As the tech industry helps exploration and mining firms extract data from more and more processes and sources, there is more need to automate analysis of that data.

The mineral exploration industry has excellent tools – such as portable XRF – for rapidly scanning core samples and converting their secrets into numerical data, but it can be very time-consuming to make sense of that data manually.

With this in mind, since 2015 CSIRO has been perfecting a software platform, now called Data Mosaic, that can process such datasets and render 3D geological models in minutes.

The heart of the solution is a mathematical procedure called Wavelet Tessellation, which identifies inflection points in a numerical signal. For drill core data these inflection points correspond to the boundaries between units with different rock properties. Machine Learning can be used to automate classification of the rock units based on multiple data types.

Furthermore, the technique takes into account both the wavelength and amplitude of features in the signal, which means it also quantifies the degree of change for each property, which could be the content of a specific element or set of elements, or a basic physical property.

«We’ve applied this technique to any number of different datasets. Any kind of downhole numerical data – it can be geochemistry, Gamma, XRF, neutron velocity, wireline logging, hyperspectral data or LIBS,» says Dr Jessica Stromberg, team leader of CSIRO’s Mineral Footprints Team, which continues to study how Data Mosaic can be improved and applied to new datasets.

The technique can even be applied to data from the drill bit, such as penetration rate, torque-on-bit, fluid flux etc. «That’s all data that is collected anyway, and is clearly related to rock properties, like hardness,» Stromberg adds. One of the more recent areas of research for Stromberg’s team is assessment of the best methods for inputting hyperspectral data into Data Mosaic, since such data can be applied to geological logging in multiple ways (raw spectral data, spectral indices and spectral mineralogy outputs).

How data can be visualized at varying degrees of resolution or domain scales

«We are continuing to investigate the most efficient and reliable use of HyLogger 3 spectral data and products in Data Mosaic for generating geological logs,» she notes.

The method is most effective for large numerical data sets, with dense, regularly sampled data, for which it will provide rapid, consistent results.

This means the software is ideal for mapping large numbers of drill holes, and the beauty of Data Mosaic in that context is that the same configuration of parameters – set according to the geologist’s knowledge of the terrain – can be repeated for every single sample.

The Data Mosaic website (https://research.csiro.au/data-mosaic/) links to detailed studies demonstrating the use of the technique: one modelling 50,000 datapoints of multi-element chemistry data from 259 drill holes at a layered mafic-ultramafic intrusion in Brazil (Fazenda Mirabela, Bahia); and another using 90,000 assay datapoints from a disseminated Ni-Cu-(PGE) sulfide orebody in Finland (The Kevitsa deposit).

But what about exploration firms that drill only a few holes each month?

In that scenario Stromberg believes Data Mosaic can still show its strength. «You may need to look at all the drill holes from a single area together, that’s where the fast processing comes in. But even on the scale of a few holes, it can be very valuable in terms of cost-saving, time-saving for geologists, interpreting the data from the drill holes and allowing faster decision-making, such as whether to continue or stop drilling, or where to drill the next hole,» she says.

While the above datasets are indeed huge, those case studies come from fairly homogenous environments. The much greater tectonic activity in the Andes likely leads to much more complex stratigraphy. Could this test the limits of Data Mosaic’s processing power?

Stromberg thinks not, and expects it to work just as well in such environments because CSIRO has tested it on drill holes and data from many different rock types and ore types.

«I would argue that in structurally complex terrains it’s even more important to nail down and correlate your geological boundaries and lithologies, and do that objectively because yes, in those cases there will be a lot of folding,» she says, adding «it also allows you to quickly test many different models to see what fits best, since it’s so quick in generating downhole logs.»

KNOCK-ON EFFECT IN OTHER AREAS OF RESEARCH

The icing on the cake is that Data Mosaic is free, available as a web app and built with accessible research code written in Python. However, a stand-alone, desktop version is available under license, and CSIRO is working one-on-one with major mining companies to help them implement it in their in-house workflows, as well as working with service companies specialize in generating this kind of data.

What’s more, CSIRO now uses Data Mosaic in-house as a standard part of its projects, it’s been adopted by the teams doing the Geological Survey of Western Australia and South Australia, and geophysical mapping firm Geotech recently used Data Mosaic to present information based on gamma logs.Stromberg herself has been able to use it in her work with the MinEx Cooperative Research Centre (MinEx CRC), where she is involved in the development of a real-time downhole assay tool based on Laser-Induced Breakdown Spectroscopy (LIBS).

While there are handheld tools on the market that allow you to take point measurements with a LIBS device, or map of small portions of core, there’s nothing for measuring long drill holes, she explains, adding that the MinEx CRC project focuses on LIBS because it’s sensitive to the entire periodic table.

The project is at an early stage of R&D, but the team’s most recent publication, led by University of South Australia PhD Student Fernando Fontana, has shown that Data Mosaic provides a fast and objective interpretation method for data from a LIBS tool, identifying boundaries and rock types based on multiple elements across intervals of 0.35mm.

Still pending in this project is work to overcome the challenges of creating a tool that can withstand the dirt, humidity and pressure in a downhole environment, as well as overcoming minerology and matrix effects that can interfere with the ability of LIBS to quantify rock types accurately and quickly.

Images from:
E. J. Hill, M. A. Pearce, J. M. Stromberg, 2021. Improving Automated Geological Logging of Drill Holes by Incorporating Multiscale Spatial Methods. Math. Geosci. 53:21–53
https://doi.org/10.1007/s11004-020-09859-0

Deja una respuesta

Tu dirección de correo electrónico no será publicada. Los campos obligatorios están marcados con *