Catalysis 2026 Issue

Operators understand qualitatively that increasing Z-additive promotes propylene formation, but the quanti- tative impact depends on catalyst activity, feed quality, and regenerator constraints. Conventional linear models rep- resent additive effects as fixed coefficients, masking these critical cross-interactions. At one refinery, a commercial target was defined for the minimum iso-butene fraction (iC₄e% [or iC₄=%]) in the mixed C₃/C₄ stream. Over several years, engineers experimented with adjusting the injection of two different Z-cat formula - tions to influence both iC₄e% and propylene yield. Results were inconsistent: periods of higher additive use improved LPG recovery but simultaneously reduced gasoline yield and upset iC₄e% control. The inability to manage these competing outcomes constrained overall margin capture during favourable propylene-gasoline spreads. What AIO learned Imubit’s AIO platform trained a high-fidelity dynamic model on multiple years of operating history, enabling it to learn the true, multidimensional response surface governing FCC ole- fin yields. The model differentiated among the various con - tributing factors, revealing previously hidden non-linearities: • Independent relationships for both Z-additive types, quan- tifying each formulation’s distinct influence on propylene, i-butene, and LPG composition. • Riser-outlet temperature effects, showing that while higher ROT increases total iC₄e flow, it dilutes iC₄ % by enhancing parallel cracking reactions. • Feed-composition sensitivity, isolating the impacts of very heavy VGO, hydrocracker bottoms, and sour feed compo - nents on additive performance.

By embedding these relationships into a closed-loop RL controller, AIO continuously optimised additive rates and severity settings to maximise profit while maintaining the specified iC₄e % limit and all regenerator constraints. Results The AIO model provided engineers with clear, quantifiable insight into how each Z-cat formulation influenced both overall gas yield and the specific iso-butene fraction (iC₄e %) in the C₃/C₄ stream. By coupling this additive behaviour with its understanding of ROT severity and cracking yields, the RL controller was able to deliver a holistic optimisation strategy, dynamically balancing additive dosage, feed composition, and severity to maximise total margin while maintaining target product quality. Across comparable FCC deployments, AIO has demon- strated approximately +$0.25 per barrel margin improve - ment and multi-million-dollar annualised benefit, achieved by unlocking the non-linear synergy between additive chemistry and operating severity. Technical insight AIO exposed that dilution effects could reduce the measured iC₄e% even as absolute iC₄e production increased – a sub - tle phenomenon masked in linear analyses. This explained why past operational adjustments, though directionally correct, had appeared ineffective when feed composition or ROT shifted. By distinguishing true conversion effects from apparent dilution, AIO enabled engineers to interpret prod- uct behaviour correctly and adjust strategy with confidence.

Case 2: Predicting long- term furnace fouling and extending run length The challenge Steam cracking and crude preheat furnaces suffer grad- ual fouling and coking that raise tube metal tempera- ture (TMT) and shorten run length. Operators traditionally respond reactively. When TMT nears its limit, severity is reduced, or coils are cleaned, often after efficiency losses have already occurred. True drivers, including feed com- position, operating severity, and pass-level heat-flux pat - terns, are rarely quantified until late in the cycle. At one European refinery, this reactive strategy led to multiple unplanned furnace shutdowns each year. Post- incident reviews linked foul - ing to crude variability, but no predictive model could

Coke

Tube (where the TMT is measured)

Process temperature inside the tube

TMT

Feed quality

Heater outlet temperature

TMT A

TMT B

TMT C

TMT D

Mass ow rate

Pass 1

Furnace dp

BFW rate

Heat transfer resistance

Boiler feed water

Pass 2 Pass 3 Pass 4

Included in model similar to Pass 1

Burners

MV Firing duty

CV = conversion

Figure 1 AIO model for heater with fouling tendency

Catalysis 2026