Real-Time Flotation Analytics
A 30-minute delay in detecting froth collapse costs up to $30,000. RxOps catches it in under a minute.
Computer vision monitoring for flotation cells
Froth condition, scraper health, and process anomalies tracked continuously
Requires no modifications to existing equipment
Schedule a demo with our experts
No commitment required at this stage
By submitting this form, you agree to the processing of your personal data.
Let us know what you have in mind
Describe your goals, and we'll outline a clear path to achieve them
By submitting this form, you agree to the processing of your personal data.
ABOUT
RxOps monitors your flotation cells continuously so recovery losses don't go unnoticed until it's too late.
RxOps is a computer vision monitoring system for flotation cells. It tracks froth condition, scraper health, and process anomalies in real time — detecting problems before they affect recovery, without modifying existing equipment.
THE CHALLENGE
Manual monitoring creates losses
that accumulate quietly
that accumulate quietly
-
Froth collapse is expensive — and slow to detectA single froth collapse event costs $5,000–$30,000 in lost recovery on a medium copper concentrator — before accounting for reagent waste and recovery time. -
5–8% reagent efficiency variability between shiftsOperator subjectivity drives consistent, compounding variability in reagent efficiency. That gap closes when monitoring is continuous rather than periodic. -
Gradual process drift erodes recovery silentlyDrift over weeks does not trigger alarms. It rarely gets attributed to its cause. It is only visible in retrospect, after the loss has occurred. -
Scraper failures cause unplanned downtimeCurrent inspection intervals are too infrequent to catch early-stage wear. Failures arrive without warning, not as a scheduled maintenance event.
HOW RXOPS WORKS
One camera per cell.
Four dimensions of monitoring.
Four dimensions of monitoring.
Cameras mount externally on existing cells. No mechanical modifications. The system runs on-site and learns your ore characteristics from day one.
01 Froth Monitoring
Tracks bubble size, froth velocity, and active surface area in real time. Provides 15–30 minute advance warning of process drift before it affects recovery.
- Bubble size distribution
Continuous measurement across the froth layer. Validated accuracy ±0.5 mm in the 0.5–50 mm range. - Froth velocity tracking
Optical flow algorithms measure froth movement. Accuracy ±0.05 m/s. Slowdown is an early indicator of collapse. - Active area measurement
Semantic segmentation measures the proportion of froth surface with active mineral loading. Accuracy ±2%. - Predictive trend forecasting15–30 minute forecast horizon for froth condition. Alerts before the process moves out of bounds, not after.
02 Scraper Сondition
Estimates blade wear, monitors motion stability, and detects pulp overflow — without physical sensors on the scraper mechanism.
- Wear estimation from videoBlade wear estimated at ±5% accuracy from camera imagery alone. No contact sensors required on rotating parts.
- Motion stability analysis
Detects irregular scraper movement patterns that precede failure — vibration, stutter, asymmetric rotation. - Pulp overflow detection
Identifies pulp creeping above the intended froth layer — a common sign of scraper underperformance or cell imbalance. - Precise defect localisation
Every anomaly mapped to exact cell coordinates, so maintenance teams know which section to inspect.
03 Anomaly Detection
Classifies froth collapse, over-aeration, scraper failure, and reagent starvation. Alerts are confidence-scored to keep false positives below 4%.
- Multi-class ML classificationFour anomaly types detected independently: froth collapse, over-aeration, scraper failure, reagent starvation. Precision 94%+.
- Confidence-scored alerts
Each alert carries a confidence score. Operators see how certain the system is before acting, reducing alert fatigue. - Sub-minute detection
30–60 second early warning window. Industry-standard manual inspection detects the same events 20–40 minutes later. - Monthly model retraining
Models retrain each month on your site-specific data. Accuracy improves over the deployment lifecycle, not just at commissioning.
04 Operator Interface
Live cell overview, shift performance summaries, and reagent recommendations. Optional closed-loop mode enables autonomous process correction.
- Live circuit overview
All cells visible simultaneously with AI overlays. Status, active anomalies, and trending metrics on a single screen. - Shift performance scorecardsEnd-of-shift summaries with reagent efficiency, anomaly count, response times, and comparison to prior shifts.
- Reagent recommendations
System generates dosing recommendations based on current froth condition and historical ore behaviour. - Closed-loop integration (optional)Direct OPC UA write-back to PID loops for reagent dosing, airflow, and pulp level. Autonomous correction without operator intervention.
VALIDATED PERFORMANCE
Accuracy validated under operating conditions
-
±2%
Froth area measurement
Field validated
-
±0.5 mm
Bubble size distribution
0.5–50 mm range
-
94%+
Anomaly detection precision
False positives < 4%
-
±5%
Scraper wear estimation
Vision-only, no sensors
-
±0.05
Froth velocity (m/s)
Optical flow method
Accuracy may decrease on highly mineralised froth with bubbles below 1 mm or heavily oily surfaces. A site-specific validation protocol is provided during the pilot phase.
BUSINESS IMPACT
Payback period depends on what you process
Based on actual field deployments on typical medium-size plants. Payback range: 6 to 14 months depending on ore type, current metal losses, and process efficiency.
Ore type / scenario
Throughput (metric tons per day)
Expected payback period
Copper concentrator (typical)
60,000 – 120,000
10-14 months
Nickel concentrator (typical)
30,000 – 60,000
8-12 months
Polymetallic / high-grade (Au, PGM, etc.)
20,000 – 40,000
6-10 months
Payback range: 6 to 14 months depending on ore type, current metal losses, and process efficiency.
What Makes RxOps Different
Four things that matter in practice
No equipment modifications
Cameras mount externally. Installation happens during normal operation — no shutdowns, no mechanical changes to cells. Commissioning does not interrupt production.
Models trained on your ore, not generic data
During the pilot, the system learns your specific ore type and operating conditions. Models retrain monthly. Accuracy improves over the deployment lifecycle.
One camera monitors froth and scraper simultaneously
The same unit tracks froth condition and scraper mechanics. There is no separate sensor system for scraper health — which is where most competing solutions stop.
Your data stays on your site
100% on-premise processing. No video or process data transmitted externally. Full model ownership, standard OPC UA / MQTT protocols, and complete API access — no vendor lock-in.
TECHNOLOGY
Hardware rated for the conditions you operate in
implementation
Pilot on 2–4 cells. Validated results before full commitment.
Five phases. Each ends with a defined deliverable before the next begins.
Site assessment
Camera placement, integration mapping, baseline OEE measurement.
Technical design
Installation
Hardware mounted during normal operation. No process interruption.
System operational
System operational
Pilot
2-week calibration on 2–4 cells, validated against manual sampling.
Initial ROI data
Initial ROI data
Scale
Full circuit coverage at 4–8 cameras per week.
Optimised models
Optimised models
Handover
Two days on-site operator and metallurgist training, documentation, support activation.
Certified team
Certified team
FAQ
Common questions
No. Cameras are mounted externally on standard housings. No mechanical modifications to cells or scrapers. Installation occurs during normal operation.
Standard housing suits dry climates with humidity below 85% and stable temperatures. Premium housing — with heated glass and internal temperature control — is recommended if you have high humidity, intense foam, extreme temperatures, or if existing cameras have experienced condensation. Our site survey determines the right configuration.
RxOps maintains a library of ore-specific models and switches automatically based on feed characterisation signals such as XRF or head grade trends. Accuracy is maintained through ore transitions without manual reconfiguration.
Entirely on-premise on edge devices and local servers. No video or process data is uploaded to external systems unless you explicitly authorise it. Full API access and model ownership are included as standard. An NDA and data protection agreement are signed before the pilot begins.
Full API documentation, admin access, and model retraining procedures are provided. Your team can manage and retrain models independently after the enablement session. There is no dependency on RxOps for ongoing system operation.
Ready to see the numbers for your operation?
We'll run through the system with your ore type, your circuit, and your current baseline — so you leave with something specific.
