moisture...

Area of Science:

• Mineral Processing
• Instrumentation and Measurement
• Machine Learning in Geosciences

Background:

• Water content is critical in mineral processing, particularly for ore transport and beneficiation.
• Existing instrumentation for measuring ore moisture suffers from slow response times and inaccuracies.
• Operational issues in mineral processing can arise from inadequate sensing or control of water content.

Purpose of the Study:

• To develop a novel bench system for fast and accurate ore moisture measurement.
• To address the limitations of current instrumentation in the mineral sector.
• To implement a TinyML model for real-time ore moisture evaluation.

Main Methods:

• Utilized the real-dual-frequency method (RDFM) to measure electrical conductivity and relative permittivity.
• Integrated bulk density, chamber level, and compression torque as input variables.
• Developed a TinyML model, specifically tree-based models, trained on experimental data.
• Employed ore compression to minimize air bubbles and enhance measurement precision.

Main Results:

• The developed bench system demonstrated a fast response time and improved accuracy in ore moisture measurement.
• Ore compression was found to significantly enhance measurement accuracy.
• Tree-based TinyML models proved effective for rapid moisture estimation.
• Experimental validation was conducted using a dataset from a mining company's laboratory.

Conclusions:

• The novel bench system provides a viable solution for accurate and rapid ore moisture determination.
• TinyML models, particularly decision trees, are suitable for real-time ore moisture analysis in mineral processing.
• Optimizing measurement through ore compression leads to more reliable results.