Aramco Yokogawa Fadhili: Autonomous AI Cuts Steam 15%

The Aramco Yokogawa Fadhili autonomous AI deployment has delivered a 10–15% reduction in amine and steam consumption and approximately 5% lower power draw at Saudi Aramco’s Fadhili Gas Plant in the Kingdom of Saudi Arabia. The results, reported by Yokogawa Electric Corporation, represent the first commercial-scale commissioning of multiple autonomous AI control agents on an acid gas removal (AGR) unit in the oil and gas sector.

FKDPP: The Reinforcement Learning Algorithm Behind the Deployment

The technology core is the Factorial Kernel Dynamic Policy Programming (FKDPP) algorithm — a reinforcement learning system jointly developed by Yokogawa Electric and the Nara Institute of Science and Technology (NAIST). FKDPP uses multiple coordinated AI agents that make control and optimisation decisions autonomously, without human input, and are designed to maintain stable operation even under previously unseen process conditions.

At Fadhili, the agents were deployed specifically on the AGR unit, which removes hydrogen sulphide and carbon dioxide from raw gas before downstream processing. Amine-based AGR systems are energy-intensive: amine regeneration requires significant steam input, and process instability drives both chemical consumption and off-spec output. The FKDPP agents target both variables simultaneously, adjusting control parameters in real time to minimise resource use while holding process quality within specification.

Aramco Yokogawa Fadhili: Three-Phase Commissioning Approach

Yokogawa implemented the solution across three sequential phases, each progressively extending autonomous control into a wider section of the AGR unit. Before any live integration, the team constructed a detailed digital simulator of the Fadhili plant. AI agents were trained entirely within this simulated environment and validated against a range of operational scenarios before being introduced to the live process — a precaution standard in high-consequence industrial deployments but rarely executed at this level of fidelity for reinforcement learning systems.

The final phase connected the AI agents to Yokogawa’s CENTUM VP integrated production control system, allowing the autonomous layer to operate within the plant’s existing safety and interlock infrastructure rather than alongside it. That integration point is commercially significant: it means the deployment does not require replacement of the underlying control architecture, lowering the barrier to replication at other Aramco facilities.

Operational Significance at Scale

Fadhili Gas Plant, located in the Eastern Province of Saudi Arabia, processes gas from multiple Aramco fields for domestic supply and export. Its AGR capacity is substantial — making the efficiency gains from the Aramco Yokogawa Fadhili deployment meaningful in absolute terms, not just percentage terms. A 10–15% reduction in steam consumption at a facility of Fadhili’s scale translates directly to reduced fuel gas consumption in the regeneration process and lower plant-level carbon intensity.

The deployment also reduces operator manual intervention in AGR control loops — a function that previously required continuous monitoring as ambient conditions shifted across the day. Freeing operators from continuous loop management allows reallocation to higher-level process oversight and exception handling.

Khalid Y. Al Qahtani, Aramco’s Senior Vice President of Engineering Services, described the Yokogawa collaboration as part of “an ambitious plan to unlock value by deploying a wide range of industrial AI applications across our operations,” citing efficiency, sustainability, and shareholder value as the three primary objectives.

IA2IA: From Automation to Autonomy

Yokogawa frames the Fadhili deployment within its broader IA2IA (industrial automation to industrial autonomy) strategic programme — a transition in which control systems move from executing predefined rules to making independent optimisation decisions based on real-time process data and learned operational experience.

The distinction matters commercially. Conventional distributed control systems (DCS) follow operator-configured setpoints and respond to deviations. Autonomous AI agents, by contrast, continuously reoptimise setpoints themselves based on a reward function — in this case, minimising amine and steam use while maintaining AGR output quality. The FKDPP algorithm’s ability to perform this optimisation under conditions not seen during training is the capability that makes it deployable in live industrial environments rather than confined to simulation.

Kunimasa Shigeno, Yokogawa’s President and CEO, confirmed the results “have already exceeded expectations” and that the deployment demonstrates Yokogawa’s readiness to extend autonomous operations to additional energy sector facilities.

For context on Saudi Aramco’s broader infrastructure strategy, see OGSN’s analysis of the Aramco Yokogawa Fadhili deployment alongside the Saudi Aramco Terminals: Ras Tanura, Yanbu & 2026 Exports profile. The trend toward AI-driven operational efficiency in oil and gas infrastructure is also examined in OGSN’s 2026 Strategic Trends Outlook.