Quick answer
When a cement cross-belt elemental analyzer starts disagreeing with laboratory XRF, do not recalibrate it immediately. First prove that both systems are evaluating the same material population over the same time window. A persistent offset can come from sample timing, non-representative grab sampling, laboratory preparation bias, belt-loading or moisture changes, analyzer geometry, instrument health, or a true calibration drift. The useful question is not “which result is right?” but “what pattern does the analyzer-to-lab residual follow, and which part of the measurement chain can create that pattern?”
Problem / job
The online analyzer and the central laboratory no longer agree closely enough for operators to trust automatic raw-mix or stockpile control.
Evidence
Use synchronized analyzer averages, lab sample timestamps, sample-point details, belt load, moisture, material changes, lab preparation method, duplicate results, calibration history and analyzer health alarms.
Decision
Separate time/sample mismatch, laboratory repeatability error, process-presentation effects and true analyzer calibration drift before changing equations or feeder-control logic.
As-found evidence capture
Freeze the diagnostic baseline before anyone changes a calibration coefficient. Export the online analyzer values for the period surrounding each laboratory sample and record whether the analyzer value is instantaneous, a rolling average or an integrated average over a defined belt interval. Record the laboratory sample collection time, exact sampling point, transport delay, preparation method and analysis method. Also capture belt loading, material bed depth, conveyor speed, moisture trend, raw-material feeder changes, stockpile transition, dust return changes and any recent maintenance on the analyzer or conveyor.
Preserve the analyzer calibration history and instrument-health log. Note detector/source warnings, temperature-control alarms, communication interruptions, software or equation changes, and any recent mechanical work that could have altered the measurement geometry. For the lab side, retain duplicate or repeat results rather than only the final reported number. A diagnostic comparison is weak if one side is a time-averaged full-stream measurement and the other side is one small sample taken from a changing stream.
Controlled comparison checklist before changing calibration
- Define the exact online averaging window that corresponds to the material sampled for the laboratory.
- Account for conveyor travel time, sampler location and sample transport delay.
- Use the same homogenized laboratory sample for duplicate preparation/analysis checks.
- Record belt loading, conveyor speed, moisture and major raw-material changes during each paired comparison.
- Verify sampler cutter/path condition and that the sample is representative of the full material stream.
- Check laboratory QC/CRM performance before treating the lab value as the reference.
- Preserve analyzer health logs, calibration revision, software/equation version and recent maintenance history.
- Collect several paired points under controlled conditions before calculating any calibration correction.
Stop/go decision: do not recalibrate if the residual disappears when timing, sampling or lab preparation is corrected. Recalibration becomes justified only when a repeatable bias remains after both measurement chains and the process presentation have been proven stable.
Field go/no-go sequence
- Freeze calibration coefficients and automatic correction logic.
- Synchronize the online averaging window to the sampled material interval.
- Run duplicate laboratory preparation/analysis on the same homogenized sample.
- Verify sampler representativeness and inspect belt presentation.
- Collect several controlled online-versus-lab pairs.
- Plot the residual against concentration, belt load, moisture and process changes.
- Correct sampling/lab/process-presentation causes first if the residual follows them.
- Only then apply the analyzer supplier’s approved calibration procedure if a repeatable bias remains.
Eight-step diagnostic workflow
1. Synchronize the material population
Start by aligning the comparison window. The laboratory sample must represent the same material that passed through the online analyzer. Account for conveyor travel, sampler location, transport delay and sample-preparation delay. If the process composition is moving, comparing a later lab sample with an earlier online average can create an apparent calibration error even when both measurements are functioning correctly.
2. Prove laboratory repeatability first
Prepare and analyze duplicate portions of the same sample using the normal plant method. If repeated lab preparation changes the result materially, the comparison cannot yet isolate analyzer drift. Check grinding, pellet or fused-bead preparation, sample homogeneity, contamination, flux/binder handling where applicable, and instrument quality-control checks. A stable online trend paired with erratic lab duplicates points to the offline chain before it points to the cross-belt analyzer.
3. Verify that the reference sample is representative
Inspect the sampling device and sample path. A spot sample taken from one region of a segregated stream may not match an analyzer that measures a much larger fraction of the belt cross-section. Check for sampler wear, partial blockage, biased cutter travel, poor mixing, segregation during transport and contamination from the previous sample. The reference must be representative before it can be used to correct the online instrument.
4. Check belt presentation and process condition
Review material bed depth, loading stability, conveyor speed, moisture, large-particle segregation and unusual voids in the stream. Compare current conditions with the conditions under which the calibration was established. A measurement system can remain healthy while process presentation shifts enough to change the residual pattern. Look for correlation between residual error and belt load or moisture rather than treating every disagreement as a constant bias.
5. Analyze the residual pattern element by element
Calculate the difference between synchronized online and laboratory results for the controlled constituents or derived moduli. A nearly constant offset suggests a different cause from a slope that grows with concentration. Intermittent spikes suggest another cause again. If all major elements shift together after one sampling event, investigate sample identity and timing. If one element develops a concentration-dependent deviation while others remain stable, review the relevant calibration behavior and spectral/instrument evidence.
6. Inspect analyzer health and change history
Review internal diagnostics, calibration checks, source/detector status, temperature stability, electronics alarms and any mechanical changes around the analyzer. Confirm that no software, regression equation or reference-data update was introduced at the time the disagreement began. Do not assume a healthy status screen proves accuracy, but use it to separate a process/sample problem from an obvious instrument fault.
7. Run a controlled comparison sequence
Collect a planned series of reference samples under stable operation rather than reacting to one discrepant point. Use consistent timing and sample handling. Compare the online average over the correct material interval with independent laboratory results and document each pair. The objective is to determine whether the residual is stable, random, process-dependent or concentration-dependent.
8. Change calibration only when the error mechanism is proven
If the controlled comparison demonstrates a repeatable analyzer bias after laboratory repeatability, sample representativeness and process presentation have been verified, then follow the analyzer supplier’s approved calibration procedure. Preserve the previous calibration and document why the change was made. If the residual disappears when sampling or lab preparation is corrected, do not compensate the online analyzer for an error that originated elsewhere.
Evidence-to-decision table
| Observed pattern | Most useful next check | Decision direction |
|---|---|---|
| Online value is smooth but lab points scatter | Duplicate lab preparation and analysis | Prove laboratory repeatability before analyzer adjustment |
| Difference appears mainly during raw-material changes | Time alignment and sample transport delay | Correct comparison window before recalibration |
| Error changes with belt load or moisture | Belt presentation and calibration-condition match | Investigate process-presentation effect |
| One sampler gives biased results while another agrees | Sampler cutter/path, mixing and contamination | Repair sampling chain |
| Stable offset persists across controlled paired samples | Analyzer calibration history and reference sequence | Evaluate approved calibration correction |
| Deviation starts after software or maintenance change | Change log, geometry and equation revision | Restore or validate changed configuration |
Verification and return-to-control criteria
Return the online analyzer to automatic control only after synchronized paired comparisons show stable, explainable agreement and the laboratory reference method itself is repeatable. Confirm that sample identity and timing are traceable, analyzer health alarms are clear, process presentation is within the intended operating condition and the calibration history is documented. If the analyzer is connected to raw-mix optimization, restore automatic feeder correction only after the measurement chain is trusted; do not use control action to hide an unresolved measurement problem.
Use the installed analyzer and plant laboratory procedures for quantitative acceptance limits. The acceptable residual depends on the instrument, constituent, calibration range, process variability and plant quality-control scheme, so this guide intentionally avoids inventing universal tolerances.
Common mistakes and repeat-failure traps
- Recalibrating from one bad pair: one mismatched time window can permanently bias an otherwise healthy analyzer.
- Treating the lab as automatically perfect: grinding, pressing, fusion, contamination and sample identity can all introduce error.
- Ignoring transport delay: the sample reaching the lab may represent a different process state from the online average selected for comparison.
- Using an unrepresentative grab sample: a small biased sample is not a gold standard for a full-stream measurement.
- Changing feeder control before measurement is proven: control-loop movement can make diagnosis harder by changing the material while the measurement problem is still unresolved.
- Deleting the old calibration: always preserve the previous state so a poor correction can be reversed and audited.
Related Infinity guides
Build a stronger cement technical library
The Infinity for Cement Organization technical package contains 900+ technical files for cement professionals. It is a one-time $249.99 purchase, with instant download immediately after payment and lifetime updates forever at no additional cost.