Blending silo and kiln feed
The last chance to fix the chemistry before it becomes clinker. It is also the stage with no symptom of its own, measured in the wrong place at most plants, and blamed for problems that were created three machines upstream.
On this page · 16 sections
- 01The one idea
- 02The parts — four measurement points, one of which matters
- 03First principles — a silo is a filter, and filters have a frequency
- 04What must be true
- 05The quantitative anatomy
- 06What goes wrong here
- 07Scenarios from the field
- 08The numbers that matter
- 09Build the model — the variance budget
- 10Upgrade paths
- 11The frontier, and who is running it
- 12What’s changing now
- 13Check yourself
- 14Mini case — the deviation that met target and the kiln that did not
- 15Go deeper
- 16Carry forward
§ 01 The one idea
The whole variance chain of A2 exists to deliver one number: the standard deviation of LSF in the kiln feed, and the target is below 1.
A kiln can absorb a certain amount of chemistry variation by burning defensively — running a little hotter than it needs to, so that the bad hours still make clinker. F6 §03 priced that: the fuel rate is set by the worst hour and everybody pays for it every hour. Beyond about 1.5 to 2 points of LSF swing over a few hours, defensive burning stops being enough and the kiln starts producing free lime, rings and quality excursions instead.
Ask a plant for its LSF standard deviation and you will usually be given a number from the raw mill outlet, because that is where the routine sample is taken. That is the wrong side of the silo. It tells you what the mill delivered, not what the kiln received, and the entire purpose of the silo lies between the two. The number you need is at kiln feed, and many plants do not routinely produce it.
§ 02 The parts — four measurement points, one of which matters
| Silo type | Blending ratio | How it works |
|---|---|---|
| Continuous, single silo | 5–7× | Air-fluidised sectors activated in sequence, drawing from many levels at once. The modern standard. |
| Continuous, two silos | 7–12× | Better, and roughly twice the capital. |
| Gravity blending, well designed | up to 15× | The high end of what silo technology delivers. |
| Batch blending | high, but discontinuous | Older design. Mix a batch, test it, correct it, discharge it. Excellent blending, poor availability, largely displaced. |
§ 03 First principles — a silo is a filter, and filters have a frequency
A blending silo removes variation by mixing material that entered at different times. How far apart in time it can mix is set by its residence time — typically a few hours of kiln feed. That single fact determines what a silo can and cannot do, and it is almost never stated.
Variation faster than the residence time is averaged away. A ten-minute excursion in mill product is mixed with material from hours either side of it and disappears. Variation slower than the residence time passes straight through. If the raw mix has been drifting for two days — because the quarry moved to a different face, or an additive feeder is out of calibration — the silo mixes drifting material with drifting material and delivers a drift.
Standard deviation is a single number that says nothing about timescale. Two kiln feeds with identical deviation can behave completely differently: one wobbling rapidly around target, the other sitting 2 points high for six hours and 2 points low for the next six.
The kiln does not care about the deviation. It cares about the excursion. A six-hour period at LSF +2 is six hours of hard burning, rising free lime and a burner operator adding fuel — and it will look identical, in the monthly report, to a day of harmless noise.
What to ask for instead: the LSF time series at kiln feed, hourly, for a fortnight, plotted. Not the deviation. Two minutes of looking at a chart will tell you what no summary statistic can, and it will tell you immediately whether you are looking at a silo problem or an upstream one.
The diagnostic rule that follows is clean: if the variation is fast, the silo is your problem. If it is slow, the silo cannot help you and the answer is upstream — in the pile (A2), the mix control (A3) or the quarry (A1).
§ 04 What must be true
The blending silo is the only major vessel in a cement plant with no instrument of its own. Material goes in, material comes out, and almost nowhere in the world is both sides measured. That is why its conditions have to be stated explicitly — there is no alarm that will tell you any of them has failed.
| What must be true | Why | Target | How you verify it |
|---|---|---|---|
| 1. The silo is measured on both sides | Blending performance is a ratio: inlet standard deviation over outlet standard deviation. One side alone tells you nothing. Almost every plant samples the mill outlet and calls it kiln feed. | H = σin / σout | Two hourly sample sets, silo inlet and kiln feed, for a fortnight. Correct for analysis error — H = √((σ²in − σ²an) / (σ²out − σ²an)) — or a precise lab will look like a bad silo. |
| 2. The inlet is variable enough for the test to mean anything | A low H can mean a silo that does not blend, or a mill that already delivered uniform meal. The two are opposite findings and the same number. | — | Establish σin first. If the incoming meal is already inside the kiln’s tolerance, the silo has nothing to do and a low ratio is not a fault. This single check prevents the most common misdiagnosis in this part of the plant. |
| 3. The deviation is held, not the mean | A kiln burns the hour it is given, not the monthly average. A silo outlet on target with a wide hourly spread produces free-lime excursions the operator corrects with fuel. | kiln feed LSF SD < 1 |
Plot the hourly series, not the summary statistic. Ask for the control chart, not the certificate. |
| 4. Extraction cuts across the whole floor | A continuous blending silo works by withdrawing from many points at once so that material of different ages mixes on the way out. Blocked or inactive aeration sectors convert a blending silo into a storage silo quietly and completely. | all sectors cycling |
Watch the aeration sequencing for a full cycle and check the blower or compressor current against the sector valve positions. A dead sector shows up nowhere else. |
| 5. The material flows as mass flow, not funnel flow | In funnel flow a central channel empties while material at the wall sits. The silo then holds two populations and discharges the newest meal first — the opposite of blending. Severe cases rathole, and in the worst case the structure is loaded in a way it was not designed for. | — | Level profile against time during a draw-down, and the discharge chemistry trend through it. A clean first-in-first-out signature is mass flow; a step change is not. |
| 6. The kiln feed rate the kiln believes is the rate it gets | Everything downstream — heat rate per tonne, specific power, clinker factor — is divided by this number. A weighfeeder that drifts corrupts every operating KPI at the plant simultaneously, and does so without producing a single alarm. | — | Ask when the kiln feed weigher was last calibrated against a known mass, not against itself. Then compare the feed total with the clinker produced and the loss on ignition. If those do not close, nothing else you compute is safe. |
Rows 1 to 5 decide how well the silo blends. Row 6 decides whether any number you compute about the plant is true. A 2% drift on kiln feed mass moves kcal/kg, kWh/t and clinker factor by 2% each, in the same direction, invisibly. It is the cheapest check in this course and the one most often skipped, because the instrument reports a number continuously and therefore looks as though it is being measured.
§ 05 The quantitative anatomy
| Metric | Target | Note |
|---|---|---|
| LSF standard deviation, kiln feed | <1 | The number the whole chain exists to produce. At kiln feed, not mill outlet. |
| LSF swing the kiln tolerates | 1.5–2 over hours | Beyond this, defensive burning stops working and free lime, rings and quality excursions begin. |
| Silo blending ratio | 5–7× single | 7–12× for two silos; up to 15× for well-designed gravity systems. |
| Silo residence time | a few hours of feed | Sets the filter’s frequency. Ask for it — most plants have to work it out, which is itself informative. |
| Free lime | 0.5–1.5% | The consequence. Its deviation is the tell, not its mean. |
| Chain requirement | — | Quarry deviation ÷ (pile × mill × silo) < 1. All four numbers, or the requirement is unverifiable. |
§ 06 What goes wrong here
| The failure | The tell, and the usual misdiagnosis |
|---|---|
| Aeration patterns drift and nobody notices. Blowers degrade, sectors block, the sequence timing gets changed during a commissioning trial and never changed back. | The silo has no symptom of its own. Nothing alarms, nothing looks wrong, the material discharges. Usual misdiagnosis: kiln instability, treated at the kiln. The only readout is kiln-feed deviation, and it is often not measured. |
| Deviation measured at the mill outlet. | The wrong side of the silo. It cannot tell you what the silo is doing, by construction. This is the single most common measurement error in this part of the plant. |
| A slow drift diagnosed as a silo failure. | The silo is blamed for not filtering something no silo can filter. Look at the time series: fast means silo, slow means upstream. |
| Reporting a standard deviation and nothing else. | Hides the excursion structure entirely (§03). A month of harmless noise and a month of six-hour swings can report the same number. |
| Proposing a new silo. | Expensive, slow, and almost never the constraint (A2 §06). The ratios multiply, so a silo bought to compensate for a poor pile has to be extraordinary. Proposing it is a reliable way to be told the real problem was the reclaimer. |
| Ignoring the fuel side. | Higher alternative-fuel rates raise variability on the fuel side of the kiln, which consumes the same tolerance budget that feed variability does. A plant pushing TSR has less room for chemistry swing than it had before, and nobody rebalances the two. |
§ 07 Scenarios from the field
Under high clinker stock, Devapur had been cycling the raw mill on and off. Instead, it held the RM-3 separator fan at 90% speed and ran continuously at reduced rate.
Result: 0.141 kWh per tonne of cement, ₹17.3 lakh a year, zero investment. In the same year’s register, that single item outperformed six of the other eight projects combined.
Two things to take from it. The obvious one: an operating rule, changed once, with no capital. The less obvious one: stop-start milling is bad for chemistry as well as for power. Every restart is a transient in mill product composition, and transients are exactly what the silo has to absorb. The energy saving is the part that got measured; the stability benefit is real and was not counted. That is the normal shape of these items.
There is no material technology change in blending silos. Aeration and fluidisation retrofits are cheap and fit inside a normal stoppage; a new silo is a capital project that is almost never justified against tighter upstream control.
What is changing is the demand placed on the silo. Higher thermal substitution rates raise fuel-side variability (A6, and now a statutory obligation under F4 §12), and the kiln has one tolerance budget shared between fuel variation and feed variation. A plant that has taken TSR from 3% to 8% has quietly reduced the chemistry swing it can absorb — which raises the return on silo discipline without anything about the silo changing.
§ 08 The numbers that matter
| Metric | Reference | Basis | Your plant |
|---|---|---|---|
| LSF deviation, kiln feed | <1 | At kiln feed. Confirm where the sample is taken. | — |
| LSF deviation, mill outlet | — | Useful for diagnosing the pile and mill. Not a substitute for the kiln-feed figure. | — |
| Tolerable LSF swing | 1.5–2 over hours | Beyond this the kiln cannot burn its way out. | — |
| Silo blending ratio | 5–7× | Single continuous silo. Achieved, not rated — and the achieved figure needs measurement on both sides. | — |
| Silo residence time | hours of kiln feed | Live capacity ÷ kiln feed rate. Sets which frequencies the silo can filter. | — |
| Free lime deviation | — | The consequence. Track it alongside kiln-feed LSF; if one moves and the other does not, the problem is not chemistry. | — |
What is not published. There is no Indian benchmark for kiln-feed LSF deviation, and no published dataset of achieved silo blending ratios. The 5–7× and 7–12× figures are from the standard process literature and describe design performance, not measured performance. Treat any silo ratio as unverified until it has been sampled on both sides.
§ 09 Build the model — the variance budget
The same chain as A2, run backwards: given a kiln-feed target, what does each stage have to deliver — and which one is actually binding at this plant?
Why drift is added rather than divided. The silo divides variation it can mix across — variation faster than its residence time. Drift slower than the residence time passes through essentially unattenuated, so it is added to the output rather than reduced by the chain. That is the whole distinction of §03 expressed as arithmetic, and it is why a plant can meet its deviation target on paper and still have an unstable kiln.
§ 10 Upgrade paths
| Tier | Intervention | What it preserves, and the honest assessment |
|---|---|---|
| 0 | Move the routine sample to kiln feed | Free. Without it the silo is invisible and every argument about it is an opinion. This is the first recommendation in the lesson and it costs a sampling-point change. |
| 0 | Plot the time series, do not report the deviation | Free. Separates fast noise from slow drift, which determines whether the answer is the silo or upstream (§03). Two minutes of chart beats any summary statistic. |
| 0 | Verify the aeration sequence against its design | A control-system check. Patterns drift and timings get changed during trials and never changed back. Costs an afternoon. |
| 1 | Stop cycling the raw mill | Devapur: hold the fan at reduced speed instead of stopping and starting. ₹17.3 lakh a year at zero investment, plus a chemistry-stability benefit nobody counted. |
| 2 | Aeration and fluidisation retrofit | Blowers, pads, valves, sequence control. Cheap, fits inside a normal stoppage, preserves the silo structure entirely. The right answer when the silo genuinely is underperforming — which the Tier 0 work will have established. |
| 4 | New blending silo | Tens of crores for a ratio of 5–7 when the pile can usually be moved for the price of a shift instruction. Almost never justified against tighter upstream control, and proposing it before the Tier 0 work is how you find out the real problem was the reclaimer. |
§ 11 The frontier, and who is running it
There is no frontier here, and the shape of the absence is the finding. Every published before-and-after number in raw-mix chemistry control was measured upstream of the blending silo — at the crusher, at the stockpile, or at the raw mill outlet. Two decades of disclosed evidence, and the silo itself is still a black box. No named plant anywhere has an online analyser on the kiln feed after the silo.
| Status | What | Plant, company, country | The number |
|---|---|---|---|
| OPERATING | The best disclosed chain, closed loop | Fateh Jang — Fauji Cement, Pakistan. SpectraFlow cross-belt analyser after the crusher, ahead of a 55,000 t circular stockpile. Commissioned October 2011. | Hourly raw mill LSF standard deviation 4.385 → 2.695 → 1.503, with validation piles at 1.32 and 0.97. Closed loop on both the crusher apron feeders and the raw mill LSF setpoint. Measured at the raw mill, i.e. the silo inlet. |
| OPERATING | The longest disclosed series | Lappeenranta — Finnsementti, Finland. PGNAA analyser before the raw mill, 2015; blending control software upgraded 2016. | Raw meal LSF SD 6.9 (2011) → 5.3 (2015) → 1.9 (2016), with cement strength SD improving across four CEM types. Two caveats travel with it: the hardware and software changes are confounded, so neither can be isolated; and again it is measured after the mill, not at kiln feed. |
| ORDERED | Analyser on the airslide | Bowmanville — St Marys Cement (Votorantim), Ontario, Canada. SpectraFlow airslide analyser after the raw mill, 450 t/h. Reported early 2025. | One result per minute against the previous ten. No performance target disclosed. Note the position: after the mill, before the silo. Even the newest installation does not look at kiln feed. |
| OPERATING | A large modern blending silo | Ragland — National Cement Company of Alabama (Vicat), USA. thyssenkrupp Polysius inverted-cone raw meal silo, 78 m, six roof inlets, eight cone outlets, on a 5,000 t/d line. Line inaugurated July 2022. | No blending factor, no kWh/t, no inlet or outlet standard deviation. The only energy statement published is “a lower amount of air at lower pressure”. A brand-new silo on a brand-new line and not one performance number. |
| ANONYMOUS | The only numeric blending factor in open literature | FLSmidth CF silo technical brochure. No plant, no date, no shown calculation. | States H of 5:1 to 10:1 is “usually required” to hold kiln feed inhomogeneity below 1% LSF, and gives a worked example of H = 8.3 “from a working CF silo”. IBAU Hamburg, Claudius Peters and Polysius publish no blending factor at all. |
| LAB ONLY | A result worth knowing and not citing as a benchmark | Nanjing Tech University, published June 2024. Scaled equipment models, magnetite tracer, raw meal from an unnamed Zhejiang plant. | Homogenisation coefficients: pneumatic silo >6, ball mill ~5, five-stage preheater ~4, rotary kiln ~2. If that holds at scale, the mill and the preheater each blend nearly as much as the silo does — which would make a silo-centred diagnosis mis-aimed. The authors state the models cannot fully simulate industrial equipment. Treat it as a hypothesis worth testing on a client, not a number to quote. |
| FAILURE | What the downside looks like | Fateh Jang — Fauji Cement, Pakistan, 29 May 2016. Raw meal silo holding about 25,000 t collapsed, damaging the coal mill area. No casualties. | The 7,200 t/d Line 2 was out for an expected five to six months. Damage estimates in the trade press ranged from PKR 2bn to 20bn — a tenfold spread, which tells you nothing was actually disclosed. The company never published a repair cost or a root cause. This remains the most recent documented cement raw meal silo collapse. |
No plant anywhere publishes a measured blending factor for its own kiln feed silo. No plant publishes a paired inlet and outlet standard deviation across the silo. No named plant has an online analyser on the kiln feed after the silo. No benchmarking body — CII, VDZ, IFC — publishes a silo aeration kWh/t line at all; CII’s nearest entry bundles it into raw mill “auxiliaries” at 1.2–4.5 kWh/t. No plant publishes a measured kiln feed weighing accuracy, and no vendor publishes short-term and long-term accuracy separately — which is exactly the split that matters, because drift, not noise, is the failure mode on kiln feed. No AI or advanced-control installation anywhere closes a loop on silo extraction or on kiln feed chemistry.
Aeration power for blending, from three independent sources that agree with each other and are all design figures rather than plant measurements: air-merge batch blending 1.5–2.5 kWh/t raw meal; controlled-flow inverted cone 0.25–0.5; multi-outlet controlled flow 0.10–0.13. A spread of roughly twenty to one between the oldest and newest concepts.
That spread is the usable teaching point, and it is also the trap. It is a technology-generation difference, not an operating gap. A plant with an air-merge silo cannot operate its way to 0.12 kWh/t; it can only replace the silo, which nobody does for an aeration saving. Use the range to place a client’s silo in its generation, then stop — the recommendation that follows is about blending performance, not about the blower.
§ 12 What’s changing now
Current as of August 2026 · refresh every six monthsTSR became a legal obligation under the Solid Waste Management Rules rather than a voluntary target (F4 §12). Alternative fuels vary far more than coal in calorific value, moisture and ash chemistry.
What it means: the kiln has one tolerance budget, shared between fuel variation and feed variation. Raising TSR spends part of it on the fuel side, which leaves less for chemistry. A silo that was adequate at 3% TSR may not be at 10%, without anything about the silo changing. Nobody rebalances the two, because they are owned by different people.Blending silo technology has not moved materially. What moves is whether the aeration is doing what it was designed to do, and whether anyone is measuring on the right side of it.
What it means: be sceptical of any proposal in this section that is mostly capital. The value here is in measurement and operating discipline, and a vendor arriving with a silo answer has probably not been shown the kiln-feed time series — because it usually does not exist.§ 13 Check yourself
§ 14 Mini case — the deviation that met target and the kiln that did not
A 1.9 Mt/yr plant reported kiln-feed LSF standard deviation of 0.94 — inside the target of 1.0 — and had reported something similar for two years. It also had free lime averaging 1.7%, refractory life of nine months, and a fuel rate about 20 kcal/kg above where the same line had run in 2021. The plant’s conclusion was that the kiln needed a burner upgrade.
The chart nobody had drawn. Asked for the hourly LSF series rather than the summary, the plant produced a fortnight of data. Plotted, it showed a clean sawtooth: LSF sat around +1.8 for roughly seven hours, then around −1.6 for the next seven, with very little fast noise on top. Standard deviation 0.94, exactly as reported. And a fourteen-hour cycle the kiln had no chance of absorbing.
Where it came from. The plant ran two limestone faces, alternating by shift because one had better haul-road access in the mornings. The pre-blend pile was short and reclaimed at high rate, so it filtered minutes rather than hours. The silo’s residence time was about four hours — less than half the drift period, so it passed the sawtooth through almost untouched, exactly as §03 predicts.
What was actually wrong. Nothing in the kiln, nothing in the silo, and nothing that a standard deviation could reveal. A quarry scheduling decision taken for haul-road reasons was arriving at the burner fourteen hours later.
What was done. The two faces were blended at the crusher by alternating within the shift rather than between shifts, which cut the drift period to under an hour — well inside the silo’s filtering range. Free lime deviation halved within a month. The burner upgrade was withdrawn.
A composite scenario. The silo ratios, the residence-time argument and the tolerance figures are from the process literature; the plant is not a real one.
The transferable move. Never accept a standard deviation for a time-varying process without seeing the time series. The summary statistic is blind to timescale, and timescale is the only thing that determines whether the plant’s existing equipment can help. Two minutes of chart, on day one.
§ 15 Go deeper
§ 16 Carry forward
- “Where is that LSF deviation measured? If it’s the mill outlet, it’s on the wrong side of the silo and it can’t tell us what the silo is doing.”
- “Don’t give me the standard deviation. Give me the hourly series for a fortnight, plotted. The deviation is blind to timescale and timescale is the whole question.”
- “A silo is a filter with a frequency. It removes variation faster than its residence time and passes everything slower straight through. So if the drift is slow, no silo helps and the answer is upstream.”
- “Your kiln can absorb about one and a half to two points of LSF swing by burning defensively — and you pay for that defence every hour, not just the bad ones.”
- “You’ve taken TSR from three percent to ten. That spends part of the same tolerance budget your feed chemistry was using. Has anyone rebalanced the two?”