Machine learning for resource modelling

Maptek DomainMCF uses machine learning to generate domain boundaries direct from drillhole sample data for rapid creation of resource models. Geologists feed in drilling data and obtain domain or grade models in dramatically less time than traditional resource modelling methods.

Breakthroughs often arise through thinking about a problem in a different way. This type of thinking - there must be a better way - led to Maptek founder Bob Johnson taking borehole plotting from a laborious manual process to a simple digital solution. Following the needs of geologists led to the computerised geological modelling that became Vulcan.

Now Maptek is pursuing a novel approach to modelling domain boundaries that is destined to again overturn established techniques. Applying machine learning to geological modelling challenges the assumption that generating a resource model is a time-consuming, onerous project.

DomainMCF sees resource models generated in minutes or hours, depending on size, data density and complexity of the deposit. Operations will benefit from multiple solutions run in parallel using high performance cloud computing.

Introducing DomainMCF - Machine learning for orebody modelling

 

Putting the geology back into geologists is a simple way to describe the outcome of machine learning development, according to Technical Lead for the project, Steve Sullivan.

‘Rest assured, decisions will still be needed from geologists but now those decisions will focus on the analysis and refinement of the resource model.

‘Replacing the onerous task of pre-defining parameters and constraints with an algorithm which learns from the best of the best heralds great news for the industry.

‘Transforming the heavy computing load needed to process massive amounts of geological data has proven worthy of the challenge. It has also thrown up some interesting side benefits - bringing with it a real advance in data validation and streamlined identification of structural faults.