Refinery profit relies heavily on the Fluid Catalytic Cracking (FCC) unit. This seven-step guide explains how industrial AI can increase profit by optimizing FCC yields in real-time, where traditional methods fall short. It covers setting KPIs, identifying opportunities, using high-quality data, training AI models, deploying closed-loop control, monitoring results, and scaling best practices across the plant for sustained gains.
The fluid catalytic cracking (FCC) unit is the workhorse of gasoline production and one of the highest-margin units in any refinery. The process converts heavy, low-value hydrocarbons into gasoline, LPG, and light olefins. A significant share of crude oil refined globally passes through FCC conversion on its way to finished fuel.
Yet even well-run FCC units leave value on the table. Feed variability, catalyst degradation, and competing process constraints create a moving target that traditional advanced process control (APC) handles imperfectly. The gap between current yields and achievable yields represents recoverable margin, and for many refineries, AI optimization is closing that gap.
FCC units rank among the most profitable and complex in any refinery. Understanding where the process loses value shows where AI optimization can recover the most margin.
Here's where FCC performance breaks down, why traditional controls miss these dynamics, and how AI optimization recovers the margin.
Fluid catalytic cracking is a chemical conversion process, not a physical separation like distillation. Inside the FCC unit, heavy gas oil contacts a hot, powdered zeolite catalyst that breaks large hydrocarbon molecules into smaller, more valuable ones. The catalyst circulates continuously between the reactor and regenerator, behaving like a fluid when suspended in hot vapor.
What makes FCC optimization so difficult is the tightly coupled loop at its core. In the riser reactor, vaporized feed meets regenerated catalyst at elevated temperatures. Cracking reactions produce gasoline-range hydrocarbons, light olefins, and lighter gases while depositing carbon (coke) on the catalyst surface.
The spent catalyst flows to the regenerator, where air burns off the coke, restoring catalyst activity and generating the heat that drives the entire cycle. Adjusting riser temperature to increase conversion also changes coke yield, which alters regenerator temperature, which feeds back into catalyst activity.
Every operating decision cascades through the system. And downstream, the main fractionator and gas recovery section split cracked vapors into products with distinct and shifting economic value, so any optimization approach has to account for the full chain.
Two FCC units running similar feed can produce meaningfully different margins. The difference comes down to how well operators and control systems navigate the unit's interdependent constraints. Value chain optimization efforts across refinery commercial and manufacturing functions keep landing on the same conclusion: conversion units like the FCC are where recoverable margin hides.
Most FCC units end up operating well inside their actual capability. At the throughput volumes a typical FCC handles, even modest improvements in gasoline yield or a small reduction in delta coke can translate to significant margin over the course of a year. Closing this gap takes a control approach that can see and respond to all these interactions simultaneously.
Conventional APC has been the standard approach to FCC optimization for decades. These systems use linear or piecewise-linear models to coordinate manipulated variables and push the unit toward constraint limits while maintaining stability. APC works well for industrial process control at moderate feed changes: it can hold riser temperature targets, manage cat-to-oil ratios, and respect regenerator limits within its model range.
For many units, APC has delivered meaningful improvements over manual operation.
The limitation is structural. APC models represent a snapshot of unit behavior under specific conditions. When feed quality drifts, catalyst activity changes, or economic targets shift, the accuracy of setpoint controls degrades. Re-tuning to match current conditions is labor-intensive and typically happens on a schedule, not in response to what the unit is actually doing.
The interactions between riser severity, coke selectivity, regenerator heat balance, and fractionator cut-points are nonlinear, and linear models approximate them only within a narrow operating window.
The biggest optimization opportunities sit exactly where APC struggles: at the boundaries between safe operation and constraint limits, during feed transitions, and across the connected system of reactor, regenerator, and fractionator. These conditions create the largest swings in margin, and traditional approaches handle them conservatively by design.
The absence of effective plantwide process control adds to the problem. FCC economics depend on downstream conditions: gasoline pool octane requirements, alkylation feed needs, and product pricing. APC optimizes the FCC in relative isolation, missing the system-level value that comes from coordinating FCC operating targets with the broader refinery.
The core difference is in how the model is built. Instead of linearized representations of unit behavior, AI optimization trains deep learning models on actual plant data, capturing the full nonlinear relationships between feed conditions, operating parameters, and product outcomes.
These models learn from years of actual plant data to represent how the unit behaves, including operating states that a first-principles model would struggle to predict. Because the model has learned from a wide range of conditions rather than being calibrated to one set, it stays accurate as the unit moves through different feed campaigns, catalyst ages, and economic environments.
In practice, AI optimization addresses FCC constraints in ways that APC can't:
The operational path typically starts in advisory mode. The AI model posts recommended setpoint changes to operator dashboards, and experienced operators evaluate those recommendations against their own judgment. This builds trust and reveals edge cases the model needs to learn from.
As confidence develops, plants can progress to closed loop optimization, where the model writes setpoints directly to the distributed control system (DCS) within operator-defined boundaries.
The AI handles the continuous, multivariable optimization that exceeds human bandwidth. Experienced operators stay focused on the judgment calls, safety decisions, and exception management that no model replicates. And when operations, planning, and economics teams share a single model of FCC behavior, the recurring arguments about crude slate changes, FCC severity targets, and light olefin routing get grounded in the same data.
That's a meaningful shift from the current norm, where each group operates with different assumptions about what the unit can actually do.
For refinery leaders looking to capture the margin their FCC units leave behind, Imubit's Closed Loop AI Optimization solution provides a data-first path forward. The technology learns from actual plant operations, trains a deep learning model on years of historical data, and writes optimal setpoints back to the DCS in real time.
Plants can start in advisory mode, building confidence and operator trust before progressing toward full closed loop optimization as alignment develops. With 90+ successful applications across process industries, Imubit works as a long-term partner, not just a technology vendor.
Get a Plant Assessment to discover how AI optimization can improve gasoline yields, reduce delta coke, and capture the full margin potential of your FCC unit.
Feed variability directly impacts FCC margins because changes in crude slate composition alter cracking behavior, coke formation rates, and product distribution. Heavier or more contaminated feeds increase delta coke and limit conversion, while lighter feeds may shift output away from the highest-value products. An AI model trained on plant data tracks these feed shifts in real time and adjusts operating parameters proactively to maintain distillation yield targets even as feed quality changes throughout a crude campaign.
AI optimization typically layers on top of existing APC infrastructure rather than replacing it. The APC continues handling base-layer process control while the AI model provides higher-level setpoint targets that account for nonlinear interactions and changing economics. This preserves existing investments and allows operators to validate AI recommendations against familiar control system behavior during the transition.
An effective FCC optimization model needs process variable data from the DCS, laboratory quality results with accurate timestamps, catalyst activity and circulation metrics, and economic inputs that drive refinery ROI, including product prices and utility costs. Plants can begin with existing data; perfectly clean datasets aren't a prerequisite. The model learns from whatever operational history is available and improves as data infrastructure matures over time.
