Saudi ArabiaEnergy and utilities2025 – ongoingSaudi Aramco
Reinforcement-learning control of Fadhili acid gas removal
Yokogawa reported 10% to 15% lower amine and steam use after AI took control of Fadhili gas treatment
Problem
Fadhili's acid gas removal process required chemical inputs and energy. Operators also intervened manually as process conditions changed.
Approach
Yokogawa introduced coordinated reinforcement-learning agents in stages to control the process. It trained the agents in a plant simulator before connecting them to the existing CENTUM VP control system.
Results
Takeaway
The agents operated through the plant's existing control infrastructure and safety functions. Detailed evaluation was still underway when the initial savings were announced.
Categories
- Size
- Large enterprise
- Industry
- Energy and utilities
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- OptimizationAI agents
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Yokogawa described autonomous AI control at Fadhili and the simulator-based preparation. Initial results showed lower chemical and energy use while detailed evaluation continued.
Full details are in the original sources.
