Every barrel of crude that enters a distillation column holds more recoverable value than most refineries capture. Small volume losses across the atmospheric and vacuum units are common where cut points run conservative or the crude slate varies. Over a year they add up to a large amount of margin.
Recovering that value does not require new hardware. It depends on three operating areas: sharper separation, tighter quality control, and real-time coordination between planning targets and tower operation. For the wider context, see process optimization in oil and gas.
Where yield is lost between the plan and the column
The largest source of lost distillation yield is rarely a single equipment failure or a dramatic upset. It is the cumulative effect of conservative operating margins built into daily operation. Because these losses are operational, they can be recovered without capital investment.
Cut points and quality giveaway
Cut points, the draw temperatures and pressures that set where each product leaves the column, are the main lever. A few degrees of conservatism on the diesel cut pushes recoverable diesel into lower-value residue. At a 200,000 barrel-per-day refinery, holding that margin across a full crude slate adds up to a significant loss of revenue over a year.
Operators run those margins for good reasons. Sample results arrive hours after the product has moved downstream. Without real-time quality visibility, the safe strategy is to keep a cushion between actual product quality and the specification limit.
That cushion is quality giveaway. It is product that consistently beats its specification, and it represents margin given up to avoid off-spec risk.
When planning and operations fall out of sync
The disconnect between planning and operations makes the problem worse. Linear-program (LP) models set yield targets from economics, and those targets are usually updated daily or weekly. Column conditions change continuously with crude quality, ambient temperature and equipment fouling. In between, the plan depends on recurring decisions, made every shift by different people with different information.
Heat exchanger fouling alone can shift flash-zone temperature by several degrees over weeks, steadily widening the gap between planned and achievable yield. Stripping performance and turnaround timing add further drift that LP models rarely capture in real time. By the time the planning team recalibrates, the column has been under-recovering for the whole interval, and that loss accumulates across operating cycles.
Feed variability makes manual correction impractical. At many refineries the crude slate changes every few days, and each change shifts the boiling curve enough to require new column conditions. A cargo of heavier crude can need different draw temperatures from the lighter slate the column ran the day before. Without fast, coordinated adjustment, the unit either gives up yield to keep safe margins or risks quality excursions that send product to reprocessing.
The traditional route to more yield is hardware: better internals, deeper vacuum or a new column configuration. Those deliver real gains, but they need capital and take time to execute. The operational losses described above can be addressed through control strategy on the existing equipment.
How real-time optimization recovers lost yield
Closing the gap means working on cut point management, quality visibility and economic alignment together. Each targets a different source of loss, and each works within the existing column.
The common thread is a model learned from the column's own operating data, guided by first principles. It captures the nonlinear process interactions that make manual optimization so hard to sustain.
Tighter cut points through multivariable coordination
Traditional advanced process control (APC) moves column parameters toward targets that stay fixed between updates. Those targets reflect conditions at the time they were set, which may not be the conditions the column faces an hour later.
A model trained on years of plant data learns how flash-zone pressure, draw-tray temperature, pumparound rate and reflux ratio interact, and how those interactions carry through to product yields and quality. Instead of holding cut points at fixed margins, it recalculates the optimal operating point as conditions change. When a crude switch moves the boiling curve, the model adjusts before the next lab result confirms what operators already suspect. That continuous recalculation captures yield that periodic optimization studies and manual moves leave behind, with no change to column hardware.
Consider a diesel draw held 3°C conservative to protect its back-end distillation specification. The margin exists because the last lab sample was taken four hours ago and conditions may have shifted. If an inferential model confirms the product is well inside specification, the cut can be tightened. That recovers diesel that would otherwise go to atmospheric residue. Applied across every product draw on the column, the recovered volume adds up.
Inferential quality models that close the sampling gap
The hours between taking a sample and acting on the result are one of the largest barriers to tighter operation. Inferential models, also called soft sensors, predict properties such as flash point, sulfur content and density from the high-frequency process signals already in the control system.
With an accurate, always-on estimate of product quality, operators or the controller can move setpoints closer to specification limits instead of holding a cushion against lab uncertainty. On a large crude column, even one degree less cushion can be worth thousands of barrels a month of recovered yield. Regular comparison with fresh lab results keeps the model calibrated as feed, upstream units and equipment condition change.
Connecting tower operation to economic targets
The largest gains come from coordinating column operation with real-time economics. When diesel margins move relative to gas oil, the optimal cut point moves too. When energy prices rise, the trade-off between heater or reboiler duty and incremental yield changes.
An optimizer can weigh these trade-offs continuously by bringing LP data, market prices and actual tower conditions into one decision. That aligns distillation operation with what the refinery needs to produce at that moment, and it coordinates across the column instead of optimizing each draw in isolation. A column run for maximum diesel yesterday may need to shift toward kerosene today because jet margins moved overnight.
A refinery can raise total yield and still lose margin if the extra product goes to lower-value streams. The target is higher-value yield from the existing equipment, gained through better operating decisions.
What changes on the operating floor
Reducing shift-to-shift variability
A refinery running four rotating shifts can see a measurable yield spread between its best and worst crews on the same unit with the same crude. One crew runs tighter cut points, another responds faster to crude changes, and a third holds more margin on nights. Over a year that spread is lost margin that no capital project can recover, because the equipment is the same for every crew.
Setpoint optimization narrows the spread by giving every shift the same optimized starting point for cut points, reflux ratios and draw temperatures. The consistency effect is often underestimated, because part of the available yield sits in the gap between a unit's best and worst operating windows.
From advisory mode to closed loop
Most deployments begin in advisory mode. The model recommends setpoint changes, and operators evaluate, accept or override them. Engineers compare the model's predictions with actual column behavior, operators see variable relationships that were not obvious before, and confidence builds on demonstrated accuracy. Advisory mode is where that evidence is gathered.
Once that confidence is in place, the controller moves to closed loop. It writes optimized setpoints to the APC or distributed control system (DCS) within boundaries the operators define, and the operating team keeps override authority at every stage. ARC Advisory Group describes this arrangement as human on the loop, in which the AI calculates and executes and the operator moves from manual controller to high-level supervisor.
Results from existing infrastructure
None of this needs new instrumentation or control hardware. The models learn from data already in the plant historian: the temperatures, pressures, flows and quality measurements operators use today. A model can process thousands of those signals at once and find opportunities that manual analysis, limited by human bandwidth, cannot sustain over a full operating cycle.
The indicators that show it is working are familiar to any operations team. Yield at the blend headers trends up. Product quality runs closer to specification limits without excursions. Energy per barrel falls as unnecessary reboiler duty comes out. Off-spec events that need reprocessing become less frequent.
Where Imubit fits
In refining, Imubit applies Map, Model and Control to units such as the crude and vacuum towers. Map traces yield and giveaway losses to the recurring decisions behind them. Model uses Deep Learning Process Control® (DLPC®) to learn a nonlinear dynamic model of the column from its own history, guided by first principles. Control takes the cut point and draw decisions to closed loop, on top of the existing APC and DCS and, inside limits set by plant engineers. In one refinery naphtha and jet complex with three fractionators, this approach delivered a 2 to 3% increase in jet recovery on top of existing APC.
Frequently asked questions
Why do conservative cut points persist when engineers know they cost yield?
The main reason is data latency. Engineers know tighter cuts would improve yield, but when lab results arrive hours after the product has moved downstream, a cushion against the specification limit is the rational response to uncertainty. Inferential models that predict product properties from continuous process signals give operators the confidence to tighten cuts without adding off-spec risk.
How is the yield improvement measured?
Many refinery deployments establish a baseline period before the controller is activated, then track yield percentage, quality giveaway and energy per barrel against it. The model keeps learning from new operating scenarios, crude switches and seasonal conditions, so the comparison continues as the system matures.
Can yield optimization work alongside existing APC?
Yes. The DCS handles regulatory control, and APC handles multivariable control toward its targets. The optimizer sits above both and recalculates the targets APC works toward as feed quality, economics and nonlinear process interactions change, which conventional control was not designed to handle. Both layers keep working together through the existing DCS.
