Free lime is the number a kiln crew waits on, and it arrives late. By the time a lab result confirms under-burnt clinker, the kiln has produced hours of material under the same conditions. Prediction closes that gap by estimating free lime from raw mix chemistry and kiln process data the plant already records. It only works when the instruments feeding it are healthy and the response is organised, which is why plants pair it with a cement maintenance management platform that keeps the data trustworthy.
Cement Quality and Process | Kiln and Raw Mix Analytics
Cement Free Lime Prediction from Kiln and Raw Mix Data
Predict clinker free lime before the lab result arrives, and keep the sensors, samplers and kiln equipment behind the prediction maintained, evidenced and acted on through one maintenance workflow.
Raw mixLSF, silica and alumina moduli
to
Kiln dataTemperature, NOx, O2, fuel, feed, speed
to
Predicted f-CaOContinuous estimate with alert band
to
Maintenance actionWork order, inspection, calibration
Why Free Lime Is the Most Watched Clinker Number
What free lime actually tells you
- Free lime is uncombined calcium oxide left in clinker when the burning zone has not fully converted the raw mix into calcium silicate phases.
- It is a readout of everything upstream: raw mix chemistry, burning zone stability, feed homogeneity, fuel quality and kiln speed.
- Published plant targets commonly sit around 0.5 to 1.5 percent, although each plant sets its own band.
- Values drifting well above the plant limit point to under-burning and risk soundness problems in the cement.
Free lime control band (illustrative, plant-specific)
Low: over-burnt, hard clinker, more grinding energy
Target band
High: under-burnt, soundness risk
Set the actual limits from your plant specification and lab history.
Cost sits on both sides of the target
Burning too hard
- Higher fuel consumption per tonne of clinker
- Harder clinker that costs more to grind
- More refractory stress and ring formation risk
Burning too soft
- Unconverted lime reaching the cement mill
- Strength variability and soundness concerns
- Rework, blending or downgraded product
The Lab Delay Problem: Before and After Prediction
Before: lab-only feedback
- Sample taken from the clinker cooler discharge
- Preparation and analysis in the lab
- Result entered and sent to the control room
- Operator adjusts hours after conditions changed
Conservative margins become the default, which usually means extra fuel.
After: prediction plus lab verification
- Model estimates free lime continuously from process data
- Alert fires when the estimate trends toward the limit
- Operator corrects early; lab samples validate the model
- Instrument faults found by the model become work orders
The lab remains the reference. Prediction fills the gap between samples.
Input Data: Which Signals Drive the Prediction
| Signal group | Typical variables | Why it matters for free lime | Maintenance dependency |
|---|---|---|---|
| Raw mix chemistry | LSF, silica modulus, alumina modulus, kiln feed fineness | Chemistry drift changes how much heat the mix needs to burn clean | Sampler, XRF and raw mill condition |
| Burning zone | Zone temperature, kiln torque or amps, flame condition | Direct indicator of burning intensity | Thermocouple and pyrometer calibration |
| Gas analysis | O2, CO, NOx at kiln inlet | NOx often tracks burning intensity | Gas analyser sampling lines and filters |
| Fuel system | Fuel flow, calorific value, ash and moisture | Ash infiltration shifts mix chemistry without warning | Feeder, weigh belt and flow meter accuracy |
| Feed and speed | Kiln feed rate, kiln speed, secondary air temperature | Sets residence time and heat balance | Feed weigh-feeder, kiln drive, cooler fans |
From Data to Decision: The Prediction Workflow
1
Clean and align data
Remove outliers and sensor dropouts; align raw mix results with the kiln time they affect.
2
Train against lab results
Use historical lab free lime as the target and hold back recent data to test the model.
3
Publish to operators
Show the estimate with a confidence band and alert limits on the control room screen.
4
Validate against every lab sample
Track prediction error and flag when the gap grows beyond an agreed tolerance.
5
Investigate the cause of drift
Process change, new fuel, or a failing instrument? Raise the right work order for each.
Choosing a Modelling Approach
| Approach | Best used for | Strength | Watch out for |
|---|---|---|---|
| Neural network soft sensor | Estimating the free lime value from many correlated variables | Handles non-linear kiln behaviour; academic work has reported satisfactory free lime and phase prediction | Needs well-cleaned data and retraining after process changes |
| Classification (good or under-burnt) | Giving operators a practical yes or no signal | Research suggests experts often prefer quality class over absolute value because of process complexity | Coarse; less detail near the limit |
| Statistical process control | Spotting drift in raw mix and burn zone stability | Simple, transparent, familiar to quality teams | Does not itself estimate the value |
| Hybrid | Model estimate plus SPC alarms plus lab validation | Balanced and easy to audit | Needs clear ownership across process and maintenance |
Root Causes Behind Free Lime Excursions
Raw mix chemistry drift
A change in limestone quality, a blending ratio shift or a worn proportioning feeder moves LSF away from target. Higher saturation needs more heat to combine the lime, so the same burning zone conditions that gave clean clinker yesterday now leave uncombined lime. Raw mill sampling and XRF reliability are therefore part of the free lime story, not a separate quality topic.
Kiln feed variability
Poor homogenisation in the blending silo, coarse residue in the kiln feed, or surging feed from a worn weigh feeder all change how evenly the material burns. Coarser quartz and limestone particles are harder to combine, which raises free lime even when temperatures look stable on the screen.
Fuel quality and flame shape
Coal ash absorbed into the clinker changes chemistry, and moisture or fineness changes alter flame length and heat release. Alternative fuel feeding adds another layer of variation. A blocked burner channel or a worn burner tip can change flame shape without any process setpoint being touched.
Heat exchange and cooling
Preheater cyclone blockages, false air leaks, calciner instability and cooler fan or grate problems alter calcination degree and secondary air temperature. Each of these arrives at the burning zone as a thermal disturbance that the free lime model needs to see correctly.
Why this matters to the maintenance team
- Many of these causes are equipment conditions: feeders, burners, cyclones, fans, seals and refractory.
- When prediction flags a trend, the first question is whether the process moved or the equipment degraded.
- A shared record of past interventions lets the team answer that question quickly instead of guessing.
Where Predictions Go Wrong: A Maintenance Risk Matrix
Failure source
Effect on prediction
Detection
Impact
Drifting thermocouple or pyrometer
Burning intensity misread
Calibration check, cross-comparison
High
Blocked gas sampling line
O2 and NOx signals lag or flatten
Inspection, signal flat-line alert
High
Fuel feeder or weigh belt error
Fuel input assumed wrong
Calibration and flow audit
Medium
Sampler fault on kiln feed
Raw mix chemistry unrepresentative
Sampler PM, lab cross-check
High
Kiln equipment change
Model assumptions no longer valid
Change records, retraining trigger
Medium
Late lab data entry
Validation misaligned in time
Timestamp discipline
Low
A Good Model on Bad Instruments Is Still a Bad Number
Keep every kiln and raw mix sensor on a schedule, with calibration and repair history in one place.
Reading the estimate alongside process context
- A rising estimate with stable raw mix chemistry usually points toward burning zone, fuel or feed behaviour.
- A rising estimate after an LSF shift points toward the raw mill and proportioning system.
- An estimate that moves while every other signal stays flat deserves an instrument check before any process change.
- An estimate that disagrees with consecutive lab results suggests retraining or a data alignment problem.
Operators trust a prediction when it behaves consistently with what they can see in the kiln. Keeping instruments reliable and documenting every explanation for a deviation builds that trust faster than any model improvement.
Turning Prediction Alerts into Maintenance Action
Process response
Operator checks fuel, feed and speed. Process engineer reviews raw mix targets. Lab runs an extra sample if the estimate is near the limit.
Instrument response
If the estimate diverges from the lab without a process explanation, raise a corrective work order for the suspect sensor, sampler or analyser.
Equipment response
Persistent burning zone instability can trace to feeders, kiln drive, cooler or refractory condition, and goes to the maintenance planner.
How Oxmaint supports this workflow
- Asset management keeps the kiln, raw mill, samplers, analysers and feeders in one register with their history.
- Preventive maintenance schedules calibration and cleaning for the instruments that feed the model.
- Work orders capture each sensor fault or drift investigation, from alert to closeout.
- Mobile inspections let technicians record checks at the kiln and sampling points on the spot.
- Dashboards and reports show repeat instrument problems and downtime, supporting decisions on spares and upgrades.
- Inventory tracks critical spares such as thermocouples, sampling filters and analyser consumables.
A Practical Rollout Timeline
Phase 1
Audit the inputs
List every sensor, sampler and analyser the model will read. Record its location, owner, last calibration and known faults. Fix obvious problems before any modelling begins, because training on faulty signals teaches the model the wrong behaviour.
Phase 2
Build and test offline
Train on historical data, compare against held-back lab results, and review errors with process engineers. Treat large misses as clues about missing variables or instrument issues rather than only as model failures.
Phase 3
Run in shadow mode
Display predictions beside lab results without asking operators to act on them. This builds trust, and exposes timing mismatches between process data and lab sampling.
Phase 4
Go live with response rules
Agree alert limits, who responds, and which work order types the alerts can create. Review prediction error weekly at first, then monthly once the model is stable.
Common Mistakes to Avoid
Treating the model as a finished product
Plan retraining after fuel, raw material or major equipment changes, and log those changes where the model team can see them.
Ignoring instrument calibration history
Keep calibration dates and repair records for each model input, so a drifting estimate can be matched to a sensor event.
Alerting without a defined owner
Assign each alert type to a role, with clear escalation from operator to process engineer to maintenance planner.
Trusting prediction over the lab entirely
Keep regular lab sampling, and use each result to score the model rather than assume it is right.
What a good response record looks like
- The prediction alert, its time and the estimated value at that moment.
- The operator and process engineer actions taken, with the lab result that followed.
- Any instrument or equipment fault found, with the work order that corrected it.
- The outcome: whether free lime returned to the target band, and how long it took.
Over a few months this record shows which alerts were real process events and which were instrument problems. That split tells the plant where to spend maintenance effort and where to adjust the model.
KPIs to Track the Programme
Prediction error
Gap between predicted and lab free lime on every sample
Time in target band
Share of kiln hours inside the plant free lime limits
Sensor calibration compliance
Completed versus planned calibrations on model inputs
Instrument-related model alerts
Drift events traced to a failing instrument
Fuel per tonne of clinker
Tracks whether tighter control reduces conservative margins
Reviewing the KPIs together
- Hold a short monthly review with process, quality and maintenance in the same room.
- Compare prediction error trends with the calibration and repair history of the model inputs.
- Decide whether each repeat deviation needs a process change, a maintenance task or a model update.
- Record decisions so the next review starts from evidence rather than memory.
Readiness Checklist Before You Deploy
Data readiness
- Lab free lime results timestamped accurately
- Kiln signals logged at consistent intervals
- Raw mix results linked to production time
- Known sensor faults documented
Maintenance readiness
- Calibration schedule for each input instrument
- Owner assigned for each sampler and analyser
- Critical spares identified and stocked
- Work order route defined for model-driven alerts
Operational readiness
- Alert limits agreed by process and quality
- Operator response steps written down
- Retraining triggers defined after major changes
- Review cadence set for model performance
Frequently Asked Questions
Can free lime be predicted without new instruments?
Often yes, because models can use variables already in the control system. Their accuracy still depends on sensor health, which is where maintenance records matter.
Does prediction replace lab testing?
No. Lab results remain the reference for training and validating the model. Prediction provides guidance between samples.
Why does model accuracy fall over time?
Raw materials, fuels, equipment condition and instruments change. Track error against each lab sample and retrain when the gap grows.
Which maintenance tasks matter most for prediction quality?
Calibration of burning zone and gas sensors, sampler upkeep and feeder accuracy checks. You can book a demo to see how these are scheduled.
Should we predict the exact value or a quality class?
Both are used. Some plants favour a good or under-burnt signal for operators and keep the numeric estimate for engineers.
Keep the Data Behind Your Free Lime Prediction Trustworthy
Organise kiln and raw mix instrument maintenance, calibration and corrective work in Oxmaint so predictions stay accurate and operators can rely on them.
