Optimising mineral processing with machine learning is at the heart of DEKA Dynamics’ approach to helping operators turn complex plant data into performance gains.
As part of its program to increase exposure to Papua New Guinea (PNG) mining professionals, DEKA Dynamics, through one of the company’s experts in the field, Doctor Stephen Rayward, offers insight into the impact of machine learning in optimising mineral processing plants through his courses in Port Moresby in March and again in July.
Modern mineral processing plants are among the most data-rich industrial environments in operation today. Thousands of sensors continuously record pressures, flow rates, densities, power draw and equipment states, while routine plant surveys generate additional information such as particle size distributions, metallurgical recoveries and chemical assays.
Despite this abundance of data, converting raw measurements into reliable, actionable insight remains a major engineering challenge for operators seeking to improve performance, reduce costs and increase sustainability.
Mineral processing optimisation has traditionally relied on techniques such as mass balancing, statistical reconciliation and periodic metallurgical test work. These methods remain essential foundations of good metallurgical practice, providing consistency checks and ensuring data integrity.
However, on their own, they often struggle to capture the full complexity of modern processing circuits, particularly in plants that experience variable ore characteristics, changing operating conditions and increasingly tight economic and environmental constraints.
In recent years, machine learning and information-driven optimisation approaches have emerged as powerful complements to conventional metallurgical tools. Rather than replacing first-principles understanding, these techniques build on it, using data to reveal patterns, interactions and inefficiencies that are difficult to detect through manual analysis alone.
When combined with flow sheet knowledge and unit operation fundamentals, machine learning models can quantify the true performance of individual pieces of equipment and their contribution to overall plant outcomes.
A key enabler of this approach is the development of plant-wide digital twins.
A digital twin is a virtual representation of the processing plant that integrates real-time operational data with process models, metallurgical relationships and physical constraints. Unlike static simulations, digital twins continuously update as new data becomes available, allowing engineers to track performance, diagnose issues and test alternative operating strategies in a risk-free environment.
For example, a digital twin can help identify whether a decline in recovery is driven by upstream grinding inefficiency, poor classification performance, reagent imbalance or downstream flotation constraints. Machine learning algorithms can analyse historical and real-time data to determine which variables most strongly influence performance, even when relationships are non-linear or obscured by noise. This enables more informed decision-making and more targeted interventions.
One of the major advantages of digital twins is the ability to experiment virtually. Engineers can assess the impact of changes in operating set-points, equipment configurations or feed characteristics without disrupting production. Bottlenecks can be identified and prioritised, control strategies refined, and trade-offs between throughput, recovery and energy consumption explored before changes are implemented on the plant floor. In an industry where downtime is costly and risk tolerance is low, this capability represents a significant step forward.
The financial implications of data-driven optimisation are compelling. While large step changes in performance are rare, even modest improvements can translate into substantial value.
Reductions of around five per cent in operating cost per unit of valuable mineral recovered are often achievable through improved stability, reduced variability and better utilisation of existing assets. These gains may come from lower energy consumption, reduced reagent usage, improved recoveries or a combination of all three. Importantly, such improvements are typically realised without major capital expenditure.
Beyond economics, machine-learning-driven optimisation also supports broader industry objectives. More efficient plants consume less energy and water per tonne of product, reduce waste generation and improve overall environmental performance. As regulatory scrutiny and stakeholder expectations increase, the ability to demonstrate data-backed improvements in efficiency and sustainability is becoming a competitive advantage.
Unlocking this value, however, requires more than just advanced software. It demands engineers who understand the process fundamentals and the data science techniques underpinning modern optimisation tools. Bridging this gap between metallurgy and analytics is a growing challenge for the industry, particularly as experienced practitioners retire and digital technologies evolve rapidly.
“Once it can be estimated how particles are being processed, this provides an instantaneous ‘snapshot’ of the unit process and by considering how these snapshots change for different operating conditions, models can be constructed directly from the data,” Rayward told PNG Mining.
Recognising this need, DEKA Dynamics offers targeted training courses in mass balancing, machine learning and information theory, specifically tailored to mineral processing applications. These courses are designed to equip metallurgical and process engineers with practical, industry-relevant skills that can be applied directly to real plant data and operating challenges.
By strengthening the capability of engineering teams, operations can maximise the return on their data investments and accelerate the adoption of data-driven optimisation strategies.
As mineral processing plants continue to grow in complexity and data availability, the integration of machine learning, digital twins and sound metallurgical principles will become increasingly central to operational excellence.
Organisations that invest early in technology and people will be best positioned to extract maximum value from their assets – safely, sustainably and profitably – in an increasingly competitive mining landscape.
This article appeared in the February/March edition of PNG Mining.
