AI for Rare Earth Element Processing

REE Assay Interpretation

Rare earth elements are not rare — they are difficult to separate. The 15 lanthanides plus yttrium and scandium occur together in nature, and their similar ionic radii and chemical properties make separation one of the most challenging hydrometallurgical problems in the industry. AI is transforming how we interpret REE assays and optimize separation processes.

Open data/ree-assay-data.csv in the code panel. Each row contains a sample with individual REE concentrations (La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Y, Sc), total rare earth oxide (TREO), thorium, uranium, and host mineral identification.

TREO Distribution Analysis

Not all TREO is created equal. A bastnaesite ore with 60% TREO might be 45% cerium oxide and 20% lanthanum oxide — the two least valuable light REEs. The economic value depends on the distribution across individual elements, particularly the heavy REEs (Gd through Lu plus Y) which command 10-100× higher prices:

AI-based economic modelling takes the full REE distribution, applies current market prices, estimates separation costs for each element pair, and calculates the net economic value per tonne of ore. This changes the cutoff grade calculation fundamentally — a 1% TREO ore enriched in Dy and Tb can be more valuable than a 5% TREO ore dominated by Ce and La. The bastnaesite at Mountain Pass (MP Materials, California) is LREE-dominant, while the monazite-xenotime ore at Mt Weld (Lynas, Western Australia) and emerging HREE projects carry more of the high-value heavies.

Thorium and Uranium Contamination

Monazite-bearing REE ores contain 5-10% ThO2 and 0.2-0.5% U3O8. This is simultaneously a problem (radioactive waste management under NRC / state agreement-state rules in the US, ARPANSA in Australia, and Euratom/national regulators in the EU) and, in some jurisdictions, a strategic interest (thorium as a potential nuclear fuel). AI-based assay interpretation must flag:

Solvent Extraction Optimization

Solvent extraction (SX) is the industrial workhorse for REE separation. A typical REE SX circuit has 50-200 mixer-settler stages arranged in multiple cascades, each performing thousands of organic-aqueous contacts per day. The operating variables — acid concentration, organic-to-aqueous ratio (O/A), pH, temperature, extractant concentration, scrub ratios — create a vast optimization space.

Open data/separation-cascade.json — it contains cascade configurations, separation factors between adjacent REE pairs, and stage-by-stage composition profiles.

Separation Factors

The separation factor (β) between two adjacent REEs determines the number of stages required:

β(Nd/Pr) = D(Nd) / D(Pr)

where D = distribution coefficient = [REE in organic] / [REE in aqueous]

Typical separation factors for adjacent lanthanide pairs range from 1.5 to 3.0 with conventional extractants (D2EHPA, PC88A, Cyanex 272). Higher separation factors mean fewer stages, lower capital cost, and lower reagent consumption. AI optimization targets:

AI-Driven Cascade Design

Traditional SX cascade design uses McCabe-Thiele diagrams for binary separations and extends to multi-component systems through sequential binary splits. This works but makes simplifying assumptions — ideal stage efficiency, constant separation factors, no interaction effects between REEs.

An AI approach using neural network-based process simulation:

At Western separation plants ramping up outside China (Lynas in Malaysia/Texas, MP Materials at Mountain Pass), AI-optimized SX parameters can reduce the number of stages required for Nd/Pr separation by around 15% while maintaining 99.5% Nd oxide purity — a significant capital cost reduction for new cascade installations and a key lever for cost-competitiveness against incumbent Chinese capacity.

pH and Temperature Control

Real-time control of pH in SX circuits is critical. A 0.1 pH unit drift can shift separation factors enough to contaminate the product stream. AI-based control:

Control loop (every 60 seconds):
  1. Measure: pH at each extraction stage, temperature, flow rates, online REE
     composition (if XRF available)
  2. Predict: downstream composition profile over next 2 hours
  3. Optimize: acid/base addition rates to maintain target pH profile
  4. Constraint: total acid consumption ≤ budget, waste acid generation ≤ treatment capacity

A practical challenge in newer REE plants is measurement — many facilities still rely on manual sampling every 4-8 hours rather than online analysers. AI-based soft sensors that estimate composition from easily measured variables (pH, conductivity, density, temperature) bridge this instrumentation gap.

Critical Minerals Strategy

Open data/critical-minerals-demand.json — it contains REE demand projections by application, current Western production, import dependency, and strategic reserve estimates.

The Western REE Position

The US, EU, Australia, and Canada collectively hold substantial REE resources but historically produced a small share of separated oxides — China has dominated midstream and downstream processing. The disconnect between resources and finished-product capacity is due to:

Critical Minerals Policy (US, EU, Australia)

Rare earths appear on the USGS list of critical minerals, the EU Critical Raw Materials Act (CRMA), and Australia's Critical Minerals Strategy. Key provisions across these frameworks:

AI for Supply Chain Risk Assessment

AI models that integrate geological resource data, production capacity, trade flow data, geopolitical risk indices, and demand projections can assess supply chain vulnerability for each critical mineral:

Supply Risk Score = f(import_dependency, supplier_concentration,
                      geopolitical_risk, substitutability,
                      domestic_resource_adequacy, recycling_rate)

For neodymium: Western import dependency historically >80%, supplier concentration (China) >85% of separated supply, low substitutability for permanent magnets → high supply risk. This quantitative risk assessment informs which elements justify domestic processing capacity expansion and strategic stockpiling, and underpins the USGS, EU CRMA, and IEA critical-minerals assessments.

Exploration Priorities

Beyond traditional placer monazite and the major bastnaesite/monazite mines, prospective REE targets across Western jurisdictions include:

AI-based prospectivity mapping — integrating geological, geophysical, geochemical, and remote sensing data — can prioritize exploration targets. The same multi-source data fusion approach from Chapter 1, retrained on REE deposit characteristics, identifies areas where geological conditions favour REE mineralization.

Key Takeaways