Using automated mineralogy data in process simulation with USIM PAC

Automated quantitative mineralogy measures the composition, size and liberation of hundreds of thousands of individual particles. Process simulation predicts how a plant will treat an ore. A research project carried out by INP-Toulouse, BRGM and CASPEO explores how to use automated mineralogy data directly as input for USIM PAC simulations, with the measurement uncertainty carried through to the results. The first results are promising. Each predicted grade and recovery comes with its probability distribution. Engineers see the expected performance and the risk around it. A paper presented at the IMPC 2026 congress in Cap Town (South Africa).

Why does ore mineralogy drive plant performance?

Ore variability is the main source of plant performance fluctuations. Two batches with the same copper grade can behave differently in flotation if the mineralogy is different. Liberation, mineral associations and particle size decide what a separation step can recover.

A simulation that only uses average grades misses this. It predicts the behaviour of an average ore that the plant never receives. A simulation based on mineralogy predicts the behaviour of the real feed.

What does automated mineralogy bring to process simulation?

Automated mineralogy analyses individual particles. It quantifies the composition, size and liberation of each one. On the case study presented in the paper, the MLA analysed about 300,000 particles. This is the most detailed description of an ore available today.

A simulation based on chemical assays cannot use this level of detail. It follows elements through the plant, such as % Cu, and loses the information on how minerals are distributed between particles.

USIM PAC was designed differently. Since its creation 40 years ago, our mineral processing software describes ore as populations of particles, each class with its own size, mineral composition and liberation state. Automated mineralogy data therefore fit its structure, and the simulator predicts how each particle class behaves at every stage of the circuit.

To use the raw MLA data directly, two points still had to be solved:

• The first is uncertainty: how do you quantify the sampling uncertainty embedded in the MLA data? An MLA measurement analyses a sample. That sample carries its own uncertainty, and the simulation inherits it.
• The second is format: how do you feed individual particles into a simulator? USIM PAC describes feed material as particle classes. Hundreds of thousands of measured particles must be grouped into classes without losing useful information.

These two questions are the starting point of a research project carried out by INP-Toulouse, BRGM and CASPEO.

How the method works

This research is the PhD work of Julien Guidihoumme at INP-Toulouse. CASPEO opened its doors to Julien and supports the project.

• Mineralogical uncertainty

The MLA particle data are resampled with the bootstrap method. The team compared this approach with the method developed by Lyman to assess sampling error from MLA data. With 5,000 resampling iterations, both methods gave close uncertainty estimates for all mineral phases.

• Particle size uncertainty

MLA has a size analysis threshold for very fine particles, around 2 µm. In the case study, these fines were significant. The method adds Monte Carlo simulation of the measured particle size distribution to cover this gap.

• Simulation in USIM PAC

A clustering method groups the MLA particles into classes that USIM PAC can use as feed. Both sources of uncertainty are then propagated through the simulation.

What this research opens for mining projects

Simulation with uncertainties gives access to the complete probability distributions of grades and recoveries. Engineers can compare process options on their expected performance and on their risk.

Benefits of advanced simulation for mining operations:

• Feasibility studies: evaluate a new ore body or a tailings deposit with a realistic range of results
• Flowsheet selection: compare two process routes on performance and on the confidence you can place in each
• Plant optimisation: understand how mineralogical variability in the feed affects the plant

CASPEO invests in this research to bring these capabilities to USIM PAC users.

Authors

• Julien Hospice Guidihoumme, PhD student – Toulouse INP (author)
• Kathy Bru, Project manager & Research engineer in process engineering – BRGM
• Stéphane Brochot, PhD. – CASPEO
• Florent Bourgeois, PhD. –  Toulouse INP-ENSIACET

Stéphane Brochot will present Julien’s work at IMPC 2026, the 32nd International Mineral Processing Congress, organised by the Southern African Institute of Mining and Metallurgy (SAIMM), in Cape Town, October 18-22. The paper is titled “Development of an original method for the direct use of automated quantitative mineralogy data for the simulation of mineral processing processes with uncertainties.”