Every cracking furnace loses efficiency across a run. Convection section fouling reduces heat recovery, coke insulates the radiant coils, and firing rates climb to compensate. The IEA's analysis of naphtha steam cracking shows a wide spread in energy intensity per tonne of high-value chemicals across the installed base. Much of the gap between average and top-performing furnaces is tied to operating variables that the plant controls.
In olefins plant optimization, the hard part is that the trade-off between severity, run length and energy keeps shifting within the same run. Each variable pulls on the others. Severity decisions affect coking rates, coking rates affect run length, and run length decides whether a throughput gain shows up in margin. The furnace at hour 400 of a run is not the furnace the run started with.
In brief
- Higher coil outlet temperature improves ethylene yield but accelerates coke deposition and shortens the run.
- One fast-coking pass can force the whole furnace offline before the other passes reach their practical limit.
- Added severity improves campaign throughput when the yield gain outweighs the production lost to earlier decoking.
- A sustained rise of 50°C above design stack temperature signals significant lost heat recovery in the convection section.
- Convection fouling raises firing rates, stresses the radiant section and turns into a furnace-wide throughput and energy constraint.
The severity, coking and run length trade-off
Coil outlet temperature (COT) is the main severity lever in a steam cracking furnace. Higher COT produces more ethylene, and it also accelerates coke deposition on the tube walls. Coke acts as thermal insulation and forces higher firing to hold temperature, a dynamic that compounds across the campaign and drives up energy cost.
As coke builds, tube skin temperature rises under the same firebox conditions. When tube metal temperature approaches its metallurgical limit, the furnace must come offline for decoking. The economic window sits between acceptable severity and the point where coking ends the run.
Each decoke costs production during the downtime. It also adds energy from thermal cycling and contributes to cumulative thermal fatigue in the tube materials. More severity improves near-term yield, and it pays when the run length penalty does not erase that value.
Pass balancing and early decoking
Pass balancing is the most underappreciated lever in that equation. If one pass cokes faster than the others, the whole furnace must decoke early, and the production the other passes could still have made is forfeited.
Most operations teams try to equalize coking rates so that the passes reach their limits together. It is one of the hardest objectives to hold manually, because the drivers of differential coking, including feedstock variability, flow maldistribution and tube condition, shift continuously during the run.
Severity and the product slate
Severity affects more than ethylene. Changing COT shifts the propylene-to-ethylene ratio and alters the value mix leaving the quench and fractionation sections. An operator who raises severity to chase ethylene can give back margin on propylene or butadiene without seeing it until the accounts close.
Furnace efficiency is therefore a question of the total value the furnace produces per run, of which energy is one part.
Convection section fouling and the efficiency cascade
The convection section recovers heat from the flue gas to preheat feed and generate steam. When it fouls, the effects spread through the whole furnace. The sequence is predictable:
- Convection fouling reduces heat recovery and forces higher firing rates.
- Higher firing increases radiant section stress and accelerates coke formation.
- Faster coking shortens run length and increases decoking frequency.
- Energy efficiency per tonne declines with each cycle.
What begins as convection fouling becomes a furnace-wide throughput and energy problem. Read against the IEA spread in energy intensity across naphtha crackers, much of the difference between plants comes down to how well they manage this cascade, how well they balance passes, and how closely they run to the real limits of the current run instead of conservative assumptions from earlier in the campaign.
Furnaces at the efficient end of the curve run closer to their actual constraints than to estimated ones, whatever the age of the equipment.
Where conventional process control reaches its limits
Advanced process control (APC) and model predictive control (MPC) have delivered real value on cracking furnaces for decades. APC stabilizes multivariable furnace operation and keeps operators on target. Its limits show when the furnace changes faster than the model can be maintained.
Cracking models contend with unknowns such as feed composition shifts, coil temperature profiles and maldistribution between coils. Few online measurements capture feedstock variation quickly and precisely enough to keep the model anchored in real time. That uncertainty matters most when operators weigh throughput against the run length penalty that follows.
Model drift and run-cycle dynamics
Fouling remains a major unresolved constraint. Throughput may rise with severity, but the value disappears if the run ends early enough to cut average throughput across the campaign. A controller that optimizes throughput without accounting for run length cannot reliably optimize margin.
Coking changes heat transfer, pressure drop and temperature profiles continuously within each run. A model identified near the start of a run degrades as the run progresses, then resets to a different baseline after decoking. Re-identifying the model after every decoke takes engineering time that most sites do not have spare. These time-varying dynamics sit outside the design envelope of conventional linear MPC.
Single-objective control and disconnected decisions
A controller tuned for ethylene selectivity can undervalue propylene yield. The decision about when to decoke often sits apart from the real-time severity decision. It is made by a different team, on a different cadence, using different assumptions about where the furnace stands in its current run. The two decisions should be linked, and in many installations they are not.
None of this diminishes the operational expertise built over decades at the board. The specific gap is that conventional approaches struggle to optimize severity, run length and energy together while conditions keep changing within the same run.
Learned control for the run-cycle trade-offs
These gaps share a pattern: nonlinear, time-varying relationships among variables that do not stay still long enough for a fixed linear model. A model trained on the plant's historical data learns those relationships directly, including feed-to-coking behavior, convection-to-radiant interactions and the severity-to-run-length trade-off as it shifts through each cycle.
Starting in advisory mode
The usual starting point is advisory mode. The model recommends setpoint changes, and experienced operators decide whether to act. On a cracking furnace those recommendations can be checked against what operators already watch: whether one pass is coking faster than its neighbors, whether stack temperature trends suggest convection fouling is progressing, and whether the severity targets from planning still reflect the furnace's current state.
Trust builds through repeated accuracy. A recommendation to rebalance a pass or adjust the steam ratio can be compared with board-level observations before anyone makes the change. The same visibility shows newer operators the reasoning behind setpoint decisions, so the model also works as a training tool.
Advisory mode is a stage on the way to closed loop. As confidence builds over several runs, the controller writes setpoints to the APC or DCS itself, inside limits set by plant engineers. The trade-off between severity, run length and energy is then managed continuously, instead of at the moments when someone has time to review it.
One view for planning, operations and engineering
A shared model changes cross-functional decisions too. When planning, operations and process engineering look at the same furnace behavior, severity targets can be weighed against run length and energy cost in one view. LP targets can be tested against the furnace's current coking state instead of an assumption from earlier in the run.
The decoke timing decision can draw on the same data that informs the real-time severity choice. That coordination is where petrochemical operations often leave the most margin on the table.
Frequently asked questions
When does stack temperature signal a real efficiency problem rather than normal variation?
Day-to-day movement is normal. The warning sign is a consistent rise above design. A sustained increase of 50°C above design stack temperature indicates lost heat recovery in the convection section. At that point the furnace is burning more fuel than it should to hold the same radiant section performance.
How can planning and operations stay aligned on severity targets during a run?
Alignment breaks down when LP targets are set at the start of a run and never revisited as the furnace changes. A practical fix is to build checkpoints into the run where operations shares the current coking state, stack temperature trends and pass balance data with planning. When both sides can see where the furnace stands, severity targets can be adjusted before the gap between plan and plant becomes a margin problem.
What does advisory mode look like in practice on a cracking furnace?
The model recommends adjustments to severity, steam ratio and pass balance based on current furnace conditions. Operators compare those recommendations with their own board readings before deciding whether to act. Repeated accuracy over several runs builds the trust needed to move to closed loop.
How Imubit applies this
Imubit applies Map, Model and Control to cracking furnaces and the wider olefins plant in chemicals and polymers. Map traces lost run length, yield and fuel to the recurring decisions behind them, including severity targets and decoke timing. Model uses Deep Learning Process Control® (DLPC®) to learn a nonlinear dynamic model of the furnace from its own history, guided by first principles, so the model tracks the furnace as it changes through the run. Control takes those decisions to closed loop, writing setpoints to the existing APC or DCS minute by minute, inside limits set by plant engineers.
