Every cement plant pushing for a higher thermal substitution rate (TSR) meets the same problem. The fuels that cut cost and carbon intensity are the hardest to burn consistently. Refuse-derived fuel (RDF) arrives at different moisture levels from one truck to the next, tire-derived fuel carries variable sulfur, and biomass shifts in calorific value with the seasons. Each load adds uncertainty to a process that rewards stability.
Cement is one of the hardest sectors to decarbonize, and IEA tracking shows fossil fuels still supply most of the thermal energy in kilns worldwide. Energy is also one of the largest cost categories in cement production, which makes fuel substitution a natural target. Each point of TSR that stalls below what the kiln could take is margin the plant does not capture.
Moving from typical substitution rates to the rates some producers already run is a control problem. At high substitution, fuel variability becomes the normal operating condition.
Why fuel variability compounds as substitution rises
RDF does not burn like coal
At higher TSR, small variations in moisture, calorific value and ash chemistry compound, because the kiln has less margin to absorb them. Particles of RDF or solid recovered fuel (SRF) leave the burner tip with different densities, shapes and chemistry than pulverized coal, and their moisture and energy density change from load to load in ways coal rarely does.
In winter, RDF moisture can exceed 20%, and some streams exceed 30%. That moisture cools the flame and produces weakly burned clinker with elevated free lime, a known path to quality problems in finished cement. Summer fuel is usually drier and easier to manage, although dust and handling problems can increase. The more useful measure is short-cycle variability. How fast moisture and calorific value change between loads sets how quickly the kiln's thermal profile has to be detected and corrected before quality drifts.
Volumetric dosing
Most fuel handling systems dose by volume. When calorific value changes between loads, the heat delivered to the burner changes even though the dosing rate holds constant.
Flame physics above 50% substitution
Above roughly 50% thermal substitution, the physics change. Higher fuel volume and lower calorific density reduce burner momentum, and the flame becomes long, soft and hard to shape. Secondary air cannot reach all of the fuel, partly burned material lands on the clinker bed, and reducing conditions begin to damage clinker quality and refractory brick.
Several European markets already run substitution rates well above the global average, while alternative fuels still supply a small share of kiln thermal energy worldwide. At those rates, kiln operation moves from manageable with careful setpoints to routine operation at the edge of the envelope.
The quality chain from burning zone to finished cement
Variable fuels destabilize the burning zone, and the effect travels through a causal chain that ends with the customer. Good clinker needs a stable burning zone temperature, generally reported in the 1,400 to 1,450°C range.
Deviations from that range impair the melt phase in which belite (C₂S) and free lime convert to alite (C₃S), the main early-strength phase. With too little melt, free lime stays above the 1.5% threshold many producers target, and alite content drops.
Alternative fuels add a second pathway. Alkali in their ash, potassium oxide in particular, affects clinker mineralogy and can interfere with the reactions that produce strong clinker. Clinker chemistry is pressed from two directions at once. Thermal instability starves the melt phase, and alkali input disrupts the reactions that depend on it. Free lime can also affect grindability, although that relationship depends on burning conditions and mineralogy.
For most plants the binding constraint is the free lime measurement itself, which comes from the laboratory with about a 90-minute lag. By the time the result confirms a problem, the kiln has produced hundreds of tonnes under the wrong conditions, and any correction is retrospective. AI-optimized pyroprocessing targets that information gap.
Where conventional kiln control reaches its limits
Models and rules that cannot follow the fuel
Advanced process control (APC) is widely used in cement, and its process models are developed and calibrated at installation. As process behavior diverges from those models, performance degrades. At high substitution the divergence is continuous. Each material change in the fuel mix shifts process behavior, and new fuel sources often lack the well-defined first-principles relationships conventional control depends on.
Rule-based automation has a different limit. When fuel energy content departs from the encoded assumptions, fixed actions under- or over-respond, and the rules cannot detect the mismatch.
Thermal lag
The long thermal time constants of a rotary kiln make both limits worse. An undetected shift in fuel calorific value works through the kiln's thermal mass for the full dead time before any sensor registers it. By then the kiln has already made off-spec clinker.
Operator turnover
As experienced operators retire, their knowledge of how a specific kiln responds to specific fuel combinations leaves with them. Plants then default to conservative setpoints that leave substitution potential unused. Capturing those patterns is a planning priority, because the window closes when those operators leave.
What learned models change under variable fuels
Moving setpoints toward the objective
APC holds variables at defined setpoints. A learned optimizer sits above it and moves those setpoints, using real-time process data, toward higher-order objectives such as minimizing specific heat consumption or fuel cost. Under variable fuels that matters, because the best setpoint combination shifts with every material change in the blend.
Predicting free lime ahead of the lab
For plants raising substitution as part of decarbonization, the central capability is real-time quality prediction. A model trained on the plant's own process history can estimate current free lime ahead of laboratory confirmation. Operators see the predicted value beside the measured one and adjust setpoints before off-spec clinker accumulates.
Treating the fuel mix as a continuous decision
Blend selection becomes a problem that is re-solved continuously instead of a fixed recipe. The model learns how combustion air ratios, the kiln thermal profile and precalciner temperature affect clinker quality under different fuel combinations, and its recommendations follow the fuel as its properties shift.
One view for operations, quality, maintenance and planning
A shared process model gives operations, quality, maintenance and planning the same predicted outcomes. Those teams often decide in isolation, each managing its own constraint without seeing how the others' choices interact.
What to watch as substitution rises
- Free lime. Many producers target levels below about 1.5% to avoid downstream cement quality problems.
- Burning zone temperature stability. Thermal swings affect clinker formation directly.
- Liter weight. A faster proxy for burn quality between lab cycles.
- Reducing conditions. Incomplete combustion of alternative fuel can signal drift before the laboratory confirms it.
From advisory mode to closed loop
Kiln deployments that last tend to start in advisory mode. The model recommends setpoint changes, and operators decide whether to accept them. A model does not hold all the pattern recognition experienced operators build over decades, and edge cases in fuel chemistry still need human judgment. Human and AI collaboration is how that knowledge gets captured before the people who hold it retire.
Advisory mode also changes how experienced and newer operators work together, because both evaluate recommendations against the same current picture of the process. As confidence builds, plants move to closed loop. The learned controller writes setpoints through the existing DCS and APC, which keep handling stabilization, so earlier investment in operational efficiency carries forward and operators can take over from the board.
How the Imubit Platform applies
Imubit applies Map, Model and Control to cement kilns. Map traces lost substitution and clinker quality to the recurring decisions behind them, such as fuel mix and combustion air. Model learns the kiln's nonlinear dynamics from its own history with Deep Learning Process Control® (DLPC®), including free lime predictions ahead of the lab. Control takes those decisions to closed loop on top of the existing DCS and APC. On one customer's kiln pyroprocess running in closed loop, the result was a 20 to 30% reduction in process variability.
