The first report in RAMMS’s In Mechanised Mining series asked a deliberately broad question: which ideas are shaping mechanised-mining research, and which problems are South African mines actually trying to solve?
The authors compared recent academic reviews with industry reports, conference discussions and interactions with mining stakeholders. That approach matters because research novelty and operational urgency are not always the same thing. The useful opportunities are often found where the two overlap.
Six signals from research
The academic literature pointed to six connected directions.
First, machine-learning methods should be combined with physical models. Mining systems operate under changing loads, geology and environmental conditions. Physics supplies structure where field data are sparse; learned components can describe effects that are difficult to formulate directly.
Second, research must move from detecting a pattern to creating an actionable insight. An anomaly score is not yet a maintenance decision, and a predicted hazard is not yet a control intervention.
Third, validation and standardisation are essential. A model that works on one dataset or machine needs a transparent domain of validity before it can be trusted elsewhere.
Fourth, interpretability remains a constraint. Mine operators and engineers need to understand why a model recommends an action, particularly when safety or production is at stake.
Fifth, transfer learning offers a practical response to limited labelled mine data. A model can first learn from simulation, laboratory tests or another asset, then be refined using the smaller amount of data available from the target operation.
Finally, broader artificial-intelligence methods may help integrate related tasks—but only if they remain grounded in reliable measurements, engineering context and a defined decision.
What industry was saying
The industry evidence showed that many mines were already digitised and connected. The limiting question was no longer simply whether data existed, but whether it was accessible and whether the operation could convert it into value.
Asset management emerged as one of the strongest value opportunities. Condition monitoring, predictive maintenance and better reliability decisions can improve production without requiring an entirely new mining method.
Access to data, however, remained difficult. Equipment manufacturers often control important machine information, while mines and researchers may receive interpreted outputs rather than the underlying data. This can limit independent analysis, interoperability and research replication.
Technical capacity is another bottleneck. Even where data are available, mining operations compete with other industries for the people needed to engineer data pipelines, models and decision-support systems.
Safety was a particularly strong South African priority. Collision-prevention systems illustrate the tension between safety and productivity: immature or overly sensitive interventions may stop equipment even when the real risk is low. The answer is not weaker safety protection, but better sensing, localisation, context and control logic.
The intersection became a programme roadmap
The report’s lasting contribution is the way it connects these signals. Hybrid models address the need for data efficiency and interpretability. Digital twins provide a structure for connecting models to physical assets and actionable decisions. Machine vision and localisation can add context to collision avoidance. Maintenance models convert equipment data into asset-management choices.
Those links helped sharpen the three RAMMS research thrusts:
- Cutting and Drilling Hard Rock;
- Productivity and Maintenance Optimisation of Mining Equipment; and
- Utilisation, Performance and Condition Monitoring of Mechanised Mining Equipment.
Digital twins are not confined to the third thrust. They can connect rock and tool measurements to cutting or drilling models, supply equipment capability and condition to productivity simulations, and support real-time utilisation, performance and maintenance decisions.
The central lesson
No technology theme stands alone. A technically strong algorithm has limited value when its inputs are inaccessible, its uncertainty is unknown, its recommendation is not interpretable or the operating system cannot act on it.
For RAMMS, the research target is therefore not digitalisation for its own sake. It is a traceable chain from the physical mining problem, through measurement and modelling, to a safer or more productive decision.