mineGPT: Seamless Data Integration and AI Services

Authors:
Andrew Kirk (Senior Consultant | Australia)
Varun Mahajan (Business Partner | Marketing)
For those mining folk wondering what an AI system might look like in their organization, here’s an attempt to explain it.
Here I have created mineGPT, its an industrial-scale assistance tool for a mine in the middle of the Pilbara? Like parking assist for your Ferrari, but on steroids!

The Building Blocks

With any AI service platform, at a 10,000 foot view, this is what you’ll find:

Data Integration & Enrichment

Data integration and enrichment is the “plumbing” to [1] ingest data from source devices and systems, then [2] catalog, classify, label, sanitize and protect in readiness for the AI core. Key issues here are data quality and real-time ingestion and enrichment needed for most AI use cases. Invariably, AI is most valuable when it’s reacting to live events at scale. Like your car’s parking assist, ‘live’ real-time data is the only way.

A mine’s data sources are numerous, but often they’re:

  • Sensor data from pumps, dump trucks, rope shovels, etc.,
  • Camera video from Camp, the Mess, the Pit, rail, port ,
  • ERP system data, and
  • HR records and policies.

The AI Core

mineGPT’s AI core is a range of data models, some pre-built, some off-the-shelf and some tailored, configured into a system. Take for example truck maintenance across several mine sites for several hundred trucks.

mineGPT’s core could consist of:

  • Model 1 consuming internet data to capture truck model data from the Komatsu product line. It uses what’s called a Large Language Model based on a transformer neural engine developed by Google to create maintenance recommendations for a dump truck.
  • Model 2 scans the mine data for each truck in motion and embeds the video into vector models (see embeddings). These will identify truck specifications, particular modifications, predict performance expectations, and generate maintenance instructions, parts orders and staff rosters for the truck fleet over the next weeks and months,
  • Model 4 uses 1 & 2 to provide the maintenance shop’s schedule over the next month with a detailed view of the week ahead, and deliver a view of a truck’s status for the maintenance supervisor. Helping to manage staffing load, provide risk level of delays, and why (e.g. parts delays),
  • Model 3 is a ‘discriminator’ model to correct bias, and check ethics and confidentiality policies are adhered to.

AI Services

Then finally there’s the human interface for consuming mineGPT’s service in the form required. Is it simple text to terminals or phones? Is it audio or a video, or is it live guidance? This is driven by the Use Case.

For truck maintenance it could be:

  • Audio to driver terminals saying “hello Jack, please slow down, we don’t want excessive breaking at the bottom of the hill”, or
  • Mixed audio, text and video. The shop Engineer requests the maintenance steps for the truck coming tomorrow. mineGPT discusses replacement parts, variations with them and confirms. Then shows the maintenance steps with both text & video instruction, ensuring quality and safety.

The mineGPT Platform

While we’ve only scratched the surface here I hope you have a better appreciation of what a mineGPT platform could look like for you.

AI is a part of our lives today. Your car’s parking assist, Google Maps directions or the grammar suggestions in Word are used by millions every day. Between the hype merchants and the cynics is a real opportunity to move Australian Mining forward.

If you’ve read this far, thank you! If it helped in some way, I’m glad. Please reach out if you’d like to explore more.

Scroll to Top