Every hydrocracker runs with the same daily tension. Operations can push conversion harder to capture middle distillate margin, or ease back to protect catalyst life. In refinery operations, that trade-off is economic long before it shows up in product yields.
Each yield lever interacts with catalyst life, hydrogen balance and equipment limits, and those interactions tighten as the run progresses. As catalyst activity declines, the operating window narrows. Configuration choices set the economic envelope, and weighted average bed temperature (WABT) ties each yield lever back to the condition of the reactor.
How configuration sets the economics
Once-through operation usually leaves more unconverted oil (UCO) available for other uses. Recycle service pushes conversion higher, but it increases exposure to ammonia, H₂S and heavy polynuclear aromatic (HPNA) buildup in the reaction loop. Teams weighing these trade-offs often look at distillation yield across the refinery as well as single-reactor conversion.
These choices matter most in the middle and back half of the run. In sour recycle service, passivation of acidic sites can force a higher WABT to hold conversion. Higher WABT then raises hydrogen demand and speeds deactivation.
That trade-off is the center of yield control. Conversion targets have value when the unit can sustain them within normal hydrogen and utility limits. Hydrogen availability often sets a hard ceiling on severity, so any WABT increase late in the cycle draws on whatever hydrogen budget remains. The same constraints carry through crude oil refining economics across the complex.
HPNA management and UCO bleed
A two-stage sweet arrangement can improve selectivity and delay the trade-off, as long as fractionation, quench performance and separator operation stay tight. Configuration differences show most clearly in HPNA management.
Recycle operation concentrates heavy aromatics in the loop, which hurts catalyst performance and product quality. Bleeding UCO controls the buildup, but the yield penalty is immediate. The right bleed depends on crack spreads, hydrogen availability and how much run length remains in the catalyst cycle. Those variables shift week to week, so the best bleed rate and recycle ratio rarely stay constant for long.
How reactor condition preserves operating room
Hydrocracking reactions are strongly exothermic. Inter-bed quench is the practical tool for keeping bed temperatures inside a usable window, and reactor design sets how precisely the board operator can use it. When distributors foul or lose effectiveness, temperature becomes uneven across the bed and severity stops being uniform.
That loss of uniformity has a direct cost. Local hot spots coke catalyst faster, while underused zones deliver less conversion than their share of reactor volume should. The unit can look stable in aggregate while its effective window narrows, because the most stressed locations reach their limit first.
WABT and the other yield levers
Raising WABT can recover conversion after catalyst aging or feed deterioration. The same move raises hydrogen consumption, gas make and the risk of aromatic condensation. Higher feed nitrogen reduces cracking activity, and heavier, less stable feeds add load to both pretreating and fractionation. Units processing heavier or more variable slates feel this most, and they tend to run tighter WABT margins through the back half of the cycle.
The UCO cut-point also affects severity, because it changes recycle composition and the quality of the material returning to the reactor.
Separator and fractionator drift
The separator and main fractionator often decide how much of that burden returns to the reactor. Slight drift in separation efficiency can leave more dissolved light ends, contaminants or heavy material in the recycle path than operators intended. A cut-point adjustment that looks minor in the tower can change recycle quality enough to raise hydrogen demand, lower selectivity and force another WABT increase a few hours later. This can happen while average reactor temperatures still look acceptable, so the yield loss does not show in reactor data alone.
These feedback loops run on different timescales. A cut-point change may take hours to register in reactor performance, while the WABT response to hold conversion shows up in minutes.
The interactions also reach into the unit's thermal network. Exchanger performance, and heat exchanger fouling in particular, decides how much preheat reaches the reactor section.
Early signs that the window is narrowing
- WABT has to rise to hold conversion.
- Quench flexibility shrinks.
- Hydrogen consumption climbs.
- Yields become more sensitive to normal feed shifts.
- Separator or fractionation disturbances carry a larger economic penalty than they did earlier in the cycle.
Why conventional control loses ground through the run
Advanced process control (APC) works well when process relationships stay close to the model it was tuned on. Hydrocrackers rarely stay there for long. Catalyst aging changes those relationships continuously, and feed quality shifts do the same. A model tuned well early in the cycle becomes less reliable as severity rises, pretreat performance changes or recycle composition drifts. For a unit processing multiple crude blends in sour recycle service, that mismatch can develop within weeks of a tuning update.
The constraints also interact. Exchanger fouling cuts preheat, fired duty rises, reactor severity shifts and hydrogen demand moves toward compressor limits. Each link is manageable on its own. Coordinating them together in real time is harder than any single control loop accounts for.
The practical result is conservatism. Operators leave conversion on the table to guard against constraints they can see one at a time but cannot evaluate together in the time available. No APC model substitutes for the pattern recognition that comes from years at the board, and no operator can weigh every interaction continuously across a full run.
The planning gap
LP targets often assume a catalyst state and constraint set that no longer match the plant. Operations compensates for the actual limits. Maintenance may defer exchanger work that operations already sees as a severity constraint, and engineering may evaluate capital projects without seeing the workarounds the unit already relies on. Each function optimizes within its own information, and on a hydrocracker the constraint picture changes faster than quarterly planning cycles can track.
A shared, current model built from the unit's own operating data gives every function the same view of actual constraints and trade-offs. Planning can revise targets when catalyst activity or hydrogen limits shift. Maintenance can prioritize exchanger or quench issues that already constrain severity. Operations can explain why temporary workarounds are losing effectiveness.
How a learned model supports hydrocracker decisions
A model trained on the unit's own operating history is useful when unit behavior moves faster than retuning cycles. On a hydrocracker it shows more than a severity recommendation. It lays out the trade-off between conversion, gas make, hydrogen draw and the room left on WABT.
Planning can test scenarios against current constraints instead of LP assumptions that may no longer hold. Experienced operators can compare recommendations with patterns they already know, and newer operators can see why one shift held conversion while another backed off. This is human and AI collaboration applied to daily hydrocracker operation.
From advisory mode to closed loop
Plants usually start in advisory mode. The model recommends setpoint changes, and operators decide whether each one fits current conditions. Trust builds as recommendations hold up across shift changes and feed conditions. Recommendations are then checked against operating boundaries and team expectations, and the plant moves to closed-loop control once the model has proved itself on the unit. Operators set the envelope and can take over from the board. Over successive catalyst cycles the plant keeps a steadier operating strategy instead of depending on whichever instincts are on the board that day.
How the Imubit Platform applies
Imubit applies Map, Model and Control to the hydrocracker as part of its work in refining. Map traces refinery margin lost through the run to the recurring decisions behind it, such as WABT, quench, recycle ratio, UCO bleed and cut-points. Model learns how those levers interact with catalyst aging and feed quality from the unit's own history, using Deep Learning Process Control® (DLPC®). Control takes those decisions to closed loop, writing setpoints to the existing APC and DCS within the unit's safe operating window and, inside limits set by plant engineers.
