Rio Tinto Centre for Mine Automation
The Rio Tinto Centre for Mine Automation (RTCMA) is a collaborative research project in partnership with Rio Tinto spanning over a decade.
Our research brings together multiple highly technical academic disciplines of perception algorithms, sensing technologies, machine learning and data fusion, operations research, stochastic optimisation and control theory.
This technology transfer into Rio Tinto is supported by a team of software engineers working closely with academics within RTCMA.
Our work has resulted in a number of major research advancements, in both fundamental and applied areas as well as technologies that have transitioned into operating mines.
Our research
Planning and optimisation
Fleet planning and scheduling is highly combinatorial, and often non-deterministic polynomial-time (NP)-hard. We're working on automation of these processes to improve equipment utilisation, production output, operating costs and carbon emissions.
Our experts: Dr Andrew Hill, Dr Mohammadreza Chamanbaz, Dr Konstantin Seiler
Our collaborator: Rio Tinto
This project has developed a new architecture for dispatching equipment in mine sites. This fleet dispatch algorithm combines Monte Carlo tree search (as seen in Alphabet Inc’s Google Deepmind AlphaGo artificial intelligence system), with dynamic time-based material flow planning, to generate real-time haul-truck assignments that optimise material movements subject to complex constraints and objectives. Our research focus is on multi-agent stochastic optimisation and high-fidelity modelling of agents and processes.
Our experts: Dr Andrew Hill, Dr William Jones
Our collaborator: Rio Tinto
Rio Tinto have the world’s first fully-autonomous heavy haul, long distance rail network. We’re developing new scheduling and planning algorithms to direct train movements, taking advantage of the autonomous trains’ performance and reliability. The research focus is on stochastic modelling, hierarchical train control and schedule optimisation algorithms.
Orebody modelling
Modelling geological formations underground is challenging due to the lack of direct visibility and cost of exploratory sampling. We're working on novel estimation algorithms to improve geological estimations, fusing together a variety of dense and sparse data from different sources.
Our experts: Dr Anna Chlingaryan, Dr Arman Melkumyan
Our collaborator: Rio Tinto
Multi-million-dollar decisions in mine planning and operations heavily rely on the amount and quality of the material that is expected to be mined. This project addresses the need to produce comprehensive and nuanced probabilistic volumetric estimates of the in-ground material distribution in 3D and continuously update it as new production data becomes available. We utilise state-of-the-art machine learning techniques to model and fuse data from multiple sources of information.
Our experts: Dr Anna Chlingaryan, Dr Arman Melkumyan, Dr Katherine Silversides
Our collaborator: Rio Tinto
Accurate representations of the geological boundaries within a deposit are required to produce high quality geological mine models. This project aims to develop a probabilistic methodology for modelling of 3D geological boundaries using exploration and production data. These will enable detailed geological surfaces to be produced that can be continuously updated as additional information becomes available. Machine learning via Gaussian Processes is used to capture the uncertainty in the boundary models.
Our experts: Dr Mehala Balamurali, Dr Anna Chlingaryan, Dr Arman Melkumyan
Our collaborator: Rio Tinto
Boundary surfaces for mine modelling are often produced using exploration data which is accurate but sparse. This project aims at deforming existing boundary surfaces to make them better aligned with denser production data that becomes available during mine operations.
Our experts: Dr Arman Melkumyan, Dr Katherine Silversides
Our collaborator: Rio Tinto
Measure while drilling (MWD) data collected in production holes can provide detailed information on the location of stratigraphic units in banded iron formation hosted iron ore deposits. Stratigraphic modelling in these deposits is typically based on sparse exploration holes, and the densely spaced production data has the potential to increase the model detail at the bench scale. Our research has focused on cleaning the data to reduce noise and employing Gaussian Processes to distinguish different geological units using the MWD.
Our experts: Dr Mehala Balamurali, Dr Arman Melkumyan
Our collaborator: Rio Tinto
Limiting the impact of outliers and appropriate treatment of subdomain regions (located within the existing domains) has the potential to significantly increase the robustness of the geological models and improve their reconciliation with the production outcomes. This project proposed machine learning based autonomous techniques for identifying latently present geological sub-regions and outliers.
Our experts: Dr Anna Chlingaryan, Dr Arman Melkumyan, Dr Richard Murphy,
Our collaborator: Rio Tinto
Classification of rock types and minerals of mine faces using hyperspectral imagery acquired from field-based platforms can provide continuous maps at unprecedented resolution and help in modelling. Analytical methods were developed and validated to map ore and waste materials and to quantify clay minerals on the mine face as potential lines of stratigraphical weakness. The hardware system was developed to enable operation of hyperspectral sensors in the challenging conditions of the mine pit.
Our experts: Dr Mehala Balamurali, Dr Andrew Hill, Dr Arman Melkumyan, Dr Konstantin Seiler
Our collaborator: Rio Tinto
Real-time control and planning rely on precise tracking of active states and available resources. This project aims to develop high-fidelity models for tracking equipment and materials once extracted from the ground, across all operations. The research focuses on advanced modelling, machine learning and data fusion techniques to handle gaps in available data.
Autonomous vehicles
Autonomous vehicles can provide increased performance and reliability, as well as allowing operations to be performed where it would be otherwise too hazardous for human operators.
Our experts: Dr Andrew Hill, Dr Arman Melkumyan, Dr Richard Murphy
Our collaborator: Rio Tinto
Automation of light vehicles with sensing payloads enables remote operation of survey and other sensing tasks. Autonomous data collection improves data quality and safety, keeping personnel away from heavy equipment and active operations. Research includes sensing technologies and perception algorithms, trajectory planning and safety of multi-agent interactions.
Our expert: Dr Andrew Hill
Our collaborator: Rio Tinto
Automation in mining is driven by safety and reliability. Robots can be utilised to perform repetitive work consistently or operate in close proximity to hazards too risky for humans. Our goal in this project is to develop new field robotic technologies to automate existing mining operations and make new processes feasible where they were previously too laborious or dangerous to attempt.
Our expert: Dr Andrew Hill
Our collaborator: Rio Tinto
Our drill automation project was the first to achieve fully automated hole pattern drilling without human intervention. Novel control and planning systems have enabled a single operator to remotely operate up to four autonomous drills, with improved safety, precision and utilisation. Rio Tinto Iron Ore currently operate seven autonomous drill rigs in production.