Вештачка интелигенција

Trening geološkog AI modela: A Fresh Perspective on Mineral Exploration

Summary

In the transition towards the Net-zero economy, one of the harsh realities is the need to discover and mine a vast amount of resources for new infrastructure. Engines will be replaced by electric motors, generators by solar and wind power […]

Trening geološkog AI modela: A Fresh Perspective on Mineral Exploration

In the transition towards the Net-zero economy, one of the harsh realities is the need to discover and mine a vast amount of resources for new infrastructure. Engines will be replaced by electric motors, generators by solar and wind power systems, pipes by wires and transformers, and gas reservoirs by batteries. Building all of this will require uncovering millions of tons of new minerals, many of which are not conveniently located due to geopolitical issues or proximity to communities.

Recently, Earth AI announced preliminary successes in uncovering this buried treasure in previously unexplored areas. The mining industry refers to these areas as “greenfield sites,” in contrast to the exploration of “brownfield sites” near existing mines. The company recently discovered a rich molybdenum deposit in Australia with double the concentration compared to the largest existing mines.

Perhaps even more impressive, their new approach achieved a success rate of 1 in 8 in uncovering mineral deposits on greenfield sites, compared to the industry average success rate of 1 in 200. Their results even slightly outperformed traditional techniques for exploring brownfield sites, where the average success rate is 1 in 20.

Training an AI geologist

Exploring mining opportunities on greenfield sites poses a much more challenging problem. Brownfield sites have been extensively studied and well-investigated, while greenfield sites lack sufficient data. The standard and desirable solution is to collect new data, which is costly and time-consuming.

The team at Earth AI hypothesized that they could replace the need for additional data with a better approach to interpreting and annotating existing data. They trained their system on 400 million geological cases known from research archives. This was no easy task. They taught their AI to learn geology. Our AI acts as a geoscientist, studying each case, extracting knowledge, and developing predictions. But it happens on a much larger scale, generating consistent and reliable predictions.

The exploration process

The exploration process consists of three phases:

1. Targeting – Models are trained on 400 million geological cases across Australia to identify areas of mineralization and highlight locations with a high probability of finding mineral systems. Teams then go to the field to sample and inspect the targets.

2. Hypothesis – Geologists study the mineral system in the field, aided by technology that enhances their understanding of the geological environment and helps formulate hypotheses.

3. Drilling – The team tests their hypothesis by drilling up to a depth of 600 meters, confirming or disproving the presence of mineralization. Each drill hole provides invaluable knowledge about the mineral system, which is then fed back into the system to form new hypotheses.

Data quality and systematic drilling approach

Earth AI has assembled a mix of experts in geology and deep learning. More importantly, they have adopted a research approach to improve the quality of geological deep learning predictions. The company has spent over six years conducting research and over six hundred field days testing predictions and gathering feedback for system improvement.

The quality of data from each geological case is crucial, but monitoring data quality is a significant challenge when working at a continental scale using hundreds of millions of data points. Therefore, Earth AI has developed a semi-automated data review system that significantly speeds up the data quality review process. For example, a region-specific software tool focuses on finding and memorizing data errors and inconsistencies to rectify them on a large scale.

FAQ:

Q: What is Earth AI?
A: Earth AI is a company specializing in using AI technology to discover mineral deposits in new and unexplored areas.

Q: What is the success rate of Earth AI’s approach compared to the industry average?
A: Earth AI’s approach has achieved a success rate of 1 in 8 in uncovering mineral deposits on greenfield sites, surpassing the industry average of 1 in 200.

Q: How does Earth AI train its AI geologist?
A: Earth AI trained its AI geologist by teaching it to learn geology through studying and extracting knowledge from 400 million geological cases in research archives.

Q: What are the phases of the exploration process?
A: The exploration process includes targeting, hypothesis formulation, and drilling.

Q: How does Earth AI ensure data quality in their predictions?
A: Earth AI has developed a semi-automated data review system that focuses on identifying and rectifying errors and inconsistencies in data to improve data quality.