The Monday morning production report lands in the digital manufacturing director's inbox at 07:00. Throughput for the previous week at the 25,000 tpd copper concentrator was 168,000 tonnes — 7,000 tonnes below the plan, representing approximately 1.4 million dollars of unrealised revenue at prevailing copper prices. The weekly reconciliation shows the shortfall traces back to three distinct flotation circuit disruptions: a collector dosage excursion on Tuesday night that dropped concentrate grade below specification for six hours, a froth depth instability event on Thursday that reduced recovery in the rougher bank by 4.2 percent, and a pH sensor drift that went undetected across two shifts and produced 1,200 tonnes of out-of-spec concentrate before the laboratory assay caught it. Each disruption was visible in the real-time process historian before it affected throughput. Not one of them generated a pre-emptive alert. The director reviews the findings, annotates the root causes, and forwards the report to the weekly continuous improvement meeting. Next week's report will show a different set of disruptions — same reactive cycle, same throughput penalty, same missed opportunity to intervene while the process data was already signalling the risk. This is the throughput ceiling that predictive scrap analytics exists to break: the structural gap between process data that contains the early indicators of scrap risk and a quality control model that only reads those indicators after the scrap has already been produced.
ML Grade Forecast · Recovery Optimization · Scarp Risk Alert · Adaptive SPC
The Throughput You Lost Last Week Was in the Process Data Hours Before the Scrap Occurred. Predictive Scrap Analytics Reads It Before You Lose It.
iFactory's predictive scrap analytics platform monitors 60+ flotation variables in real time, forecasts grade and recovery risk 2-6 hours ahead, and delivers ranked alerts that enable digital manufacturing directors to lift throughput 15-25% while reducing scrap events and reagent costs — sustaining Cpk above 1.67 across every ore zone and production campaign.
15-25%
Throughput increase reported by mining operations using predictive scrap analytics in flotation circuits — combining availability, recovery, and quality gains into measurable production lift
60+
Flotation process variables monitored per circuit — reagent dosages, air flow, froth depth, pH, pulp density, feed grade, and hydrodynamic conditions correlated simultaneously by the ML model
2-6 hrs
Early warning lead time predictive scrap analytics provides before a grade or recovery failure materialises — the intervention window that reactive quality control never opens
30-40%
Reduction in process disruptions and off-spec events when flotation circuits are managed with multivariate predictive models instead of univariate SPC limits and reactive assay follow-up
The Three Value Levers That Predictive Scrap Analytics Pulls for Throughput
Throughput in a flotation circuit is the product of three interdependent factors: the mass of ore processed per unit time, the recovery of valuable mineral from that ore, and the grade of the final concentrate. Predictive scrap analytics improves all three simultaneously by connecting the data that already exists across the circuit — the DCS historian, the online analyser readings, the reagent dosing logs — into a single forecasting model that alerts operators and quality leaders before any of the three factors degrades. The result is not incremental improvement on one variable. It is a compounding effect across all three that produces the 15-25 percent throughput lift that digital manufacturing directors report after deployment.
1
Scrap Prevention Recovers Lost Production Capacity
Every tonne of out-of-spec concentrate represents not just a quality failure but a throughput loss — the ore was mined, ground, and processed through the flotation circuit but produced no revenue. In a concentrator processing 25,000 tpd, a 2 percent scrap rate means 500 tonnes per day of lost productive capacity. The ML model forecasts scrap risk 2 to 6 hours before it occurs by detecting the multivariate pattern of an impending grade or recovery failure — a specific combination of reagent dosage drift, froth depth instability, and feed characteristic change that no single-variable alarm catches. The digital manufacturing director, alerted before the scrap occurs, can authorise a reagent adjustment or circuit reconfiguration that keeps the production running in-spec. The capacity that would have been lost to scrap is recovered as saleable concentrate. This is the first and most direct lever: preventing the throughput loss that reactive quality control accepts as unavoidable.
Recovers 300-700 tpd of lost capacity
Reduces off-spec events by 60-80%
↓
2
Grade-Recovery Optimisation Maximises Revenue Per Tonne
Grade and recovery exist in a direct trade-off in every flotation circuit: higher recovery typically means lower grade, and higher grade typically means lower recovery. The optimal operating point shifts continuously as feed characteristics, reagent effectiveness, and hydrodynamic conditions change. Fixed setpoint operation — the standard approach in reactive quality control — forces a suboptimal compromise that leaves throughput on the table. The predictive model identifies the optimal grade-recovery frontier in real time by correlating current process conditions against historical metallurgical outcomes, and recommends the setpoint adjustments that maximise net revenue. When feed grade drops, the model signals that maintaining the same recovery target will pull concentrate grade below spec — and recommends a controlled recovery reduction that keeps grade at specification while minimising the tonnage loss. When feed conditions improve, the model signals that the circuit can sustain higher recovery without grade risk, recovering additional concentrate tonnes that the fixed setpoint would have left in the tailings.
Dynamic grade-recovery frontier optimisation
2-4% recovery improvement without grade penalty
↓
3
Availability Improvement Reduces Unplanned Downtime
Flotation circuit disruptions — froth collapses, reagent system failures, pH excursions, air supply fluctuations — cause unplanned downtime that directly reduces throughput. In reactive quality control, these disruptions are detected when the process is already in distress and the circuit must be stabilised before production resumes. Predictive scrap analytics detects the precursor patterns that precede disruptions: a froth stability index declining over 30 minutes, a reagent pump flow trending toward the low alarm threshold, an air valve position drifting off setpoint. The system alerts the operations team while the disruption is still preventable — not after it has already forced a production stop. The cumulative effect on availability across a quarter is measurable: fewer unplanned stops, shorter disruption recovery times, and a higher proportion of scheduled operating time producing on-spec concentrate.
30-50% reduction in disruption-related downtime
3-6% availability improvement
Throughtput Optimisation · Predictive Analytics · Digital Manufacturing
Scrap Prevention, Grade-Recovery Optimisation, and Availability Improvement Compound Into 15-25% Throughput Lift. The Data Is Already in Your DCS.
iFactory's predictive scrap analytics platform activates all three value levers simultaneously — reading the same DCS data your team already collects and adding the pattern-recognition layer that transforms reactive quality control into a predictive throughput optimisation system.
The Digital Manufacturing Director's Maturity Model for Flotation AI
Digital manufacturing directors responsible for concentrator performance report that the transition from reactive to predictive quality management follows a consistent progression. Each stage builds on the previous one — adding analytical capability, data integration depth, and operational autonomy — while delivering incremental throughput gains that compound as the organisation moves through the maturity curve.
Stage 1: Reactive Quality Control
Quality is managed retrospectively through shift log reviews, laboratory assay follow-up, and corrective action reports generated after scrap events. The process historian contains rich data, but it is accessed only for incident investigation — never for forward-looking risk assessment. Throughput is constrained by the delay between process disruptions and their detection. Typical Cpk: 1.1-1.4. Typical throughput: baseline minus 10-15 percent avoidable losses.
Stage 2: Monitoring and Dashboards
Real-time dashboards display key process variables — reagent dosages, froth depth, air flow, pH, pulp density — with static SPC control limits. Operators and quality leaders can see current process state and receive alerts when individual variables breach limits. However, univariate SPC produces high false alarm rates during ore zone transitions, and the relationship between combined variable patterns and impending scrap risk remains invisible. Throughput improves modestly as reaction time to single-variable alarms shortens. Typical Cpk: 1.3-1.5.
Stage 3: Predictive Scrap Analytics
An ML model correlates all monitored variables simultaneously and produces a forward-looking scrap risk forecast with 2-6 hours of lead time. Alerts are ranked by confidence and projected impact — the quality leader sees the specific variable combination driving the risk and the recommended corrective action. Self-tuning SPC limits adapt to ore zone changes and reagent regime shifts, eliminating false alarms from normal process variability. Throughput improvement accelerates as scrap prevention, grade-recovery optimisation, and availability improvement compound. Typical Cpk: 1.67-1.85.
Stage 4: Autonomous Optimisation
The predictive model feeds setpoint recommendations directly into the DCS as operator guidance or, in fully validated scenarios, closed-loop control. Reagent dosage adjustments, air flow changes, and froth depth setpoints are optimised continuously by the model, with human oversight focused on exception management rather than routine setpoint decisions. The digital manufacturing director monitors circuit performance through a strategic dashboard showing throughput trajectory, Cpk trend, and model accuracy — not individual process variables. Throughput reaches and sustains the 15-25 percent improvement plateau. Typical Cpk: 1.8+.
The Director's Implementation Roadmap: From Data Audit to Throughput Lift
Digital manufacturing directors who have deployed predictive scrap analytics in flotation circuits report that the implementation follows a repeatable four-phase sequence. The timeline from project start to measurable throughput improvement is typically 12 to 16 weeks for the initial deployment on a single circuit, with subsequent circuits onboarding faster as the data infrastructure and model training pipeline are established.
1
Data Audit and Connectivity
Week 1-3. Audit available process historian data, identify data gaps, establish connectivity to DCS, online analysers, and reagent dosing systems. Most sites have 80%+ of required data already streaming.
2
Model Build and Shadow Mode
Week 4-8. ML model trained on 6-18 months of historical data with assay outcomes. Deployed in shadow mode alongside existing quality control — forecasts generated but not yet used for decisions. Accuracy validation against actual outcomes.
3
Live Alerts and Operator Training
Week 9-12. Live alerts activated for quality leaders and operators. Training sessions on alert interpretation, confidence scoring, and intervention protocols. First throughput improvements visible within 2-3 weeks of go-live.
4
Scale and Optimise
Week 13-16 onward. Model retrained with live intervention data, improving accuracy. Expanded to additional flotation lines and integrated with grinding and thickening optimisation. Continuous throughput improvement trajectory established.
Reactive vs Predictive: The Same Flotation Circuit After One Year
The difference between reactive quality control and predictive scrap analytics is not visible in a single shift comparison. It accumulates across production quarters as prevented scrap events, optimised grade-recovery trade-offs, fewer disruption-related stops, and a Cpk that holds above 1.67 even through ore zone transitions and seasonal feed variability.
Throughput
168,000 tpd average. 6-12 off-spec events per quarter. 3-5% throughput loss to scrap and disruption recovery.
Cpk Range
1.1-1.5. Fluctuates with ore zone transitions. Reactive limit adjustments lag process changes by days.
Reagent Cost
Fixed schedule dosage. Over-dosed in stable periods, under-dosed during transitions. 8-12% excess reagent consumption.
Quality Management Model
Corrective action after scrap events. Manual incident reporting. Retrospective root cause analysis.
Predictive Scrap Analytics
Throughput
193,000-210,000 tpd average. 0-2 off-spec events per quarter. Scrap prevented by 2-6 hour forecast lead time.
Cpk Range
1.67-1.85. Self-tuning limits adapt to ore zone changes automatically. Interventions before capability drops below 1.33.
Reagent Cost
Dynamic dosage guided by grade risk forecast. 8-15% reagent cost reduction while maintaining or improving Cpk.
Quality Management Model
Preventive intervention based on forecast risk. Automated audit trail. Continuous improvement from model retraining.
"
We deployed predictive scrap analytics on our copper flotation circuit in Q2 last year. The business case was built on a 12 percent throughput improvement target. By Q4, we were running 19 percent above the pre-deployment baseline. But the number that matters most to me is not the throughput percentage — it is the fact that we have gone from seven off-spec concentrate events per quarter to one in the past two quarters combined. The scrap that we are not producing is the purest form of throughput improvement because it comes with no additional ore mining, no additional grinding energy, and no additional tailings management cost. The model finds the risk in data we were already collecting. We just were not reading it together.
— Digital Manufacturing Director, Copper Concentrator — 25,000 tpd Sulphide Flotation Circuit, ISO 9001 and ISO 14001 Certified Operation
Conclusion
Predictive scrap analytics transforms the digital manufacturing director's relationship with flotation circuit performance. Instead of receiving the weekly production report that accounts for throughput lost to scrap events that already happened, the director receives a forward-looking risk assessment that shows where throughput is at risk in the next 2 to 6 hours — with ranked alerts that specify the variable combination driving the risk and the recommended corrective action. The shift is structural: quality management moves from corrective action after scrap events to preventive intervention before scrap occurs, and throughput moves from a number that is reported after the fact to a number that is actively protected in real time.
The data infrastructure required already exists in every modern concentrator. The DCS historian records every reagent dosage change, every froth depth reading, every pH measurement, every air flow adjustment. The online analyser produces a continuous stream of head grade and concentrate grade data. The laboratory information system holds years of assay results that map process variable patterns to metallurgical outcomes. What has been missing is the analytical layer that reads all these data streams simultaneously and recognises the multivariate signatures that precede a scrap event — the specific combination of reagent drift, froth instability, and feed characteristic change that no single-variable alarm can detect. That analytical layer is what predictive scrap analytics provides.
iFactory's predictive scrap analytics platform is built for digital manufacturing directors managing flotation operations in copper, gold, iron ore, and other mineral processing circuits — delivering ML-driven grade and recovery forecasts with 2-6 hour lead times, self-tuning SPC that adapts to ore zone transitions automatically, automated audit documentation for ISO compliance, and dynamic reagent optimisation intelligence that reduces cost while improving throughput and Cpk simultaneously. Book a Demo to see the platform running on a flotation use case matched to your circuit configuration, or talk to an expert about a free throughput and process capability assessment for your operation.
Frequently Asked Questions
The Scrap That Reduced Last Quarter's Throughput Was in Your Process Data Hours Before the Assay Confirmed It. Get a Free Throughput Assessment.
iFactory's predictive scrap analytics platform forecasts flotation grade and recovery risk 2-6 hours ahead, activates the three value levers that compound into 15-25 percent throughput lift, and sustains Cpk above 1.67 through ore zone transitions and seasonal variability. A free assessment shows your current throughput improvement opportunity based on 12 months of your existing process historian data.
