African Mining October 2026 | Page 18

• TECHNOLOGY

HOW AI AND NEW SENSORS ARE CHANGING PGM MILLS

By Sharon Mdaka
AI, new instrumentation and advanced control aim to deliver double-digit gains in milling and flotation

As South African PGM producers face rising energy costs, harder ore and volatile prices, Mintek argues the path forward is integrated: better measurement, smarter control, and AI that adapts in real time.

At its PGM Industry Day, South Africa’ s national mineral research organisation, Mintek, showcased three technologies working as one system, from the instruments that‘ see’ the process, to the advanced process control that stabilises it, to the AI that optimises it. Starting with the Control Foundation: 20 Years of FloatStar™ and MillStar™. The organisation’ s executive manager for Measurement and Control, David Phillpotts, explained that the journey in measurement and control started with instrumentation during the early years of the institution or in the 80s. The big leap came with industrial model predictive control and the rise of the personal computer in the 2000s.
That investment paid off in flotation. As the industry moved to process the UG2 reef, 63 kilotons of chromite-locked PGMs, circuits became larger and far more complex.“ The industry saw the benefits from our FloatStar™ control system, and currently it is installed on 35 local flotation plants,” Phillpotts said.
Based on metallurgical trials, the institution estimates that the software has unlocked“ extra tens of billions of rands in extra PGM recovery” on top of the UG2 resource. Today, the focus is on‘ missing measurements’, new sensors and algorithms needed for advanced control. That includes a decade-long effort to accurately measure xanthate in solution, plus work in machine vision and other process measurements.“ We continue to keep researching new ways to
Supplied by Mintek unlock value on PGM plants,” Phillpotts said.
According to Thabiso Chautsane, senior Process Control Engineer, if control needs data, the organisation is building the sensor to provide it. Chautsane introduced Granumetrix™, an on-stream particle size analyser that automates the traditional sieving method. The motivation is simple: milling is energy intensive, and you can’ t optimise what you can’ t measure. Existing laser and light diffraction instruments cost R1-million – R5-million, need frequent calibration, and require ideal conditions and diluted samples.
The institution is now working with other divisions on additional AI-driven online advisors and optimisers.
The product takes a different approach.“ Sieving is a technique that stood the test of time. It’ s known to be the most reliable
method,” Chautsane said. The unit delivers results every 15 – 20 minutes, handles larger samples, and works from millimetres down to micrometres without dilution. Testing of the alpha prototype showed wet-sieving results deviated by only 5 % from manual methods. With patent and trademark protection filed, the organisation is now analysing site data to fund a beta unit.“ By creating instruments such as these, we will enable the optimisation of this process,” Chautsane said.“ Obviously, there will be more money for any operation that uses this system.”
Dr William Shipman, 4IR R & D Leader, said that on top of control and measurement sits optimisation. Shipman detailed the institution’ s AI module for milling circuits, built on the MillStar™ suite. Stabilising advanced process control delivers 1 – 4 % throughput. The AI layer pushes further.“ When you start really doing optimisation, we’ ve seen in some cases 16 % improvement in throughput,” Shipman said. The system estimates“ grind curves” in real time, how mill filling, speed, power and grind relate. As ore hardness changes, the AI makes small, safe adjustments to hunt for the maximum, while modelling constraints like particle size to avoid upsets.
Tested on a difficult circuit already running MillStar™, on / off trials showed a clear throughput shift. The objective can be tuned for throughput, net smelter return, energy efficiency, or carbon footprint.“ These circuits are energy hogs, and you can’ t see inside them,” Shipman said.“ This gives us a way to adapt as the ore changes.” The institution is now working with other divisions on additional AI-driven online advisors and optimisers.
Together, the three pieces tell a story: Phillpotts’ control platforms stabilise the plant, Chautsane’ s Granumetrix™ provides the missing real-time data, and Shipman’ s AI turns that data into money. For an industry under pressure, it’ s a fully South African tech stack aimed at one goal: doing more with less. •
16 • African Mining • October 2026 www. africanmining. co. za