What I have learned between the design office and the site

This is one of the pieces of work I have done: a roadmap for applying AI in the development programme of the Azadegan oil field, using an AI-Pattern approach, to raise productivity and reduce risk. Here is a short summary for orientation.

Today’s problem is operational ambiguity. Data is scattered across systems, ageing equipment makes root-cause diagnosis difficult, and decisions are often made after the event. The goal is a dependable pattern: a data foundation and tag standardisation, real-time monitoring and decision support for operations, and continuous learning from feedback and results.

Four objectives, one success criterion: measurable operational value. Higher recovery through data-driven optimisation of production, reservoir and water injection; lower operating cost (OPEX) by reducing unplanned shutdowns and reactive maintenance; lower development risk through better decisions on new packages; and greater safety by reducing people’s presence in high-risk locations. The rule was that every use case must be tied to an operational KPI, a business owner and a valid baseline.

Two environments, two logics. In the producing (brownfield) part, the aim is more production, less water cut and fewer stoppages; data exists but is scattered, and payback is shorter. In new developments (greenfield) the aim is low-risk design with lower life-cycle cost; there is less data but it can be designed right from the start. The proposed sequence: a fast pilot in brownfield, and a Smart-Ready requirement in greenfield.

One pilot, one owner, one decision. A pilot on one selected production unit (GOSP), with a clear KPI and baseline, in three use cases: predictive maintenance of pumps, compressors and rotating equipment; forecasting production decline and water cut to prioritise engineering action; and anomaly detection linking real-time data to operator and maintenance routines. A 20 percent reduction in downtime was the proposed target, to be confirmed after the baseline is measured.

New development Smart-Ready from day one. Four layers: reservoir decisions and the field development plan (FDP) with development scenarios and optimal well placement; well design and drilling with risk prediction and reduced non-productive time (NPT); facilities and instrumentation with sensor placement, data standards and remote-operation capability; and at the centre, a Digital Twin for monitoring, simulation and continuous improvement.

Four implementation phases: data foundation (inventory of SCADA, Historian and CMMS, tag standardisation, security and access); fast pilot (GOSP selection, model implementation, validation with operators against the baseline); integration into new development (Smart-Ready requirements in EPC and design-led models); and sustainment and learning (model performance monitoring, change management, training and expanding use cases).

The decision I proposed at the end was simple: form a transformation task force and select the pilot GOSP. And the boundary I kept throughout is the one written on the AI page: AI strengthens the engineer’s decision; it does not replace engineering ownership and judgement.

A wrong decision can be corrected, because at least it has made the path clear. A decision that is never made silently removes one option from the table every week.

On the critical path, postponing a decision is itself a decision: the decision to let time choose for us. By the time the meeting finally happens, there are often no longer two options, only one — and it is the most expensive.

The question I ask before every weekly meeting is simple: which decision, if not made by the next meeting, costs us one of our options? That one is the first item on the agenda.