The short version: Refineries are not replacing their process simulators with AI. They are blending the two. Emerson’s AspenTech business has deployed its Aspen Hybrid Models in Aramco’s refinery planning, combining first-principles engineering models with AI, and reports yield and quality predictions up to 98.5% accurate in key units. It is the clearest example so far of how AI is entering process engineering: on top of the physics, not in place of it.
What happened
Emerson and Aramco announced in 2026 that Aspen Hybrid Models had been built into Aramco’s existing refinery planning setup, creating what they describe as one of the largest multi-site, multi-period planning models in the industry. Aramco used thousands of converged simulation cases, built on first-principles models and calibrated with real plant data, to create the nonlinear models its planners now use.
Early work focused on catalytic reforming units, the CCR and Platformer, and is meant to be the base for a wider rollout across Aramco’s refining network. The 98.5% figure is company-reported, not an independent benchmark.
What a hybrid model is
- Physics first. The base is a first-principles model in tools like Aspen HYSYS or Aspen Plus, built on mass and energy balances, kinetics and thermodynamics.
- AI fills the gaps. Machine learning, trained on plant data, captures behavior the equations miss, such as catalyst aging or complex reactor effects.
- Stays accurate longer. Plants drift over time. Hybrid models are designed to update with less manual tuning, so planning models do not go stale.
- Usable by more engineers. AspenTech says its AI Model Builder lets engineers at different skill levels build these models without a data science team.
AI in the control room too
AspenTech is also putting an AI advisor called AVA into its DMC3 advanced process control software. Repsol’s La Pampilla refinery in Peru has trialed it on two crude distillation units and two vacuum distillation units, where it translates what the multivariable controllers are doing into plain guidance for operators and engineers.
Why refiners care
Refinery margins depend on running the right crude through the right units to make the right products. A planning model that predicts yields more accurately means better crude buying and fewer surprises. Reactor models in particular have been hard to build and keep current, and AspenTech claims hybrid models can cut the time to maintain reactor simulations by up to half and reduce reliance on outside consultants.
Who should pay attention
Refiners, petrochemical and chemical plants, and the engineering firms that support them. That includes the Gulf Coast cluster around Houston, Beaumont, Port Arthur and Lake Charles, home to a large share of US refining capacity (see our US oil refineries map).
The catch
A hybrid model is only as good as the first-principles model under it and the plant data it learns from. Bad instruments, unlogged process changes or a weak base simulation will produce confident wrong answers. Someone who understands the unit still has to judge whether a prediction makes sense.
What it means for hiring
AI in refining rewards process engineers who know both the simulator and the unit, and can tell when a model is wrong.
- Aspen HYSYS and Aspen Plus experience is still the foundation. Hybrid models sit on top of those first-principles models.
- Engineers who can clean and judge plant data are worth more as models learn from it.
- APC and planning engineers who understand both the controller and the economics are hard to find and hard to replace.
Hiring for a refinery or chemical plant? Read how to hire a process engineer, or see our chemical engineering recruiters page.
Interview questions for process engineers
- How do you calibrate a unit simulation against plant data, and how do you know when it has drifted?
- A model predicts a yield that looks too good. What do you check first?
- Tell me about a time bad instrument data led you to the wrong conclusion.
- Which Aspen tools have you used, and for what kind of units?
Frequently asked questions
What are Aspen Hybrid Models?
Process models from AspenTech that combine first-principles engineering simulation with AI trained on plant data, used in tools such as Aspen HYSYS, Aspen Plus and refinery planning software.
How is AI used in oil refineries?
Common uses include production planning and scheduling, more accurate unit and reactor models, advanced process control advisors and predicting equipment failures before they happen.
Will AI replace process engineers?
No. The leading tools are built on engineering models and still need engineers to build, calibrate and question them. AI makes good process engineers more productive.
More stories like this are on our AI for Engineers hub.
