From wells to FPSOs: where AI is entering offshore operations

Artificial intelligence appeared across several discussions at ROG.e 2026, but some of the most relevant examples were tied to specific offshore and engineering applications.

Petrobras presented the SAPIENS BOT, developed to support the classification of oil production-loss events. According to the project team, E&P operations generate around 6,900 loss events per year, with more than 1,000 possible combinations of causes.

The system combines machine learning with a large language model trained around Petrobras’ classification rules. In the Espírito Santo operating unit, the team reported 92% accuracy, an 81% reduction in processing time per event and 60% fewer reclassifications.

Human validation remains part of the process: when the two models disagree, the specialist receives both recommendations and makes the final decision.

Well engineering was another area discussed at the event.

The SAVIA project uses intelligent agents to automate part of the complexity assessment performed during well design, while Cortex integrates information from more than 15 systems and databases related to well construction.

According to ANP, Cortex achieved 77% accuracy in complex queries and provides traceability to the original documents used in its responses.

AI and digital tools are also being applied to asset integrity.

Ativo360, developed with Petrobras and Tecgraf, combines digital twins, predictive models, computer vision, 3D models and immersive inspection. The platform is already used across more than 120 units, according to ANP.

In subsea, Petrobras and CESAR also presented work involving digital twins for rigid-riser fatigue and fatigue and CO₂ stress-corrosion cracking in flexible pipes, combining operational data with physical models and automated simulations.

FPSO operators are following a similar path.

During ROG.e, MODEC Global Digital & Analytics Director Rei Yasumuro said the company’s Lighthouse predictive-maintenance platform has helped maintain fleet uptime above 98%, reduce downtime by more than 30% and deliver returns of up to 25 times the investment.

These cases show different approaches: generative AI, machine learning, engineering rules, historical databases, computer vision and digital twins.

They also leave an important question for the offshore industry:

As these systems move closer to decisions involving wells, production and asset integrity, how should responsibility be divided between automation, engineering validation and human decision-making?