Mining automation and field survey analytics

Autonomy and teleremote retrofits

Most sites run mixed fleets: OEM autonomous haulage beside older machines that will run for another decade. A retrofit adds drive-by-wire interfaces on the machine's CAN bus, perception from lidar and radar, positioning from RTK GNSS on the surface or lidar and UWB localisation underground, and a teleremote station with low-latency video for tasks that are not yet automated. Safety-related functions, such as emergency stop and the boundary of the autonomous operating zone, are engineered as hardwired or safety-rated paths that do not depend on the autonomy computer.

ISO 17757:2019 sets safety requirements for autonomous and semi-autonomous machines in earth-moving and mining, covering the machine, its control system and the supporting infrastructure (standard listing). Control functions are analysed against functional safety practice from IEC 61508 and ISO 13849, with hazard analysis, failure modes analysis and a safety case owned by the site. Remote control on its own falls under ISO 15817 rather than ISO 17757.

Drill and loader automation

Drill automation covers auto-levelling, hole navigation against the drill plan and automated drilling cycles. Its most valuable by-product is measure-while-drilling data: penetration rate, torque, pull-down pressure and rotation speed, normalised into rock hardness indices that feed blast design and domain models. For underground loaders, automated tramming and dumping run inside isolated zones, with an operator taking over by teleremote for bucket filling, and payload estimates logged per cycle.

Collision avoidance and vehicle interaction

EMESRT frames vehicle interaction controls as nine levels: levels 1 to 7 are site and operator controls, and levels 8 and 9 are technology controls, advisory alerts at level 8 and automatic intervention to slow or stop the machine at level 9. Proximity detection draws on GNSS vehicle-to-vehicle messages, UWB and RFID tags for personnel and light vehicles, and radar. We build the data side: interaction events and near-miss heat maps by location and shift, nuisance alarm rates (which erode operator trust fastest), and scenario-based test results. We integrate and analyse collision avoidance systems; we do not certify them.

Fleet data and fixed-plant predictive maintenance

Fleet management data on cycle times, queuing and payloads feeds dispatch optimisation. On fixed plant, failures announce themselves early if you listen:

• Conveyors. Acoustic sensing along the belt finds failing idler bearings, thermal imaging catches hot idlers and pulleys, and belt-rip and misalignment detection protects the most expensive component on the system.
• Crushers and mills. Vibration spectra are tracked at bearing defect frequencies, motor current signature analysis exposes electrical and mechanical faults, and liner wear is modelled against throughput.
• Scheduling. Each asset is compared with its own baseline, and remaining-life estimates are expressed in operating hours so repairs land in planned shutdowns.

Underground ventilation and atmosphere

Ventilation on demand drives fans and regulators through variable-speed drives from where people and diesel equipment actually are, using tag tracking and gas readings, instead of running the whole mine at worst-case airflow. Gas and atmosphere monitoring covers methane, carbon monoxide, carbon dioxide, oxygen, nitrogen dioxide, hydrogen sulphide and diesel particulate, with sensor calibration and bump-test records stored against each instrument so readings stand up to an inspector.

Comms above and below ground

Open pits increasingly run private LTE or 5G. In remote Australia, the ACMA issued area-wide licences in the 3.4-4.0 GHz band in 2023, suited to private networks such as those at mine sites, and apparatus licences remain an option. Underground, leaky feeder radiating cable carries VHF or UHF voice, fibre backbones carry Wi-Fi or LTE, and mesh nodes follow development headings that move every week. Remote bores, pumps and tailings instruments report over LoRa, as described on our long-range comms page, with satellite backhaul for exploration camps.

Edge inference in dust, heat and vibration

Cameras and lidar on mobile plant face dust, spray, glare and constant shock. We run quantised models, as described on our sovereign AI page, on fanless, conduction-cooled compute rated for extended temperature and vibration, use multi-echo lidar returns to see through dust, add air knives or lens heating where optics foul, and benchmark latency at enclosure temperature.

Survey: photogrammetry, LiDAR and volumes

Drone photogrammetry plans ground sample distance and image overlap from the accuracy the job needs, adds oblique passes for highwalls, and uses LiDAR where vegetation or texture-poor surfaces defeat image matching. Accuracy comes from survey control: ground control points and independent checkpoints surveyed with RTK or PPK GNSS, with the root-mean-square error on checkpoints reported with every deliverable. Stockpile volumes are computed by differencing surfaces against an agreed base, and we report how much of the tonnage uncertainty comes from bulk density rather than geometry. Repeat surveys and truck IMU data measure haul road rutting and potholes, and pit progress is compared with design. Drone flights are regulated by CASA; see our drones and robotics page.

Slope, wall and tailings monitoring

Geotechnical monitoring fuses sources that each see part of the picture: prisms read by robotic total stations, ground-based slope stability radar, satellite InSAR (Sentinel-1 radar data is free through the Copernicus Data Space Ecosystem), extensometers and piezometers. We put them on one time base and one coordinate frame, compute velocity and inverse-velocity trends, and route alerts through the site's trigger action response plans.

Tailings storage facilities need continuous evidence: vibrating-wire piezometers for pore pressure, inclinometers, seepage flow, pond level and beach width, logged over LoRa where cabling is impractical. The Global Industry Standard on Tailings Management, launched in August 2020 by ICMM, UNEP and PRI, and the ANCOLD guidelines on tailings dams shape what is monitored and how it is reviewed. We supply the monitoring system and the data; interpretation and decisions sit with the Engineer of Record.

Mineral mapping, geophysics and prospectivity

Hyperspectral and multispectral imagery maps alteration minerals through their short-wave infrared absorption features, using spectral unmixing to separate mixed pixels. Geophysics processing covers magnetics (reduction to the pole, derivatives and analytic signal), gravity (Bouguer and terrain corrections) and electromagnetics (conductivity-depth imaging and inversion). Machine learning prospectivity models combine these layers with geology and geochemistry, and we validate them with spatially blocked cross-validation, because random splits leak information between neighbouring cells and flatter the model. Each map ships with an uncertainty map.

Core logging, assay QA/QC and reporting

Computer vision on core tray photographs registers depth, detects fractures for RQD estimation, classifies lithology and flags intervals for a geologist's review, alongside hyperspectral core scans where available. Assay QA/QC analytics track certified reference materials on control charts, blanks for contamination, and field, coarse and pulp duplicates, and flag batches for re-assay. Under the JORC Code (2012), Public Reports of Exploration Results, Mineral Resources and Ore Reserves must be based on, and fairly reflect, information prepared by a Competent Person with at least five years of relevant experience. We prepare and check the data; the Competent Person signs off.

Digital twins, rehabilitation and closure

A site digital twin combines survey surfaces, the block model and live telemetry, so planners can test a change before committing to it. For rehabilitation and closure, repeat imagery tracks vegetation cover against analogue reference sites, repeat surface models measure erosion, and water quality loggers record runoff and seepage over the years that closure criteria require.

Questions