Every shift on an ethane cracker faces the same tension. Push coil outlet temperature (COT) higher to raise ethylene yield, or ease back to extend run length and delay the next decoke. The trade-off is old, but the margin environment around it has changed.
Ethane-based producers owe their cost advantage to the feed cost gap between naphtha and ethane. At the same time, the global ethylene downcycle has driven curtailments and raised closure risk for higher-cost capacity. The IEA's Oil 2024 report describes substantial new petrochemical capacity still entering the market, concentrated in China and the Middle East, so the pressure is not temporary. For an ethane cracker, each point of ethylene yield is now a larger share of a thinner margin, which moves olefins plant optimization up the list.
The severity trade-off at the furnace
COT is the main yield lever under direct plant control. Ethane cracking furnaces commonly run with COT in the 1,400 to 1,600°F range. The U.S. Department of Energy puts ethylene yield from ethane cracking at 80 to 84%, against 29 to 34% for naphtha, so small COT changes move margin directly. Higher COT also accelerates coke deposition. As coke insulates the tube walls, tube metal temperatures (TMT) climb toward the 2,000 to 2,100°F range.
Naphtha crackers spread their margin across propylene, aromatics and C4 coproducts. Ethane crackers depend almost entirely on ethylene, so there is no coproduct revenue to absorb a severity miss or a feedstock change. An ethane cracker running consistently below its achievable yield gives up margin every operating hour.
The best severity target is not a fixed setpoint. It shifts with ethylene price, ethane cost, decoking expense and maintenance windows. Early in a run, coke is thin and TMT headroom is wide, so higher severity costs little run length. Late in the run, the same push drives TMT up at a steeper rate and compresses the remaining production window. Each shift has to judge whether today's yield gain justifies a shorter run.
What decoking costs
Decoking is often treated as a simple trade of one more day of run length against one more notch of severity. In practice the cost stack includes lost ethylene production, steam and fuel for the decoke itself, and added work for maintenance teams.
Tube stress and equipment life
Thermal cycling through shutdown, decoke and restart adds tube and refractory stress over time. Plants that run at higher severity for short-term yield can pay for it later through tighter metallurgical limits, wider inspection scope or earlier tube replacement.
Over many cycles, accumulated stress can narrow the gap between a furnace's rated TMT limit and the effective limit that tube inspections support. That shrinking envelope constrains every run that follows.
Constraints that move across the complex
When one furnace comes down early, feed is redistributed and constraints move quickly. Limits can shift to the compressors, quench system, refrigeration or fractionation. Product quality risk can move to units that were stable an hour earlier.
The fouling penalty
Preheat and convection section fouling reduce heat recovery and raise firing. Higher firing means higher operating cost and higher CO₂ emissions. The harder cost to manage is what higher firing does to the TMT trajectory and run length, especially when fouling is uneven across the train.
Fouling progression is neither uniform nor fully measurable with standard instrumentation, so it adds uncertainty whenever severity moves closer to limits. In many plants that uncertainty becomes a conservative operating buffer that persists until the next decoke resets the furnace. Tying severity decisions to their energy effect, covered under cracking furnace efficiency, is one way to put a cost on that buffer.
Where conventional control falls short
The severity and run-length trade-off needs precise control that keeps up with a changing furnace. Conventional advanced process control (APC) was not designed for that.
Model drift
APC model accuracy on ethylene crackers tends to degrade as the run progresses. Key model parameters drift as feed quality assumptions, fouling and coil maldistribution go unrecalibrated. With different degrees of fouling across individual coils, estimating each coil's temperature profile from model-based approaches alone becomes unreliable.
That model error cascades. Controllers get detuned, analyzer feedback gets discounted, and soft constraints stand in for true boundaries. Operating points drift further from the economic optimum than they need to.
Control precision
Under conventional control, COT can wander several degrees around its target, and further during large disturbances. Every degree of variation has to be covered by a margin against the TMT limit. Tighter control lets a plant hold higher average severity and longer average run length without violating constraints.
Coupled dynamics
Cracking is a strongly coupled multivariable process. Furnace outlet temperatures affect downstream separation, and product quality depends on residence time, temperature profile and feed composition together. Single-loop PID control cannot coordinate those interactions, and the furnace-to-fractionation interface usually runs as two separate control domains.
Optimizing across the whole run
The core change is the objective function. Conventional control optimizes today's yield per pass. A controller built on a model learned from the plant's own data can optimize cumulative ethylene output across the full decoking cycle, counting the severity cost that shows up later as a shorter run. That difference matters most when furnace behavior shifts mid-run.
A learned model can capture the nonlinear relationships between COT, coke deposition rate, TMT progression and downstream separation at the same time. Because it learns from the furnaces' history from start of run to end of run, it reflects how the furnace behaves late in the run as well as early. Where coil-level measurements exist, it can track individual coils instead of furnace averages and see which coils are approaching limits faster than others, so the controller can adjust firing accordingly.
Instead of reacting once TMT nears its limit, the model forecasts the TMT trajectory under current conditions, and the controller sets severity to maximize total ethylene across the run.
Coordinating constraints across the complex
A furnace may have TMT headroom while the complex is limited by compressor head, refrigeration horsepower, demethanizer overhead loading or C2 splitter capacity. Pushing severity can also tighten acetylene or methane constraints that show up downstream before they look urgent at the furnace.
A learned model does not remove those limits. It makes the trade-offs explicit and consistent across furnaces at different fouling states, across shifts whose crews face the same constraint picture, and across competing objectives, by estimating the outcome of a move before it is made. When planning, operations and engineering share one model of plant behavior, the decision-making process rests on the same data.
From advisory to closed loop
The controller sits on top of the existing APC and DCS. It receives live data and writes setpoints through the established control infrastructure. Human and AI collaboration usually starts in advisory mode, with operators validating recommendations before any automatic move. As the recommendations prove out, the plant moves to closed loop, with operators setting the limits the controller works within and able to take over from the board.
Common questions
Why does the best cracking severity change when feed composition is stable?
Coke accumulation changes heat transfer and the TMT response as the run progresses, even with consistent ethane quality. The same COT target produces a different constraint picture late in the run than it did at the start. Ethylene price, downstream separation loading and the timing of planned maintenance also shift the economics of pushing severity against protecting run length. Feed variability adds to these effects.
Can a learned controller work alongside existing APC on an ethane cracker?
Yes. It is layered on top of the existing APC and DCS. It reads live operating data and writes setpoints through the same pathways operators already use. The DCS keeps running the regulatory loops, and APC keeps handling multivariable constraint control.
How does preheat fouling affect ethylene yield beyond energy cost?
Fouling lowers the temperature of feed entering the radiant section, so firing usually rises to hold COT. Higher firing increases thermal stress and can speed the TMT rise, which shortens the run between decokes. Fouling also adds uncertainty to severity decisions, because heat recovery loss is rarely uniform across the train.
How the Imubit Platform applies
Imubit applies Map, Model and Control to ethane cracker furnaces and the wider olefins plant as part of its work in chemicals and polymers. Map traces lost ethylene and run length to the recurring decisions behind them, such as severity targets, firing and decoke timing. Model uses Deep Learning Process Control® (DLPC®) to learn each furnace's nonlinear behavior through the run from its own history, guided by first principles. Control takes those decisions to closed loop, writing setpoints to the existing APC and DCS, inside limits set by plant engineers.
