Hourly blindness
Between samples, the burn drifts with no quality feedback at all.
Agent A3 · Quality sensing
Know the clinker before the lab does.
The clinker-quality agent predicts free lime, alite and belite phases, litre weight, and defects continuously, turning hourly lab samples into a live signal the kiln agent can act on.
±0.12%
Free-lime prediction error
42min
Lead time vs. lab
−38%
Free-lime rejects
24/7
Continuous, not hourly
You steer the kiln on data that's already an hour old.
Free lime and phase come from lab samples pulled once an hour. By the time a bad result lands, the plant has already made an hour of off-spec clinker.
After-the-fact
A high free-lime result means the last hour is already suspect.
Strength scatter
Phase variability shows up weeks later as 28-day strength scatter.
Warranty risk
Unsound cement erodes product quality and warranty exposure.
Perception
A soft-sensor for clinker chemistry.
The agent learns the mapping from process, raw-mix, and burn signals to lab-measured quality, then predicts free lime and phases continuously with calibrated uncertainty.
- Free-lime predictor updated every minute, not every hour.
- Phase estimation for alite (C₃S) and belite (C₂S).
- Auto-calibration to each new lab sample as it arrives.
Free lime (pred.)0.86 %
Alite / C₃S64 %
Belite / C₂S16 %
Litre weight1,320 g/l
Confidence± 0.12 %
Control
It closes the loop with the kiln agent.
Prediction is only useful if it changes the burn. The quality agent feeds the kiln agent a live free-lime and phase target error so corrections happen before the clinker is made.
- Live target error streamed to the kiln agent.
- Drift alerts when predicted quality leaves spec.
- Lab reconciliation keeps the model honest over time.
Target free lime1.0 %
Predicted error−0.14 %
Kiln correctionissued
Next lab in18 min
Model driftnominal
From signal to on-spec clinker.
Ingest signals
Fuse raw-mix, process, and burn data into feature streams.
Predict quality
Estimate free lime and phases with calibrated uncertainty.
Feed the kiln
Stream target error so the burn corrects before clinker forms.
Reconcile
Auto-calibrate to each lab result to hold accuracy.
What the quality agent predicts
Primary targets
Free lime, alite (C₃S), belite (C₂S), litre weight.
Secondary
C₃A, C₄AF, SO₃, LSF back-calculation, coarse/dusty defect flags.
Lead time
Typically 30–45 minutes ahead of the next lab sample.
Inputs
Raw-mix chemistry, kiln process tags, burner camera, cooler data.
Calibration
Continuous reconciliation to XRD / free-lime titration results.
Output
Streamed to the kiln agent and to your historian / dashboards.
What continuous quality moves
Legacy control vs. Clinkum autonomy
Manual & rule-based
- Hourly lab samples
- Free lime known after the fact
- Operators infer quality by feel
- Strength scatter found weeks later
- No feedback between samples
Clinkum
- Minute-by-minute prediction
- Free lime pre-empted
- A calibrated soft-sensor everyone trusts
- Phase held tight, strength stabilized
- Continuous closed-loop feedback
“We used to fly blind between lab samples. Now free lime is a live number, and the kiln corrects before the clinker is even made.”
Quality Manager · Grinding & integrated group
Common questions
No. The lab remains the reference. The agent predicts between samples and reconciles to every lab result, so the two reinforce each other.
Typical prediction error is around ±0.12% after calibration, with a calibrated confidence band that widens when inputs are uncertain.
Raw-mix chemistry, kiln process tags, and ideally a burner camera and cooler data. It works with your existing instrumentation.
Yes, as a soft-sensor feeding your dashboards, but it is most valuable closed-loop with the kiln agent.
Adjacent agents
Make clinker quality a live signal.
Pilot the soft-sensor against your lab for a month and watch the free-lime prediction track. Then close the loop.