CASE STUDY
Datarock: Extracting new value in mining using predictive and generative AI
DiUS helped Datarock build a cloud-based, AI-powered platform that extracts geological and geotechnical data from core imagery for transformational insights.
At a glance
When DiUS partnered with Solve Geosolutions on a project, the two companies successfully reinvented the process for geologists to analyse drill core samples to generate geological information. However, they also discovered an opportunity to build a new software product, which leverages both generative AI and traditional machine learning tecniques, to improve the mineral discovery process.
Our services:
Meet Datarock (Solve Geosolutions)
Datarock started life as Solve Geosolutions – a small Melbourne-based data analytics consultancy that specialised in the application of powerful machine learning workflows to a variety of geoscientific problems in the exploration and mining industry.
Solve’s aim was to develop and implement data science-based solutions that improved the way companies mine and explore large and complex multivariate datasets. In addition to designing end-to-end solutions, Solve wanted to empower its clients to think about their data in new and innovative ways through collaborative engagement and training.
The Challenge
A more responsive and accurate way to plan and manage operations
In today’s mining industry, fast access to reliable data about the geology of a deposit is critical when making operational decisions about field work at a mine. Even small improvements to the geological process can have a significant effect on efficiency and productivity—and ultimately profitability.
A key part of the exploration process involves drilling core samples at different locations and getting a geologist to manually inspect the samples and report on the geological features such as veins and textures. This task can be tedious and error prone and conclusions can be highly subjective. Additionally, the lag between drilling a sample and receiving the results impacts the numerous interrelated processes and decisions on how economically important metals can be efficiently extracted from a mineral deposit.
The team of mathematical geologists at Solve Geosolutions identified an opportunity to automate the analysis of drill core samples using an unexploited datasource: the store of high resolution digital photos that are taken of the core samples as a record of the job.
What we did
Geology from imagery reinvents discovery process
In order to create better outcomes from core image analysis, Solve needed assistance from a partner that had specialist expertise with deep neural networks to help with the complex image processing — so they reached out to DiUS.
Starting small with a project to detect veins in half-core, DiUS fast tracked the project by leveraging an open source deep neural network architecture suitable for object detection and segmentation problems. In addition, the team was able to further speed time to results by working closely with the Solve geoscientists, using a small dataset and heavily relying on augmenting images to train the model.
The results were impressive. The machine-learning powered solution detected veins faster, more consistently and often at a higher or equal quality to a geologist. More importantly, it provided the ability to more completely investigate the geology of a mineral deposit, as until now it has not been feasible for a geologist to dedicate the time to analyse the data at the same resolution.
In order to further improve the accuracy of models, DiUS explored ways in which generative AI could take drill core images and smoothen the texture in such a way that other important geological information could still be preserved. To do this, we applied an edge-preserving image smoothing GAN (Generative Adversarial Network) in a daisy chain solution with the existing predictive AI model. After some fine tuning of the model, our generative AI solution worked extremely well and provided accurate results. Another success was the fact that only a small dataset of around ten images was required to rollout the solution.
Datarock – Cloud native, computer vision solution
Solve Geosolutions and DiUS partner to transform the mining industry
Results for Datarock
Opening new and revisiting old areas of mineral exploration
Recognising that miners have similar needs and similar standards of image repositories, Solve and DiUS formed a joint venture to build a platform, Datarock – a new and innovative cloud-based platform to automatically analyse drill core images using image segmentation technology that allows exploration and mining companies to get consistent insights from image sources.
It’s estimated that Australian mining companies alone commonly hold up to ten years of suitable photos, between 150,000 and 300,000 per deposit, and that machine-learning powered analysis using DataRock is a highly cost-effective investment to uncover value in old deposits.
“DiUS brought product development and specialist computer vision machine learning capabilities together with Solve Geosolutions’ data science expertise in exploration and mining to deliver a richness of geological insight from core imagery that’s never been economically viable before.
In the face of increasing industry challenges such as low commodity prices and declining grades, DataRock provides accurate, fast and consistent high resolution information about a mineral deposit’s geology that aids decision making throughout the entire mining cycle, delivering important productivity and throughput savings to a mine’s bottom line.”
– Liam Webb, CEO, Datarock
To operationalise the image analysis of geological drill core samples as a service, DataRock leverages the power of the AWS Platform and latest AWS Machine Learning Services, including Amazon Sagemaker, in a completely serverless approach. It is secured using AWS IAM along with Amazon Cognito for identity management and protected by Amazon API Gateways.
DataRock provides access to a range of geological use cases and the SaaS product has been designed to drive maximum utility through the easy addition of additional geological use cases.