Skip to content

Control algorithms at the Fadhili plant: announced energy gains face lack of baselines and independent audit

Share
Control algorithms at the Fadhili plant: announced energy gains face lack of baselines and independent audit

Listen to this article

Read by Anchor

Operational reports released about Aramco’s Fadhili gas plant have reopened a highly sensitive file in the energy sector: to what extent can the claimed savings from industrial artificial intelligence be independently verified? Yokogawa, in partnership with Aramco, announced the launch of a suite of AI agents for autonomous control within the plant’s acid gas removal unit, citing early results that include a 10-15 percent reduction in amine and steam consumption, about a 5 percent cut in electricity use, and a decline in manual operator interventions despite changing surrounding conditions. However, as the announcement itself notes, these figures are still under detailed assessment and represent a technology-supplier claim that has not yet undergone neutral external audit.

The agents were deployed in three successive phases, with the models first trained on a digital simulator of the facility, then integrated with the production control system “Centum V” to ensure that approved safety standards remained in force. This application does not arise from a vacuum; it is based on a dynamic policy-programming algorithm with a core developed jointly by Yokogawa and researchers at Japan’s Nara Institute of Science and Technology, and published in the proceedings of the IEEE Automation Conference in 2018.Construction of the programmed reward functionBy embedding precise engineering knowledge of industrial processes, as affirmed by Professor Takamitsu Matsubara, this formed the basis that allowed the algorithm to manage a real distillation column at GSR for 840 continuous hours at the start of 2022, and to be officially adopted by Innos Materials in March 2023 to achieve a 40 percent reduction in steam and emissions amid external temperature swings of up to 40 C.

This documented engineering record gives the control algorithm a rare technical credibility, yet it does not by itself prove the validity of the savings percentages reported at the Saudi facility. The industrial landscape in the Kingdom has seen parallel announcements that lack independent verifiability, ranging from Aramco’s digital-leadership talk about the “Simafor” network in September 2026, which claimed realized value exceeding five billion dollars in tech investments without specifying a baseline or methodology, to Ma’aden’s claim of roughly 2 200 hours saved per month via a Microsoft Office-based assistant package without measuring the conversion of hours into actual productivity, and up to corrosion-monitoring of insulators with Corogen Radar, which explicitly stated that its results reflect the supplier alone and do not constitute an external examination.

This contrast leads us to strict evaluation criteria that separate a press release from a measurably engineered outcome, requiring four integrated elements: a clear temporal baseline against which improvement is measured, an explanation of the calculation methodology, third-party review by a regulatory, academic or independent operator body, and an operational record extending beyond the initial launch announcements.Absence of defined baselinesMakes it impossible to distinguish genuine operational efficiency from carefully chosen comparison periods or number rounding that favors the supplier.

This decisive standard changes the rules for operations managers and industrial control engineers at refineries, petrochemical plants and mines across the Gulf, Egypt and the wider region. It is no longer acceptable to sign contracts for autonomous-control agents based on isolated percentage figures; engineering procurement departments must now require documentation of reference measurement periods and data audits during hot summer months with peak loads, as well as training local staff to engineer reward functions within models to ensure performance reliability. Financial evaluators must also refrain from recording estimated savings as completed assets in budgets before neutral engineering audits are finished.

The real shift from traditional industrial automation to fully autonomous operating systems will not be measured by promotional numbers from tech firms, but by how ready industrial plants are to deploy measurement methodologies and invite neutral auditors to examine results. When savings figures withstand scientific methodology and independent audit, artificial intelligence moves from a marketing material to a reliable pillar of energy security and industrial production efficiency.

Don't miss the next story

Subscribe for updates