IAP2-23-028

Machine Learning for Innovative Slope Design to Radically Decarbonise Open Pit Mining

The goal of achieving net zero by 2050 worldwide requires mining essential metals for electrification, e.g. copper, and building renewables, e.g. lithium, silver and nickel for EV batteries, iron for wind turbines, and many minerals for solar panels. In the UK, new open pit mines for the extraction of lithium (Cornish Lithium and British Lithium) in Cornwall and tin and tungsten in Devon (TungstenWest) are in the advanced planning stage. “To accelerate the growth of the UK’s mineral production domestic capabilities” and to “re-establish the UK as a skills leader in mining” are key objectives of the recently launched UK Government “critical mineral strategy” published this year by BEIS [9]. However, mining emissions are currently estimated at around four gigatonnes of CO2 equivalents per year, amounting to 8% of CO2 emissions globally [1]. Therefore, it is vital to establish disruptive technologies to substantially reduce mining’s carbon footprint.

Most mining happens via open pit mines. The overall steepness of open pit mine slopes (so-called pitwalls) plays a crucial role in determining a mine’s carbon footprint since carbon emissions are proportional to the amount of rock excavated. The overall steepness of pit slopes in open pit mines plays a key role in determining a mine’s carbon footprint since carbon emissions are proportional to the amount of rock excavated. Typical overall inclinations for mine pitwalls range from 30 to 55 degrees from the horizontal (the inclination being dictated by the strength of the rock(s) and the depth of excavation), so a huge amount of rock is excavated purely to ensure the stability of the pitwalls, i.e. to avoid pitwall failure. The adoption of optimal slope shapes for the mine pitwalls without compromising safety, i.e. the same level of Factor of Safety as for planar slopes is guaranteed, enables overall steeper pitwalls to be achieved so that the same amount of orebody can be extracted excavating much less rock and as a consequence less carbon footprint [1]. By systematically adopting topologically optimal pitwalls to design open pit mines globally (currently >5,000 worldwide), the mining industry could save 200M tonnes of CO2-equivalent carbon emissions annually and $7B in direct excavation costs globally. These numbers have been calculated from 4 mine case studies published in leading international mining Journal where the mine pitwalls were designed by OptimalSlope’s slope optimiser instead of traditional planar pitwalls, achieving reductions of 600,000 tonnes of CO2 emissions per mine on average (see refs: [2-5]).

OptimalSlope Ltd, the project industrial partner, is a start-up founded in 2021. Its novel slope optimiser has already demonstrated significant traction due to its potential to deliver radical decarbonisation to the hard-to-abate mining sector. The company has been selected to take part in two UK business incubators: the Greenhouse (now “Undaunted”) of the Centre for Climate Change and Innovation of the Grantham Institute at Imperial College and Better Futures of the Greater London Authority. Additionally, OptimalSlope has been admitted into the BHP (the largest mining company worldwide) powered incubator programme for innovation in mining TAD (Think & Act Differently), won prestigious industry awards such as the first prize in the category “innovation to be validated” at Peruminhub (September 2022) and is a London City winner in the Technation Rising star 5.0 programme. These prestigious industry awards and two grants from Innovate UK clearly indicate that leading mining industry experts see the game-changing potential of OptimalSlope and the need for its slope optimiser.

Methodology

Currently, the OptimalSlope slope optimiser requires an execution time ranging from 30 minutes for a single-layer slope to several hours/days for slopes with several layers (most common case). A radical runtime speed up (10- 50 times) is needed to achieve a slope optimiser product that may keep up with users’ runtime expectations to enable the adoption of OptimalSlope technology in the mining industry to achieve the aforementioned ambitious carbon reductions.

The ultimate impact sought by this project is to develop a responsible and trustworthy AI system in the form of a fast slope optimiser for the design of open pit mines that will contribute substantially to the decarbonisation of the mining industry.

The following are the project objectives to achieve such impact:

– Determine suitable ML-based techniques which can be used to deliver a radical runtime speed up (10- 50 times) to OptimalSlope current slope optimiser [6-8].
– Determine novel ML-based techniques to enhance the runtime for the second-generation slope optimiser with additional functionalities.

Project Timeline

Year 1

i) Acquiring and profiling an enriched dataset of optimal slopes and shapes (output set) linked with their lithologies (input set).
ii) To obtain this, the student must undergo a labelling process using the current version of the slope optimiser as a starting point for determining optimal slopes and shapes. This way, the student can extract meaningful features from these datasets that could link to optimal slope shapes.

Year 2

i) Choosing appropriate ML algorithms to determine the optimal slope of open pit mines, which can be included in existing slope optimisers such as OptimalSlope.
ii) Selecting both supervised learning and unsupervised, i.e. reinforcement learning, algorithms to establish correlations between the independent inputs (dataset) from Year 1, e.g. lithological sequences, geotechnical parameters such as statistics, survival probabilities, hydrological conditions and geological discontinuities and FoS, and the optimal slope profiles (output).

Year 3

i) The final step would be to improve even further slope optimisers with additional functionalities, such as simulating geological discontinuities.

Year 3.5

i) Thesis write-up;
ii) Dissemination at various conferences/workshops.

Training
& Skills

The School of Engineering requires each student to collect at least 60 PGRDP credits, corresponding to attendance in-school delivered workshops, taught modules and other activities that display further engagement. The training in (a) research skills and techniques and (b) research environment are provided through four mechanisms: (i) a programme of taught modules; (ii) internal training ‘workshops’ that focus on key geographical research skills and techniques; (iii) input from supervisors; and (iv) School and academic Group seminars by visiting and internal speakers and presentations by postgraduate students themselves.

In addition to generic training offered by the University, the School provides training through a series of in-house ‘workshops’. Engineering research postgraduates normally take the following Workshops: ‘Scientific Writing’, ‘Research Ethics (Theory)’, ‘Data Management’, ‘Time Management’, ‘Document Management – Content and Layout’, and ‘Introduction to Learning and Teaching’ during their first year. Also, it is envisaged that the student will undertake from 2 to 4 taught modules depending on the academic background of the appointed student of the MSc in ‘Geotechnical Engineering’, ‘Software Engineering’ and ‘Machine Learning’. Most of these modules are delivered in one or two intensive weeks, so well-suited for PhD students.

Research training continues through the second and third years, and is based around a number of themes: (i) Recognition and validation of problems; (ii) Demonstration of original, independent and critical thinking, and the ability to develop theoretical concepts; (iii) Knowledge of recent advances within the research field and in related areas; (iv) Understanding relevant research methodologies and techniques and their appropriate application within the research field; (v) Ability to analyse and critically evaluate findings and those of others; and (vi) Summarising, documenting, reporting and reflecting on progress.

Bespoke technical training will also be provided by the research supervisors (machine learning and advanced programming, numerical and analytical modelling of jointed rock masses, and geomechanical principles for the design of pitwall profiles) and technical staff in the School of Engineering.

References & further reading

1. Cox B., Innis S., Kunz N., Steen J. (2022) The mining industry as a net beneficiary of a global tax on carbon emissions. Communications Earth Environ., 3: 17.
2. Utili S., Agosti A., Morales N., Valderrama C., Pell R., Albornoz G. (2022) Optimal pitwall shapes to maximise financial return and decrease carbon footprint of open pit mines. Mining Metallurgy & Exploration, 39: 335–355, https://doi.org/10.1007/s42461-022-00546-8.
3. Agosti A., Utili S., Tasker J., Zhao C., Knights P., Nerhing M., Zia S. (2023) The effect of carbon tax and optimal profiles on profitability and emissions of open pit mines. Mining Technology, 1-16, https://doi.org/10.1080/25726668.2022.2122336.
4. Agosti A., Utili S., Gregory D., Lapworth A., Samardzic J., Prawasono A. (2021) Design of an open pit goldmine by optimal pitwall profiles. Canadian Institute of Mining Journal, 12(4): 149-168, https://10.1080/19236026.2021.1979382.
5. Agosti A., Utili S., Valderrama C., Albornoz G. (2021a) Optimal pitwall profiles to maximise the Overall Slope Angle of open pit mines: the McLaughlin mine. ACG Proceedings of Slope stability in mining conference 2021, 69- 82, Perth (Australia) https://doi.org/10.36487/ACG_repo/2135_01.
6. Rafiev A. et al., (2022) Visualisation of Machine Learning Dynamics in Tsetlin Machines. International Symposium on the Tsetlin Machine (ISTM), Grimstad, Norway, pp. 81-88.
7. Wheeldon A., Shafik R., Rahman T., Lei J., Yakovlev A., Granmo O.C. (2020). Learning automata based energy-efficient AI hardware design for IoT applications. Philosophical Transactions of the Royal Society A, https://doi.org/10.1098/rsta.2019.0593 .
8. Ghazal, O., Singh, S., Rahman, T., Yu, S., Zheng, Y., Balsamo, D., Patkar, S., Merchant, F., Xia, F., Yakovlev, A. and Shafik, R. (2023). IMBUE: In-Memory Boolean-to-CUrrent Inference ArchitecturE for Tsetlin Machines. ISLPED 2023, in press.
9. BEIS (2023) Resilience for the future: the United Kingdom critical mineral strategy.

Apply Now