AI Algorithm Helps Optimize AGR, and Save Amine, Steam and Power

Key Takeaways

• Yokogawa AI cut amine and steam use by 10‑15% at Fadhili
• Power consumption dropped roughly 5% after autonomous AGR control
• FKDPP cut craft beer fermentation 28%, from 336 to 240 hrs
• AI agents trained in simulation, then integrated with Centum VP for safety

Pulse Analysis

The petrochemical sector is increasingly turning to advanced machine‑learning techniques to push the limits of traditional control loops. Yokogawa’s FKDPP algorithm, a reinforcement‑learning framework co‑developed with NAIST, creates autonomous agents that can predict optimal set‑points for acid‑gas‑removal units. By first training these agents in a high‑fidelity digital twin, the company ensures that safety constraints are respected before any live deployment, and the seamless hand‑off to the Centum VP system preserves existing alarm and interlock logic.

At Aramco’s Fadhili plant, the AI‑driven AGR optimization delivered tangible resource savings: amine solvent and steam usage fell by up to 15%, while electricity demand dropped about 5%. These reductions not only lower fuel and utility bills but also shrink the plant’s carbon footprint, a growing priority for both regulators and investors. Moreover, the improved process stability lessens the need for operator intervention, freeing skilled staff for higher‑value tasks and reducing the risk of human error.

Beyond oil and gas, the same FKDPP methodology proved its versatility in a brewing context, where a temperature‑schedule generated by the algorithm cut fermentation time by 28%. This cross‑industry success underscores how generative AI and robust data governance can unlock efficiencies in any continuous‑process operation. As more firms adopt similar AI agents, the industry can expect faster innovation cycles, tighter integration of structured and unstructured data, and a new baseline for operational excellence.