CS-22
ASX Diversified Mining Group — AI Ore Processing Optimisation
An ASX-listed diversified mining group operating gold, copper, and nickel operations across 4 Western Australian sites — processing 28Mt of ore annually — was losing significant value to suboptimal flotation and comminution circuit performance. Manual process control, based on experienced metallurgists making decisions from lagging assay results (2–4 hour laboratory cycle time), resulted in ore recovery rates 6–12% below theoretical maximum. Anicalls' AI Ore Processing Optimisation Platform increased average ore recovery by 8.4% across all 4 operations, generating $112M in additional annual revenue without increasing throughput or operating costs.
+8.4%Ore Recovery Improvement
$112MAdditional Annual Revenue
23%Energy Cost Reduction
840%Year 1 ROI
Business Challenge
The Ore Recovery Value Loss Challenge
Recovery Rates 6–12% Below Theoretical Maximum
Metallurgical assessments across all 4 operations confirmed that actual ore recovery rates were 6–12% below the theoretical maximum recovery achievable from the ore body, given the mineralogy and processing circuit design. At 28Mt annual throughput and blended head grades, each 1% recovery improvement was worth approximately $13.3M in additional annual revenue. The gap between actual and theoretical recovery was driven by delayed response to ore variability — the processing circuit was optimised for average ore characteristics but real ore feed varied significantly hour-by-hour.
2–4 Hour Laboratory Cycle Creates Blind Flying
Process control decisions were made by metallurgists based on assay results from the laboratory — with a 2–4 hour cycle time from sample to result. During those 2–4 hours, the ore feed could have changed significantly in head grade, mineralogy, hardness, and liberation characteristics — but the processing circuit parameters (mill speed, cyclone pressure, flotation air rate, reagent dosing) remained locked at settings optimised for the previous ore batch. The result: perpetual sub-optimal performance as the circuit lagged the ore body by 2–4 hours.
Expert Knowledge Concentration Risk
The highest recovery periods at all 4 operations correlated directly with specific shift metallurgists who had 15–25 years of experience and deep intuitive understanding of their individual circuit's optimal operating envelope. When those metallurgists were on annual leave or resigned (two key individuals resigned during the year), recovery rates dropped 3–5% — and recovery variance increased significantly. This knowledge concentration risk was unquantified but materially significant: the departure of two individuals cost an estimated $8M in foregone recovery in the months before their institutional knowledge could be transferred.
Energy Overconsumption in Comminution
The comminution circuits (SAG mills, ball mills) — the single largest energy consumers at each operation, accounting for 60–70% of site electricity costs — were operated at fixed parameters that did not respond dynamically to ore hardness variability. When softer ore was processed, the mills were consuming energy at rates optimised for hard ore — wasting 15–25% of comminution energy input. Annual energy waste across all 4 operations: $18M. Eliminating this waste while maintaining throughput and recovery required parameter adjustment response times of minutes — impossible with human-manual control.
Solution Delivered
Anicalls AI Ore Processing Optimisation Platform — 4-Site WA Deployment
Real-Time AI
Real-Time Process Optimisation AI
AI process control models — trained on 8 years of historical process data (6.2TB per site) correlated with hourly assay results, ore characterisation data from the mine geology block model, and expert metallurgist shift notes — continuously optimise all controllable process parameters in real time. Models update recommendations every 60 seconds based on 340 live sensor inputs per circuit (feed rate, particle size, pulp density, pH, reagent flows, bubble size, froth depth, concentrate grade). Recommendations are autonomously applied within pre-approved control ranges; edge cases are escalated to metallurgists for approval.
- 340 live sensor inputs per circuit
- 60-second parameter optimisation cycle
- 8-year historical data model training
- Autonomous control within pre-approved ranges
Ore Sensing AI
Predictive Ore Characterisation
AI ore characterisation models predict incoming ore head grade, mineralogy, and hardness 2 hours before it reaches the mill — using ore block model geology data, drill core assay interpolation, and blast fragmentation image analysis (AI camera systems on the primary crusher analysing ROM fragment size distribution as a proxy for ore hardness). This 2-hour prediction window enables the circuit to pre-position parameters for the incoming ore batch — rather than reacting 2–4 hours after the ore enters the circuit, eliminating the core recovery lag that characterised manual control.
- 2-hour ore characterisation prediction window
- Drill core + block model geology integration
- AI ROM fragment size analysis (crusher cameras)
- Pre-emptive circuit parameter positioning
Energy AI
AI Comminution Energy Optimisation
AI comminution optimisation models — running on SAG mill and ball mill control systems — continuously optimise mill speed, load, and media filling levels against measured ore hardness and throughput targets. Soft ore detection via power draw monitoring enables automatic mill speed reduction, maintaining throughput and recovery while reducing energy draw. Average energy reduction: 23% per tonne processed, saving $14M annually across all 4 operations. Bond Work Index prediction from ore characterisation AI enables advance energy scheduling for site power procurement optimisation.
- SAG + ball mill real-time optimisation
- Soft ore detection via power draw AI
- 23% energy reduction per tonne processed
- Bond Work Index prediction for energy scheduling
AI Workforce Deployment
The Anicalls AI Metallurgy Operations Team
AI Process Control Agents — 4 Sites
Dedicated AI Process Control Agents at each of the 4 operations — each managing the full processing circuit from primary crusher through to final product dewatering. Operating continuously, each agent monitors 340+ sensor inputs, generates 60-second optimisation recommendations, applies approved autonomous adjustments, and escalates complex ore transitions or equipment anomalies to the metallurgist on shift. The agents have effectively codified the institutional knowledge of the group's top-quartile metallurgists — enabling consistent top-quartile performance on all shifts, regardless of individual operator experience level.
AI Laboratory Intelligence Agents
AI Laboratory Agents integrate all laboratory assay results as they are produced — correlating actual recovery performance with the AI model's predictions, identifying model drift (where ore characteristics have changed materially from the historical training data), and triggering model recalibration when prediction accuracy falls below threshold. Laboratory assay results — which previously arrived as a paper printout 2–4 hours after sample submission — are now integrated into the process control model within 8 minutes of analysis completion, closing the control feedback loop from hours to minutes.
AI Metallurgical Performance Agents
Daily metallurgical performance reports generated by AI — covering recovery performance versus theoretical, energy efficiency per tonne, reagent consumption efficiency, and model recommendation compliance rates — replaced the weekly manually-compiled metallurgical balance reports previously produced by the site metallurgists. Daily AI reports are delivered to site management and the group's central processing technology team, enabling faster identification and sharing of best-practice operating conditions across all 4 operations. Cross-site learning (optimal settings from Site A for a specific ore type shared to Site B) is now automated — previously dependent on occasional metallurgist knowledge-sharing sessions.
Technologies Used
The AI Technology Stack Deployed
Platform
MineProcess AI Platform™ — Australian Edition
Mining-specific AI process control platform — pre-integrated with all major industrial control systems (ABB, Siemens, Rockwell Automation DCS/PLC), OSIsoft PI Historian (the standard process data historian in Australian mining), LIMS laboratory systems (LabWare, SampleManager), and geological block model formats (Vulcan, Datamine, Leapfrog). All AI models run on edge computing hardware at each site — critical for remote WA operations where satellite internet latency (180–340ms) is incompatible with 60-second control cycle requirements. Privacy Act (Australia) compliant for operational data.
- ABB + Siemens + Rockwell DCS integration
- OSIsoft PI Historian integration
- Edge computing deployment (low-latency control)
- Vulcan + Datamine block model integration
Vision AI
Computer Vision Ore Sensing
Industrial computer vision systems at primary crushers and conveyor transfer points — using 4K cameras with AI image analysis to measure ROM fragment size distribution (proxy for ore hardness), ore colour (indicator of oxidation state and mineralogy), and conveyor load distribution (for belt loading optimisation). Vision-based ore characterisation provides a real-time proxy measurement in the 2–4 hours before assay results are available — enabling the process model to make informed parameter adjustments based on visual ore indicators while awaiting laboratory confirmation.
- 4K industrial camera systems (12 per site)
- ROM fragment size distribution analysis
- Ore colour oxidation state detection
- Conveyor load distribution optimisation
Reinforcement Learning
Reinforcement Learning Process Models
Deep reinforcement learning models — trained using 8 years of historical process data and reward functions weighted toward metallurgical recovery maximisation and energy efficiency — continuously refine their process parameter recommendations based on observed outcomes. Each 60-second control decision is logged with its outcomes (recovery improvement, energy draw, concentrate grade) and fed back into the model — enabling ongoing self-improvement without scheduled retraining cycles. Models at all 4 sites share learning from ore type transitions — creating a network learning effect across the group's operations.
- Deep reinforcement learning (8 years training data)
- Recovery + energy multi-objective reward function
- 60-second feedback loop for model improvement
- Cross-site network learning effect
Quantified ROI
The Revenue and Efficiency Impact at 12 Months
$112M Additional Revenue, 840% ROI
The 8.4% ore recovery improvement across 28Mt annual throughput at blended commodity prices generated $112M in additional annual revenue. Energy cost reduction contributed a further $14M in annual savings. Total annual financial benefit: $126M. Platform investment: $15M (implementation and 3-year licence). Year 1 net benefit: $111M. Year 1 ROI: 840%. Payback period: 43 days from go-live. The $112M revenue addition at no throughput increase represents pure margin improvement — with the group's EBITDA margin improving from 34% to 38% at the processing operations included in the programme.
23% Energy Reduction: $14M Annual Saving
The comminution energy optimisation reduced energy consumption per tonne processed by 23% — saving $14M annually across all 4 operations. At the group's Scope 2 emission factor for Western Australian grid power, the energy reduction also avoided 28,000 tonnes CO₂e annually — contributing to the group's CDR (Consumer Data Right — AU) and voluntary climate commitment emissions reduction targets. The energy efficiency improvement also reduced power demand charge peaks — the most expensive component of the group's electricity tariffs — by 18%, producing an additional $2.4M in tariff savings not included in the $14M headline figure.
Reserve Life Extension: Equivalent to 2.8 Additional Years
The 8.4% recovery improvement — recovering metal that was previously lost to tailings — is economically equivalent to extending the group's ore reserve life by 2.8 years, since the same in-situ ore body now yields 8.4% more contained metal. For the group's nickel operation specifically (facing declining ore grades from 2026), the recovery improvement is critical — generating $38M additional nickel revenue annually that partially offsets the grade decline impact, extending the economic mine life by 4.1 years at current nickel prices.
Business Outcomes
Australian Mining Group Transformation
ASX Rating
ASX Re-Rating on AI-Enabled Productivity
The $112M revenue addition — equivalent to a 12% revenue increase at no additional throughput — contributed to an ASX analyst re-rating of the group from Neutral to Outperform by three major brokers. The group's ASX-listed shares traded at a 22% premium to the pre-programme price at 12 months post-implementation. Analyst reports specifically cited the AI ore processing programme as evidence of operational excellence and management quality — a factor that commanded a valuation premium versus peer miners not deploying AI process optimisation.
Sustainability
28,000 Tonne CO₂e Avoidance
The 23% energy reduction avoided 28,000 tonnes CO₂e annually — contributing to the group's published net zero commitment and satisfying the Scope 2 reduction targets in the group's CDR-aligned climate risk disclosure. The energy efficiency improvement was cited in the group's TCFD (Task Force on Climate-related Financial Disclosures) report as a key near-term decarbonisation initiative, alongside the group's renewable energy procurement programme. Reduced energy intensity per tonne of ore also improved the group's positioning on the S&P Global Corporate Sustainability Assessment — contributing to ESG index inclusion.
Knowledge
Metallurgical Knowledge Preservation
The AI process models — trained on the operational decisions and outcomes of the group's top metallurgists over 8 years — effectively codified and preserved institutional processing knowledge that was at risk of loss through retirement and resignation. The group's two most experienced metallurgists (combined 38 years site experience) transitioned to "AI training and oversight" roles — formalising their knowledge into the AI models and providing oversight of model recommendations rather than performing manual process control. Knowledge transfer that previously took 5–8 years of apprenticeship now happens in 4–6 months of AI-supervised operations.
Executive Testimonial
"Mining has always said the ore body is fixed — you can't change what nature put in the ground. What Anicalls proved is that what you do with that ore body is not fixed at all. Our 8.4% recovery improvement is pure value that was already there, going to tailings every day because we couldn't react fast enough to ore variability. The 43-day payback was extraordinary — but what really changed our view of AI is the reserve life extension. We effectively got 2.8 years of additional ore reserve from our existing deposits. That's generational value for the business."
Metrics Dashboard
12-Month Processing Performance Scorecard
+8.4%Ore Recovery Improvement
$112MAdditional Annual Revenue
23%Energy Reduction per Tonne
$14MAnnual Energy Cost Saving
43 daysInvestment Payback Period
840%Year 1 ROI
2.8 yrsEquivalent Reserve Life Extension
28,000tCO₂e Annual Avoidance
Recover the Value Going to Your Tailings
See how MineProcess AI Platform™ can improve your ore recovery, reduce energy consumption, and generate $100M+ in additional revenue from your existing throughput — at no additional mining cost.