Key Takeaways

  • Physical density testing (water immersion, wax coating, gas pycnometry) has no dedicated software of its own — it’s a bench measurement, not a computational one.
  • The AI opportunity is predicting density from cheaper, faster proxy measurements instead of running the physical test on every sample.
  • A January 2026 peer-reviewed study, A Multi-Proxy Framework for Predicting Ore Grindability, shows portable XRF, Leeb hardness, and hyperspectral imaging can stand in for slower physical rock-property tests with real predictive power.

TL;DR

You don’t AI-ify the density test itself — you train a regression model on hyperspectral, XRF, and hardness proxy data to predict density (and related grindability parameters) without running the physical test on every single sample.

How Do I Predict Density With AI?

Density testing itself — weigh dry, weigh submerged, compute specific gravity — is a bench procedure with no software layer to automate; it’s fast and cheap enough per-sample that nobody’s built a dedicated product around it. Where AI actually helps is reducing how many physical density (and related grindability) tests you need to run at all. A multi-proxy framework published in Minerals (January 2026) evaluated Leeb rebound hardness, Cerchar abrasivity index, portable XRF, and hyperspectral imaging as rapid, low-cost stand-ins for slow physical rock-property tests on ore samples from two Canadian open-pit mines, and found geochemical features plus HSI-based mineralogical attributes carried real predictive signal for grindability class.

The practical workflow: run your portable XRF and hyperspectral scans (many sites already do this for grade and clay estimation — see our post on handheld XRF and hyperspectral core scanning), collect a smaller calibration set of physical density/hardness tests, and train a regression or classification model (gradient boosting or random forest works fine at this data volume) to predict density and grindability class from the cheap proxy measurements. This is genuinely emerging — a published research result, not yet a packaged commercial tool — so expect to build the pipeline yourself or work with a metallurgical consultant rather than buying it off the shelf.

Try It With Geocluster

Building and validating a proxy-prediction model like this means pulling together hyperspectral scans, XRF assays, and physical test results from different systems into one training set — the kind of multi-source data harness Geocluster is designed to help you assemble.