Petroleum engineers require rapid, reliable insights to detect and interpret production changes, ranging from deliberate operational adjustments like choke modulations to subtle, reservoir-driven shifts that may signal critical anomalies.
Traditional methods, reliant on complex, handcrafted metrics and error-prone reporting, demand constant maintenance and often miss subtle changes.
On Wednesday afternoon (December 03), at the Dig X Subsurface 2025 conference, Lukas Mosser, Advanced Data Scientist at Aker BP, will unveil how Vision-Language Models (VLMs) transform production time-series monitoring.
Aker BP’s ‘Well Informed’ platform uses Vision-Language Models (VLMs), AI that interprets time-series plots with text prompts, to generate clear, structured reports that flag system changes based on risk levels, enhancing decision-making.
Tested on 1,000 labeled time series, with one-third anomalies, VLMs achieved a 0.94 recall for detecting changes, without manual feature engineering.
Aker BP sees this novel VLM application as a first step toward eliminating maintenance-heavy, handcrafted features, paving the way for smarter, adaptive production monitoring.
Integrated into Aker BP’s North Star Digital Architecture, this AI-driven solution detects undefined errors and slashes maintenance needs.
The program can be found on the conference website.