Honeywell puts a per-barrel value on its AI plan for Dangote’s refinery

Honeywell Technologies is launching Forge Plant Cognition at Dangote’s Lekki refinery in Nigeria. Announced on 26 October, the platform brings operational data, process expertise and AI into decisions about the plant’s performance.[1]

Honeywell’s preliminary estimate is more than 29 cents of operating benefit per barrel, or above $60 million a year at full-capacity throughput. Those are projected gross benefits, excluding implementation costs.[1]

The proposed gains come from identifying bottlenecks, improving throughput and product yield, anticipating equipment problems and giving operators access to specialist process knowledge.[1]

The per-barrel figure makes the commercial argument easier to examine. A small gain repeated across a large volume can support a substantial annual benefit. The investment decision still has to account for the cost of putting the system into service and the amount of production over which that benefit is earned.

The models have a history inside the refinery

This extends a relationship that already reaches into operations. On 9 April, Honeywell announced connected services and performance monitoring for several of Dangote’s core processing units. The arrangement combined digital tools with UOP engineers’ expertise to identify possible problems and recommend operating changes.[2]

That earlier announcement also included operator training simulators built from digital twins of the refinery. Their purpose was to let staff rehearse realistic situations and test actions safely. The model of a plant can therefore serve a training purpose as well as helping engineers analyse operation.[2]

The new platform includes NVIDIA’s PhysicsNeMo framework among its tools.[1]

NVIDIA describes PhysicsNeMo as a framework for models that combine physical relationships with simulation and observed data. Such surrogate models approximate the behaviour of a physical system, allowing engineers to explore its response more quickly. The framework supports creating and validating digital twins across domains such as fluid flow and structural mechanics.[4]

A useful refinery model needs to represent the connections between processes: improving the output of one unit is of limited value if another unit cannot handle the resulting flow. That is why the economic question is about the performance of the connected plant, rather than the quality of a single prediction.

An expansion changes the baseline

A separate, larger project predates this AI launch. In November 2025, Honeywell announced that Dangote had selected its process technology, services, catalysts and equipment for an expansion from 650,000 barrels a day of capacity to 1.4 million. That agreement also aimed to widen the range of crude the refinery could process and make greater use of existing equipment.[3]

New physical capacity and better operation can both change a refinery’s results. If the plant processes more crude after an expansion, a higher total output alone will not isolate the contribution from AI. The comparison needs to account for what equipment was available and how it was being used.

Honeywell’s estimate expressly depends on refinery configuration, crude mix, utilisation, existing performance, data quality, product margins and implementation.[1]

Those dependencies connect the software to the business. A change in crude or product prices can alter the value of an operating improvement, even if the recommendation itself stays the same. The useful comparison will track performance under comparable inputs and operating conditions as the tools enter use. That is how a preliminary per-barrel estimate can become an operating result against which to judge the investment.

Report an error or suggest a correction