You’ve probably heard about machine learning and artificial intelligence, but you may not know that some mines are already using machine learning to predict downtime and to automate tasks traditionally undertaken by engineers and scientists.
Machine-learning algorithms are considered the next step for digital mine transformation. The good news is that machine-learning algorithms are often platform agnostic and can be deployed using edge and/or cloud computing. Some mines have already moved beyond descriptive analytics and visualisation platforms and into machine-learning prediction and artificial intelligence.
PETRA Data Science’s Forestall algorithms use real-time PI data to automatically warn operators and maintenance crews of downtime hours in advance. Forestall algorithms can predict Geho pump pressure spikes 24 hours before they happen. The probability of a pressure spike is displayed in real time using the company’s data-visualisation platform. Algorithms can also be coded into the plant’s digital control system (DCS) or mobile fleet to enable end-node processing.
Forestall algorithms are also being used to predict SAG (semi-autogenous grinding) mill overloads and to predict processing plant performance using geological data. PETRA Data Science’s Prodfinder data integration breaks down data silos and enables these types of machine-learning predictions to be developed.
MOVUs’ FitMachine uses machine-learning algorithms to monitor machine health. FitMachine is like a Fitbit for industrial equipment. This custom-sensor cloud and analytics solution is designed to provide visibility, utilisation and condition health of industrial equipment.
Machine-learning artificial intelligence has a wide range of potential applications in the mining industry. An example of this is its use in the assessment of ore fragmentation in underground and open-pit operations. FRAGx algorithms, developed by PETRA, for example, utilise 3-D mapping point cloud data (e.g. uGPS Rapid Mapper, 3-D laser mapping or MVS) to automatically assess ore fragmentation in less than a minute. The algorithms have been trained to automatically remove concrete floors and shotcrete from the assessment, and are unaffected by dark, wet, dusty and humid underground mine conditions. Until now, ore-fragmentation assessment required an hour of manual processing by geotechnical engineers, off-site processing and/or very high lighting.
In the future, these algorithms can also be trained to carry out geotechnical inspections using 3-D mapping data (digital mine surveys). For example, the algorithms could be trained to recognise:
- Cracked shotcrete; and
- Plate deformation, missing plates and mesh bagging.
While these algorithms won’t be as accurate as professionally trained geotechnical engineers, they have the potential to cover large areas quickly as part of digital mine surveys (using 3-D mapping). Together, 3-D mapping and machine-learning artificial intelligence offer an array of exciting opportunities for geotechnical engineering, not just for underground, but also for open-pit wall inspections and assessment.
Ozius Spatial’s Naxia is a machine-learning artificial intelligence solution for environmental monitoring. Naxia consumes existing field data and uses remote-sensing algorithms, supported by machine learning to enable large tracts of land to be analysed for environmental risk, consistently and efficiently without needing to be on-site.
Edge computing may also help miners deploy machine-learning algorithms more quickly and without the need for high-speed internet. When people think about machine learning they sometimes assume it needs cloud processing, but edge computing and the internet of things (IoT) enable algorithms to be processed where the things are. Accordingly, edge computing enables machine-learning processing on the mine site. In contrast, cloud processing requires data to be sent off-site, and therefore requires fast and reliable internet connectivity – a common problem for remote mine sites.
Machine-learning algorithms can be coded into fixed plant control systems (e.g. DCS and PLC) and mobile fleet edge processing solutions, such as iVolves’ Maintenance Manager, which is capable of running machine-learning algorithms on the machine or using cloud processing.
Digital mine transformation
With the huge volumes of IoT data generated by mine sites these early machine-learning case studies are only the start of the value to be derived from machine-learning prediction and artificial intelligence.