Designing the raw mix
Four numbers decide what the clinker will be, how hard the kiln has to work to make it, and what the cement will be worth. They are set by whoever owns the additive budget, defended by whoever owns the fuel budget, and revisited by almost nobody.
On this page · 16 sections
- 01The one idea
- 02The parts — four materials, four numbers
- 03First principles — LSF is a stoichiometric ratio, not an index
- 04What must be true
- 05The quantitative anatomy — a worked mix
- 06What goes wrong here — the burnability trap
- 07Scenarios from the field
- 08The numbers that matter
- 09Build the model — a four-component raw-mix solver
- 10Upgrade paths
- 11The frontier, and who is running it
- 12What’s changing now
- 13Check yourself
- 14Mini case — the strength complaint that cost three years of fuel
- 15Go deeper
- 16Carry forward
§ 01 The one idea
Raw-mix design is a cost decision disguised as a chemistry decision. Every one of its targets can be met more than one way, and the ways differ in what they cost and who pays.
Raising the lime saturation factor makes stronger cement, which the sales team wants. It also makes a mix that is harder to burn, which the kiln pays for in fuel and in refractory life. Both consequences are real; only one of them appears in the decision. And because the raw-mix targets are usually set once, at commissioning or after a quality complaint, the cost keeps being paid for decades by people who were not in the room.
“When was your LSF target last changed, and what was the reason?” If the answer is a quality problem — a strength complaint, a failed batch, a new customer specification — ask what the fuel rate did in the following quarter. Very often nobody looked, because the two numbers live in different reports.
§ 02 The parts — four materials, four numbers
A raw mix is normally built from four components, and the fourth is often optional.
| Component | Typical share | What it supplies |
|---|---|---|
| Limestone | 85–93% | Lime. Everything else is a correction to what the limestone lacks. This is why A1 governs this lesson. |
| Clay or shale | 4–12% | Silica and alumina together. The workhorse corrective, and the cheapest if there is a deposit nearby. |
| Iron source laterite, iron ore, mill scale, red mud | 0.5–3% | Iron oxide, which fluxes the melt. Small quantity, large effect on burnability. |
| Silica source sand, quartzite | 0–4% | Silica alone, when the clay does not supply enough. The component most likely to cause a burnability problem — see §06. |
| Sweetener | variable | High-grade limestone bought in to lift the lime when the quarry's grade falls. Pure operating cost, and the most direct link between A1 and this lesson. |
Those four proportions are chosen to hit three ratios and one constraint:
| Ratio | Formula | Target | What it controls |
|---|---|---|---|
| LSF lime saturation factor | 100 · CaO ⁄ (2.8 SiO₂ + 1.18 Al₂O₃ + 0.65 Fe₂O₃) | 92–98 | Strength, and how hard the kiln has to burn. The master variable. |
| SM silica modulus | SiO₂ ⁄ (Al₂O₃ + Fe₂O₃) | 2.3–2.7 | How much liquid forms in the burning zone. High SM means less melt, harder burning, thin coating. |
| AM alumina modulus | Al₂O₃ ⁄ Fe₂O₃ | 1.3–1.6 | How viscous the melt is, and therefore how the coating behaves and how fast the clinker sets in the cooler. |
| MgO, alkalis | — | MgO <6% R₂O low | Hard constraints from A1 §03. Not design variables — you inherit them from the deposit. |
§ 03 First principles — LSF is a stoichiometric ratio, not an index
LSF looks like an arbitrary weighted formula. It is not. It is the ratio of the lime you have to the lime the other oxides can chemically combine with, and each coefficient is a molar ratio in disguise.
Silica wants to end up as tricalcium silicate, C₃S — three CaO for every SiO₂. Three times the molar mass of CaO (56.08) divided by the molar mass of SiO₂ (60.08) is 2.80. That is the first coefficient, and it is simply the mass of lime needed to saturate a unit mass of silica. The alumina and iron coefficients come out of the same arithmetic for C₃A and CᾔAF.
An LSF above 100 is not a hard target — it is an impossible one. Free lime is guaranteed by stoichiometry, and no amount of fuel, residence time or burner tuning removes it. A plant fighting persistent free lime at LSF 100+ is fighting arithmetic.
Below 100, free lime is a kinetic problem rather than a stoichiometric one — the lime can combine, and whether it does depends on temperature, time and how finely the silica is divided. That distinction is the whole of §06, and it is the first thing to establish when a plant reports a free-lime problem: is this chemistry or is this burning?
And what the LSF choice does to the product
§ 04 What must be true
Raw-mix design is where the clinker is decided, so its conditions are stoichiometric rather than mechanical. Five of them, none of which depends on the equipment installed.
| What must be true | Why | Target | How you verify it |
|---|---|---|---|
| 1. There is enough lime, and not more than enough | LSF is the ratio of lime present to lime the acidic oxides can chemically combine with. Above 100 the surplus has nowhere to go. | LSF 92–98 | The hourly LSF series at kiln feed. Above 100, free lime is arithmetic and no fuel rate removes it — establish which regime the plant is in before diagnosing anything else. |
| 2. There is enough melt to form alite | C₃S does not form by solid-state reaction at any useful rate. It forms through a liquid phase whose quantity is fixed by the silica modulus and whose viscosity is fixed by the alumina modulus. | SM 2.3–2.7 AM 1.3–1.6 |
Both moduli on the clinker, not the raw meal. High SM shows up as thin coating and short refractory life; low SM as rings. |
| 3. The silica is burnable, not merely present | Burnout is diffusion-limited. Finely disseminated silica in clay reacts; coarse quartz above about 45 microns has a much longer path into the melt. | quartz <45 µm | A petrographic or sieve analysis, which most plants have never run. A grade table shows how much silica there is and says nothing about whether it can be burned. |
| 4. The target is held to a deviation, not a mean | The kiln burns defensively against the worst hour. A mix on target with a wide spread costs fuel every hour of the month. | LSF SD <1 at kiln feed |
Standard deviation at kiln feed, which is after the silo, not before it (A5 §02). And the hourly series plotted, because a deviation is blind to timescale. |
| 5. The corrective system has range left | Four ratios ride on about 8% of the mix. A feeder running at the top of its range has no capacity to absorb the next change in the deposit. | feeders mid-range | Feeder set points against their design range, and the limestone analysis the system was sized on. Holding the window by running the feeders flat out works until it does not. |
Rows 1 and 2 are chemistry and can be checked from a lab sheet in minutes. Row 3 needs a measurement almost nobody has. Rows 4 and 5 are the ones that decide whether today’s performance survives next year’s deposit.
If a plant reports persistent free lime, work them in order: is the mix saturated (row 1), is there melt (row 2), is the silica burnable (row 3)? Only after all three come back clean is it an operating problem, and only then is the kiln the right place to look.
§ 05 The quantitative anatomy — a worked mix
A realistic Indian four-component mix, and what it produces. The point of showing it in full is that none of this requires software — it is four multiplications and three divisions, and you can do it in front of a client.
| Component | % | CaO | SiO₂ | Al₂O₃ | Fe₂O₃ | LOI |
|---|---|---|---|---|---|---|
| Limestone | 92.0 | 46.0 | 10.0 | 2.5 | 1.2 | 37.5 |
| Clay | 4.6 | 2.0 | 58.0 | 18.0 | 8.0 | 10.0 |
| Laterite | 1.2 | 0.5 | 12.0 | 12.0 | 62.0 | 10.0 |
| Sand | 2.2 | 0.3 | 92.0 | 3.0 | 1.5 | 1.5 |
| Clinker, after loss on ignition | — | 65.4 | 21.6 | 5.1 | 3.5 | — |
Run Bogue on that clinker and you get C₃S 62, C₂S 15, C₃A 8, CᾔAF 11 — a conventional Portland clinker. Notice how little of the mix is anything other than limestone: 8% of the raw mix carries all four control ratios. That is why small changes in the correctives move the chemistry so much, and why an error in the additive feeders shows up as a chemistry excursion within hours.
The Bogue equations assume the clinker reached equilibrium, that the four phases are pure, and that no minor components interfere. None of those is exactly true. Real clinker analysed by X-ray diffraction typically shows C₃S a few points different from Bogue and often higher.
Bogue survives because it is computable from an oxide analysis every plant already has, hourly, for free. Use it for direction and comparison, never as a measured phase composition — and if a plant is making a decision that turns on two or three points of C₃S, that is a decision that needs XRD, not arithmetic.
§ 06 What goes wrong here — the burnability trap
One failure dominates this lesson and it is worth stating as a sequence, because that is how it happens.
One. A strength complaint, or a push into a premium product, leads to the LSF target being raised — say from 94 to 97. The reasoning is sound: more C₃S, more early strength.
Two. The mix is now closer to lime saturation, so more of the silica has to be fully converted. Conversion happens by solid-state diffusion into the melt, and it is slow.
Three. If the silica arrives as coarse quartz — from sand rather than clay, or from a quartz-veined part of the deposit — the diffusion distance is much larger. Coarse quartz above about 45 microns is disproportionately hard to burn. The combination of high LSF and coarse silica is the classic burnability trap.
Four. Free lime runs marginal, so the kiln burns harder. Heat rate rises. Burning-zone temperature rises. Refractory life falls. NOx rises.
Five. None of it is attributed to the mix decision, because the decision was three years ago and the fuel rate is reported by a different function. The plant now believes it has a kiln problem.
| Other failures | The tell |
|---|---|
| LSF controlled to a target but not to a deviation. | The mean is on target and the hourly spread is wide. F6 §03’s argument applies exactly: the fuel rate is set by the bad hour. Ask for the standard deviation, not the average. |
| Chemistry corrected at the mill instead of at the pile. | Feeder trims chasing an analyser signal, hour by hour. It works, and it means the pre-blend pile is not doing its job (A2). Fixing it upstream is cheaper and more stable. |
| The additive is chosen on delivered price alone. | A cheaper iron source with more alkali or more moisture can cost more in the kiln than it saves at the gate. The additive decision belongs to whoever owns the fuel line, and it usually does not. |
| SM raised to save on iron and alumina correctives. | Less melt, harder burning, thinner coating, shorter refractory life. It is a real saving on the additive line and a larger loss on two others. |
| Nobody has checked the quartz size distribution. | A grade table shows SiO₂ percentage. It says nothing about whether that silica is finely disseminated in clay or present as 100-micron quartz grains. Only a petrographic or sieve analysis distinguishes them, and the difference is worth more than most projects in this chapter. |
§ 07 Scenarios from the field
Orient applied machine learning to raw-mix optimisation across 44 process parameters and 91 quality-control parameters. Result: 1,314 Gcal a year, for ₹6.33 lakh. A two-month payback.
Two things worth noticing. First, the money: ₹6.33 lakh is a rounding error against any capital project in this course, and it is on the raw mix rather than the kiln — where almost every AI proposal in cement is aimed. Second, why it worked at all. Raw-mix optimisation is a genuinely high-dimensional problem with a fast, cheap, abundant measurement (hourly XRF) and a controllable actuator (feeder ratios). That is close to the ideal setting for machine learning, and it is why this project cost ₹6 lakh while kiln-optimisation projects cost crores and disappoint.
The consultant’s read: when a client asks where to start with AI in a cement plant, this is the honest answer, and it is not the answer the vendor will give.
A cross-belt online elemental analyser on the crusher or mill feed measures composition continuously rather than hourly. CII puts its direct energy effect at −3 to −5 kcal/kg.
That number understates it, and writing the business case on it is a mistake. The analyser’s value is not the energy it saves directly; it is that it is the enabling instrument for tighter LSF control, for stacking against quarry variability (A2 §10), and for any control scheme of the Chittapur kind. Sold on −4 kcal/kg it looks marginal. Sold as the measurement that unlocks a class of interventions, it is one of the better-evidenced retrofits in the value chain. Frame it as instrumentation, not as an energy project.
§ 08 The numbers that matter
| Metric | Target | Note | Your plant |
|---|---|---|---|
| LSF | 92–98 | Above 100 guarantees free lime by stoichiometry. | — |
| LSF standard deviation, kiln feed | <1 | The number that actually matters. Measured after the silo, not before (A5). | — |
| Silica modulus | 2.3–2.7 | Higher means less melt and harder burning. | — |
| Alumina modulus | 1.3–1.6 | Sets melt viscosity and coating behaviour. | — |
| Free lime | 0.5–1.5% | The kiln’s report card, measured hourly. | — |
| Coarse quartz | >45 µm | Disproportionately hard to burn. Ask whether anyone has measured it. | — |
| Raw meal fineness | 12–15% R90 | Residue on a 90-micron sieve. Finer costs mill power and buys burnability — a direct A4 trade. | — |
| Corrective share of mix | ~8% | All four control ratios ride on this fraction, which is why feeder accuracy matters so much. | — |
| Cross-belt analyser | −3 to −5 kcal/kg | CII. Understates it — the value is control, not energy. | — |
§ 09 Build the model — a four-component raw-mix solver
Change the proportions and watch the moduli and the clinker phases move. The oxide analyses are typical Indian values and can be overwritten from the client’s own lab sheet.
Try this. Drop the limestone CaO from 46 to 43 — a grade decline of the kind A1 §09 models — and watch LSF fall out of window. Then restore it by raising the limestone proportion, and notice what that does to the silica modulus. Every correction has a second effect, and the reason raw-mix design is done with a solver rather than by hand is that the four ratios are not independent.
§ 10 Upgrade paths
| Tier | Intervention | What it preserves, and the honest assessment |
|---|---|---|
| 0 | Re-examine the LSF target | Free. Ask when it was set, why, and what the fuel rate did afterwards. If it was raised for a quality reason that no longer applies — a customer lost, a product discontinued — it may simply be wrong. This is the highest-value item in the lesson and it costs a meeting. |
| 0 | Measure the quartz size distribution | A petrographic or sieve analysis. Establishes whether a burnability problem is chemistry or physics, which determines whether the answer is the mix or the mill. |
| 1 | Control LSF to a deviation, not a mean | A reporting and control-loop change. Preserves everything. Pairs directly with A2 — the pile has to deliver the raw material for this to be achievable at all. |
| 1 | By-product correctives | LD slag, red mud, mill scale, marble slurry, fly ash. Cheaper than mined additives, and slag in particular arrives decarbonated (A1 §07). Constrained by supply, transport and alkali content — not by technology. |
| 2 | Cross-belt online analyser | The standard retrofit here. Preserves the whole circuit; installs on a belt. Write the case on control capability, not on CII’s −3 to −5 kcal/kg, or it will look marginal and lose. |
| 2 | Model-based raw-mix optimisation | Chittapur: ₹6.33 lakh, two-month payback. Needs the analyser or reliable hourly XRF, and needs feeder control that responds. The best-value software project in the plant, and the least glamorous. |
| 3 | Additional corrective storage and feeders | Civil and mechanical work in a shutdown. Justified when the quarry has changed enough that the original two-corrective design cannot hold the window — which is increasingly common as A1 §12’s newly auctioned blocks come on. |
§ 11 The frontier, and who is running it
Raw-mix control is the one place in the plant where artificial intelligence has produced results that stand up, and the reason is in §06: a cheap abundant measurement and a fast actuator. The frontier here is closed loop — software that writes setpoints rather than suggesting them.
| Status | What | Plant, company, country | The number |
|---|---|---|---|
| OPERATING | Fully closed-loop plant control | Roanoke — Titan America, Virginia, USA, with Optimitive. Raw mill, kiln and finish mills. | AI applying setpoints in closed loop, fully operated by AI across three circuits. The company discloses only “returns of investment in less than a year in most cases”. No performance number — which is itself the thing to notice. |
| OPERATING | Closed-loop raw mill | Barcelona — Cementos Molins, Spain, with Optimitive. Six manipulated variables on an FLSmidth DCS. | From 340 t/h at 3,600 kW: +14.6 t/h throughput and 5–10% off kWh/t, fineness up 5.59%. In service since 2017. Implies about 10.6 kWh/t at the mill motor. Vendor-disclosed. |
| OPERATING | AI quality prediction at scale | alcemy — live at 45 cement plants in 18 countries. Named customers: Amrize, Buzzi Unicem, Cimpor, CRH, Molins, Titan, Votorantim. | More than a third already run it in closed loop. Cement reduction in concrete of 5–30 kg/m³. July 2026. No individual plant is named, which limits what you can verify — but the installed base is large enough that a client’s group may already have it somewhere. |
| OPERATING | Clinker strength prediction | Anhui Conch, China — company-level, no plant named. | 28-day strength predicted to within 1 MPa at better than 85% accuracy, with 1% kiln fuel reduction claimed alongside. |
| NUMBER WITHOUT A RESULT | Group-wide AI deployment | Holcim with C3 AI — 45 plants live, target over 100 within four years, 40+ countries. | No plants named and no quantified result disclosed, anywhere. This figure is repeated constantly. Use it as an example of a number that carries no verifiable performance claim — the deployment count is real, the benefit is unevidenced. |
Two gaps run through every entry above. Nobody discloses closed-loop uptime — what fraction of mill or kiln hours the system actually holds control. A system running 30% of the time delivers 30% of the benefit, and that is the number that decides the business case. Ask for it in writing.
And no named cement plant has published a case for routine XRD control rather than XRF. The technique is clearly in use — Rietveld phase analysis is standard laboratory practice — but nobody has published a plant-level control benefit. If a decision turns on two or three points of C₃S (§05 warning on Bogue), that is the measurement you need and the evidence base for it is thinner than the vendor literature suggests.
§ 12 What’s changing now
Current as of August 2026 · refresh every six monthsAs LC³ moves into commercial production, kaolinitic clay becomes something a plant sources, calcines and stores — which puts it in this lesson as well as in A9. India has had the standard for three years.
What it means: anyone still saying “we’re waiting for the BIS standard” is three years out of date, and that sentence is a useful test of how current a client’s technical view is. The real constraints are clay quality, calcination energy and market acceptance — not the standard.The deposits coming on stream are lower grade and more variable (A1 §03). A corrective system designed for a 46% CaO limestone has less headroom at 43%.
What it means: ask when the corrective system was last resized, against which limestone analysis. Many were designed at commissioning and have never been revisited, and the plant is holding the window by running the feeders at the top of their range — which works until it does not.§ 13 Check yourself
§ 14 Mini case — the strength complaint that cost three years of fuel
A 1.4 Mt/yr plant lost a large infrastructure customer in 2022 over 3-day strength. The technical response was correct and quick: LSF target raised from 94 to 97, sand addition increased to hold the silica modulus, and the product met specification within a month. The customer came back.
What also happened. Heat rate went from 728 to 751 kcal/kg over the following two quarters. It was attributed to a coal-quality change, which had also occurred, and the attribution was never tested. Refractory life in the burning zone fell from 11 months to 8. Free lime deviation widened.
What the review found three years later. The sand used to hold the silica modulus was a local quartzite with a coarse grind — a large fraction above 45 microns. High LSF plus coarse quartz is the burnability trap of §06, arrived at in two separate and individually sensible decisions. Neither decision was wrong. Nobody had ever looked at them together, because the mix belonged to quality and the fuel rate belonged to production.
The arithmetic. 23 kcal/kg at ₹1.90 per ’000 kcal and a 0.68 clinker factor is about ₹30 per tonne of cement — roughly ₹4.2 crore a year on 1.4 Mt, and it had been running for three years. Against that, the customer was worth having.
What was actually done. Not a reversal — the strength requirement was real. The sand was replaced with a finer-ground silica source and the raw meal fineness tightened by two points of R90, at a cost in raw-mill power of about 1.1 kWh/t. Heat rate recovered to 736. The trade was made deliberately, in one place, for the first time.
A composite scenario. The ₹1.90 conversion, the 0.68 clinker factor and the 45-micron quartz threshold are sourced; the plant is not a real one.
The transferable move. When a plant tells you its heat rate changed, ask what changed in the raw mix in the preceding two quarters. The two numbers live in different reports and are owned by different functions, and the causal link runs in a direction nobody is looking.
§ 15 Go deeper
§ 16 Carry forward
- “LSF is a stoichiometric ratio, not an index. It’s the lime you have over the lime the silica, alumina and iron can chemically take. Above a hundred, free lime is arithmetic — no amount of burning removes it.”
- “When was the LSF target last changed, and what did the fuel rate do in the two quarters afterwards? Those numbers live in different reports.”
- “Six points of LSF is about fifteen points of C₃S, and every one of them is bought with fuel and refractory life. That’s a commercial trade, not a technical one.”
- “Has anyone measured the quartz size distribution? Your grade table tells me how much silica you have, not whether it’s burnable.”
- “If you want to start with AI, start on the raw mix, not the kiln. Orient did it for six lakh with a two-month payback, because that’s where the measurement is cheap and the actuator responds.”