Every ethylene complex runs on the same clock. From the moment a pyrolysis furnace returns to service after decoking, coke starts to build on the radiant tube walls, tube metal temperature (TMT) starts to climb, and the window for holding target severity narrows. The decisions operators make inside that window set yield, fuel consumption, and the date the furnace comes offline again.
The stakes go beyond one furnace. With overcapacity pressing on olefins margins, producers need more value from the assets they already run. Holding severity while managing coking rate across several furnaces is one of the clearest furnace efficiency levers available without new capital.
How coke accumulation reshapes every furnace decision
The operating principle of a pyrolysis furnace is straightforward. Set feed rate, severity and steam ratio so that every pass cokes at the same rate and reaches maximum TMT on the planned decoking date. In practice that target keeps moving.
As coke thickens on the tube inner walls, heat transfer falls. The distributed control system (DCS) responds by firing more fuel to hold coil outlet temperature (COT). TMT rises as the insulating layer grows, and the gap between the current tube temperature and the metallurgical limit narrows week by week.
That narrowing accelerates. Each additional increment of coke produces a larger TMT response than the one before, so headroom disappears faster late in a run than early in it. Operators who extrapolate from early-run TMT trends can underestimate how quickly the limit arrives.
Raising COT to hold severity also raises TMT. As the run goes on, the furnace reaches a point where target severity cannot be held without exceeding safe tube temperature. Coke accumulation lowers the achievable severity ceiling, and TMT becomes the binding constraint.
How downstream fouling compounds the energy penalty
The effects do not stop at the radiant section. Fouling in the transfer line exchangers raises backpressure and increases compression energy downstream. Fouling in the convection section reduces feed preheat, and operators compensate by firing harder.
The extra firing adds thermal fatigue to the tube materials, which can shorten tube life independently of the coking cycle. A furnace that started its run at target severity and moderate fuel per tonne of ethylene ends it consuming more fuel per tonne at lower selectivity. The gap between the planned and the actual decoking date is lost ethylene production that no downstream unit can recover.
How pass imbalance shortens the run
A pyrolysis furnace has several parallel radiant coil passes. If one pass cokes faster than the others, the furnace may have to come offline for decoking before the remaining passes reach their full run potential. Each premature decoke means lost production time and a higher operating cost per tonne of ethylene.
The economics of decoke timing are asymmetric:
- Pulling a furnace too early leaves capacity unused. Tubes that still had weeks of headroom produce nothing during the decoke.
- Running too long risks a forced shutdown on a TMT exceedance, or tube damage that extends the outage well beyond a standard decoke.
Operators usually err on the conservative side. That is the safer choice, and it means the fleet rarely runs at its theoretical maximum availability.
Why coking rate is harder to equalize than it looks
Equalizing cracking and coking rates across passes depends on precise control of the steam-to-hydrocarbon ratio and of flow distribution to each pass. Small imbalances create asymmetric coking patterns that shorten the run.
The objective is to equalize the coking rate itself, which is more than volumetric balance, and the coking rate cannot be measured directly during normal operation. Tube skin thermocouples sample specific locations, and the real hot spot on a tube may sit between measurement points. That gap adds uncertainty to every run-length decision.
Feedstock variability makes it harder. A different naphtha composition can crack differently at the same COT. The yield distribution shifts and the coking rate changes even though no operator-controlled variable has moved. The same furnace on the same setpoints can behave measurably differently on Tuesday's naphtha than on Monday's.
Firebox burner uniformity adds to the problem. Uneven burner performance causes local overheating that accelerates carburization in specific tubes. Firebox balancing is a tube life question as much as a process control question, and because burner condition interacts with coking rate, furnace maintenance decisions and process optimization decisions are never fully independent.
Why fleet coordination exceeds conventional control boundaries
A typical ethylene complex runs several furnaces, usually with one in decoke, one in hot standby and the rest producing. Each producing furnace sits at a different point in its coking cycle, runs a different feed composition and operates at a different severity. The combined cracked gas feeds shared separation and compression trains.
That arrangement means the plant is never at steady state. Conventional advanced process control and model predictive control are typically designed around steadier assumptions. A complex with furnaces at different stages of coking, on feeds of varying composition, shifts continuously, which is one reason plantwide control has remained hard to achieve in olefins operations.
Where measurement gaps and feed allocation collide
The measurement infrastructure reinforces the limit. Severity is not measured directly. It is calculated from product ratios or online analyzer results, and a gas chromatograph is often shared across several furnaces, so feedback for any one furnace arrives intermittently. Between readings, controllers interpolate while feed composition and coking state continue to change.
Feed allocation creates its own dependencies. Changing feed selection changes fouling rate and run length. When one furnace goes offline for decoking, the others absorb the lost throughput, which can push them harder and accelerate their own coking.
None of these variables moves independently. Furnace severity affects downstream column loading, feed allocation drives coking rate, and co-product economics shift the best severity target. No model replaces decades of operating experience on a cracker complex, but the number of interacting variables across a furnace fleet is more than any operator or conventional controller can track at once.
What changes when a model follows the full degradation cycle
The degradation loop is a multivariable problem that never settles. Coke accumulation raises TMT, increases fuel demand, compresses severity and unsettles downstream operation at the same time. A model learned from the plant's own operating data fits that pattern better than one built on idealized assumptions about how the process should behave.
A model trained on each furnace's history can learn the relationships between operating parameters and coking behavior for that furnace. Tracking coking state continuously, it can manage constraints as TMT headroom changes, and it can coordinate fleet-level feed allocation based on where each furnace sits in its cycle.
The same view supports decoke planning. When each furnace's coking state, TMT trajectory and remaining headroom are known, decoke timing can be coordinated across the fleet instead of decided furnace by furnace. The aim is to keep decokes staggered so plant-level throughput stays more consistent, an idea central to self-optimizing petrochemical plants.
Inferential models built on operating data can also estimate severity from the available process measurements between analyzer readings. That narrows the gap between analyzer timing and the control actions that depend on it, a recurring problem in olefins plant operations.
Starting in advisory mode
Advisory mode is usually where operators test the model against what they know. The model recommends setpoint changes, and operators decide whether to act. An operator who has run a furnace through dozens of campaigns often has a feel for how that unit responds to a feed change or a burner adjustment. Advisory mode lets that operator compare the model's predictions with experience. Where they agree, confidence builds. Where they diverge, the difference often surfaces assumptions worth examining on both sides.
As recommendations prove out, plants move to closed loop. The controller writes setpoints to the APC or DCS, and operators can take over from the board, within limits they set.
The shared view also matters at crew change. When the model shows why one furnace can take more severity while another needs protection, operators at every experience level can see the coking-state reasoning behind each recommendation. Operator training benefits from seeing that logic applied to real cases. Because the model learns from operating outcomes, it keeps a record of the process states and decisions that produced better results, which supports knowledge transfer as experienced operators move on.
How the Imubit Platform applies
On chemical and polymer plants, the Imubit Platform applies Map, Model and Control to the cracker furnace fleet. Map traces lost run length, ethylene yield and fuel to the recurring decisions behind them, such as severity targets, feed allocation and decoke timing. Model uses Deep Learning Process Control® (DLPC®) to learn each furnace's nonlinear dynamics through the run from its own history. Control takes those decisions to closed loop, inside limits set by plant engineers, writing setpoints to the APC and DCS already in place.
