Honeywell Technologies launches Forge Plant Cognition at Dangote refinery with NVIDIA and Dell
- Forge Plant Cognition debuts at Dangote Petroleum Refinery in Lekki, one of the world’s largest single‑train refineries.
- Honeywell estimates AI-enabled optimization could deliver more than $0.29 per barrel, equating to over $60 million gross benefit per year at full throughput.
- The solution is built with NVIDIA AI models and Dell AI Factory infrastructure, using digital twins, vision AI, and physics‑based models for optimization and anomaly detection.
- Honeywell and Dangote plan to raise refinery capacity from 650,000 to 1.4 million barrels per day within three years and target a workforce that is 70% Nigerian over time.
Overview
Honeywell Technologies has launched Forge Plant Cognition, an industrial AI platform aimed at refining, petrochemical and LNG operations. The platform combines process expertise, industrial automation and AI to identify constraints, anticipate issues and convert insights into operational actions rather than just recommendations.
Technology and partners
The solution uses Dell AI Factory infrastructure and NVIDIA open models and libraries: agents powered by Nemotron, Omniverse libraries for digital twins, Cosmos/Metropolis for vision AI, and the PhysicsNemo framework for physics‑aware models and optimization workflows. Dell Services supports deployment.
Deployment at Dangote
The platform is launching first at Dangote’s Lekki refinery to integrate operating data with Honeywell UOP process knowledge, accelerate troubleshooting and support operator decision‑making. The deployment also aims to develop local engineering capability and help Dangote reach a 70% Nigerian workforce as it expands.
Anticipated benefits
Preliminary estimates indicate potential operational benefit of more than $0.29 per barrel, which at full capacity translates to over $60 million gross per year for the facility; that estimate is gross of implementation costs and dependent on refinery configuration, operating conditions, data quality and other factors.
Source: Honeywell