Vacuum distillation units lose VGO yield between turnarounds as flash zone pressure, heater limits, and ejector performance tighten together, often beyond what fixed linear models can track. AI optimization learns these nonlinear relationships from plant-specific data and coordinates setpoint changes across interacting variables in real time. Plants can begin in advisory mode and progress toward closed loop operation, recovering VGO yield and protecting refinery margin as conditions change.
Vacuum distillation units rarely lose value in one dramatic event. More often, margin slips away as column pressure rises, heater limits tighten, and wash section performance drifts between turnarounds. By the time the operating window has narrowed enough to show up in the monthly numbers, the unit has already given back yield that an earlier intervention could have protected.
A vacuum distillation unit (VDU) takes the atmospheric residue from the crude unit and recovers vacuum gas oil under reduced pressure, separating distillates that would otherwise crack thermally if pushed through atmospheric distillation. About 80% of U.S. refineries operate a vacuum distillation unit, and for those plants the VDU is one of the largest single contributors to overall refinery margin through its yield of feedstock for downstream conversion.
That economic weight sits inside an environment where margin recovery has become harder. Downstream earnings dropped roughly 50% in 2024 versus 2023, sitting approximately 60% below 2022 peaks. Against that backdrop, every tonne of atmospheric residue that stays in vacuum residue instead of converting to vacuum gas oil (VGO) is value the unit doesn't recover, and decisions about cut depth and steam rates ripple directly into refinery operations and downstream conversion economics.
Recovering that value means balancing flash zone pressure, coil outlet temperature, wash section integrity, and ejector performance at the same time. Those variables interact nonlinearly, degrade at different rates between turnarounds, and shift with every crude slate change. Traditional control architectures handle each variable in isolation. VDU economics depend on how those constraints tighten together.
VDU performance usually slips between turnarounds as several operating constraints tighten at once. Recovering VGO yield depends on managing those interactions as feed quality and equipment condition change.
The sections below trace how those constraints build and where AI optimization changes the response.
VDU flash zone economics shape VGO yield through tightly coupled variables. Flash zone temperature, flash zone pressure, and stripping section performance each set part of the operating window, and every adjustment to one changes the constraints on the others. That's why VDU economics rarely respond to single-variable moves.
Higher flash zone temperature increases vaporization and VGO yield, but it's bounded by a hard ceiling that protects the heater coils and the residue pumparound. Lower flash zone pressure also increases vaporization, but it depends on ejector and condenser capacity that may already be operating near their limits. Stripping steam reduces hydrocarbon partial pressure to enable deeper cuts, but it also loads the overhead ejector system and pulls more energy through the refinery furnace circuit.
Deep-cut operation tightens the rest of the unit. Heavy vacuum gas oil (HVGO) quality becomes harder to hold as cutpoints extend. Wash section integrity matters more, and the margin between productive operation and coking narrows. At the board, these trade-offs translate into judgment calls operators make every shift: how aggressively to push cutpoints when the wash zone has already been stressed, how much steam to add when the ejectors are showing fatigue, and when to pull back to protect the unit. Those calls vary by crew, by shift, and by recent operating history, which is one reason yield variation between shifts often becomes a meaningful contributor to lost margin.
VDU performance degrades between turnarounds because several limits tighten together, not because of any single isolated failure. The pattern repeats across most refining operations: limits that would each be manageable in isolation interact, and each compensating move adds load somewhere else in the unit.
Preheat train fouling develops after startup, and as furnace heat exchanger effectiveness declines, furnace firing duty rises to maintain feed temperature. As heater tube skin temperatures rise from coke deposits, operators may lower firing severity or operating temperatures to protect tube metallurgy. Lower coil outlet temperature means less flash zone vaporization, so stripping steam rates often increase to compensate.
That adds more load to the ejector system. The wash zone comes under stress next. Reduced coil outlet temperature and variable overflash create inconsistent liquid distribution, and dry spots can develop where coking begins. Operators may increase overflash to protect HVGO quality, but that adds load to the rectification section and narrows flexibility further.
Later in the run, vacuum tower pressure often creeps upward as ejector performance weakens. At that point, the unit can be pinned against coil outlet temperature, HVGO quality, and column pressure all at once. A crude blend change on already-degraded equipment makes the operating window tighter still, and the cumulative effect of these interactions plays out across the crude oil refining process well beyond the VDU itself.
Traditional VDU control reaches its ceiling because the unit behaves nonlinearly while conventional model predictive control uses fixed linear step-response models built during commissioning. Flash zone behavior, vapor-liquid equilibria, and wash bed response don't move linearly as feed and operating conditions shift, so the mismatch between model and reality grows as the run progresses.
The problem deepens through the cycle. A conventional model doesn't adapt on its own as crude slates change and equipment degrades, so it becomes less representative of actual process gain, timing, and interaction effects. By mid-cycle, the recommended moves can be directionally correct but quantitatively off, and operators end up overriding the controller more often than the original deployment assumed they would.
The coordination gap between the crude distillation unit (CDU) and the VDU matters just as much. Many refineries run the two units as separate advanced process control (APC) applications, but the economic problem crosses that boundary. Shifting separation load between atmospheric and vacuum service, or balancing heater duty across both units, requires coordinated action across a shared constraint set, an issue that compounds across refinery process challenges more broadly.
Real-time optimization layers add economic targets, but those targets still rely on steady-state assumptions. Targets that look valid for one steady-state condition may not hold under actual plant dynamics. For a VDU that is moving with crude changes and equipment degradation, that limitation leaves value unrecovered.
No AI optimization technology replaces the pattern recognition that comes from decades at the board. But a VDU with several moving constraints can exceed what any operator or linear controller can continuously balance across all variables.
AI optimization adapts to VDU process changes through models built on plant-specific data that learn the nonlinear relationships fixed linear models miss: how flash zone conditions, wash section behavior, ejector performance, and downstream product quality move together. When crude properties shift, the model can adjust targets based on patterns in similar past transitions, rather than waiting for a re-tune. And instead of moving one variable at a time and watching the response, it can coordinate setpoint changes across interacting variables inside the unit's safe operating envelope.
A shared model across the CDU and VDU also changes how cross-functional teams work. It can show whether shifting separation load from atmospheric to vacuum service captures more total value under current conditions, and how that decision interacts with distillation yield optimization further downstream. Planning teams and process engineers see how cutpoint decisions affect downstream economics, and maintenance decisions become easier to prioritize when degraded equipment has a visible yield cost. That visibility tends to reduce the friction between groups that have historically pulled toward their own targets rather than the whole system.
Implementations that build trust usually start in advisory mode. The AI recommends setpoint changes, and operators decide whether to act on them. Senior operators test the recommendations against their own judgment, while less experienced operators see not only what changed but why a particular trade-off made sense under those conditions. Planning and economics teams use the same model to run scenarios on cut depth, steam rates, and crude blend changes before they hit the board, and process engineers track how catalyst, exchanger, and ejector performance evolve over time.
Many plants then move from advisory recommendations to supervised deployment with operator validation before advancing to closed loop operation as confidence builds. For a VDU where compound degradation erodes value between turnarounds, continuous adaptation matters. It keeps the unit operating closer to the best available point that current limits allow, and it does so within boundaries the operations team defines and can adjust.
For refining operations leaders seeking to recover the VGO yield and margin that compound VDU constraints erode between turnarounds, Imubit's Closed Loop AI Optimization solution offers a path forward. The Imubit Industrial AI Platform learns from actual plant data and writes optimal setpoints to control systems in real time, giving teams a dynamic view of the VDU and its interactions with the CDU as conditions change. Plants can begin in advisory mode, where operators evaluate recommendations against their own expertise, then progress toward supervised deployment and closed loop operation as confidence builds.
Get a Plant Assessment to discover how AI optimization can recover VGO yield and protect VDU margin between turnarounds.
Crude slate variability makes vacuum distillation unit optimization harder because the flash zone doesn't respond the same way to every feed, even when crude properties look similar on paper. Flash zone temperature, pressure, and stripping steam interact, so a crude switch mid-run on already-degraded equipment can tighten the operating window further. Under those conditions, single-variable adjustments become less effective because one correction often creates another limit somewhere else in the unit, particularly when heavy oil processing feeds enter the slate.
VDU ejector performance affects VGO yield by setting the pressure the flash zone must work against. When ejector performance weakens, effective vacuum tower pressure rises, feed vaporization falls, and VGO recovery drops. The effect is rarely isolated. Ejector limitations often combine with heater constraints, wash section stress, and HVGO quality pressure, which is why yield recovery depends on balancing the whole unit rather than treating vacuum performance as a standalone issue, and why coordinated refinery quality management matters across the conversion train.
AI optimization can work alongside existing advanced process control (APC) on a vacuum distillation unit. APC can continue handling stabilization, while the AI layer addresses economic target-setting and coordination across interacting constraints. That matters when current crude properties, equipment condition, and cross-unit economics no longer fit the assumptions built into fixed linear models. Many plants build trust gradually by starting in advisory mode before moving toward supervised deployment and closed loop operation, an approach that aligns with broader refinery ROI optimization priorities.
