Autonomous control has reached a new milestone at Aramco’s Fadhili gas plant

In a development that marks a significant step toward industrial autonomy in the energy sector, Aramco and Yokogawa Electric Corporation have commissioned multiple autonomous control AI agents at the Fadhili Gas Plant in Saudi Arabia

The deployment targets the acid gas removal (AGR) unit - one of the most dynamic and energy-intensive sections of a gas-processing facility - and demonstrates how reinforcement-learning-based control can enhance efficiency, stability, and sustainability in large-scale operations.

At the core of the installation is Yokogawa’s Factorial Kernel Dynamic Policy Programming (FKDPP) algorithm, a reinforcement-learning (RL) method co-developed with the Nara Institute of Science and Technology. FKDPP enables AI agents to derive optimal control strategies without predefined models, allowing them to handle multivariable interactions, nonlinear behaviour, and fluctuating process conditions. Unlike classical model-predictive control or PID strategies, FKDPP agents can adapt to scenarios not explicitly included during training, giving them the robustness required for mission-critical energy applications.

SAFETY-CENTERED IMPLEMENTATION

Given the AGR unit’s sensitivity and the need to maintain strict safety margins, Yokogawa implemented the solution in three structured phases. The project began with the development of a high-fidelity plant simulator to train the AI agents and evaluate their responses to a wide range of process disturbances. This virtual environment allowed engineers to validate the agents’ stability, boundary-handling behaviour, and ability to meet operational constraints without exposing the live plant to risk.

Once the agents demonstrated reliable performance in simulation, Yokogawa proceeded with a stepwise rollout across various subsections of the AGR unit. Ultimately, the agents were integrated with the existing Centum VP distributed control system, ensuring that the underlying safety and interlock functions of the plant remained fully intact.

MEASURABLE EFFICIENCY GAINS

Preliminary results from the Fadhili Gas Plant show compelling performance improvements. The AI-driven optimisation has reduced amine and steam consumption by 10–15%, while overall power usage has dropped by roughly 5%. Operators report improved process stability under varying ambient conditions, along with a substantial reduction in manual interventions. These gains are especially notable in AGR systems, where feed composition variability and thermal effects often challenge conventional control strategies.

Beyond cost savings, reductions in steam and power usage directly support Aramco’s broader decarbonisation initiatives by lowering the facility’s energy footprint. Less amine degradation and improved thermal stability also contribute to lower maintenance demands and enhanced long-term reliability.

A PATH TO INDUSTRIAL AUTONOMY

For Aramco, the deployment is part of a wider effort to expand industrial AI across its asset base.

“Aramco has embarked on an ambitious plan to unlock value by deploying a wide range of industrial AI applications across our operations,” says senior vice president of Aramco Engineering Services, Khalid Y. Al Qahtani. “The collaboration with Yokogawa is one of many initiatives that focus on improving efficiency, enhancing sustainability, and generating more value for our shareholders. It reflects how the company is harnessing advanced technology, including AI, to elevate its performance and reinforce its position as a technology leader in the energy sector. We look forward to building on this important milestone, as we explore further adoption of cutting-edge solutions that will contribute to a new era of industrial innovation.”

Yokogawa’s leadership echoed these sentiments, highlighting the project as a flagship example of its Industrial Automation to Industrial Autonomy (IA2IA) vision.

With the Fadhili project, autonomous control has reached a new level of maturity in the energy sector. For engineers and decision-makers, the deployment offers a blueprint for how AI-driven control – when supported by simulation, rigorous validation, and integration with established DCS frameworks - can unlock measurable performance improvements in complex process environments.