FEED stands for Front End Engineering and Design. FEED is intended to substantiate the plant's guaranteed functionality and reliability, the project timetable and budget, and identify risks and ways of mitigation. These points form the basis for the Final Investment Decision (FID) on project viability.

Does FEED deliver what is expected?

The mega-projects statistics show that over 75% suffer from substantial cost overrun and delays (over 30%). The situation is even worse in the oil and gas industry. Clearly, the FEED methodology does not work in practice, constrained by a budget and time.

The reason for that may be found in the AACE Estimate Classes for the FEED costs accuracy prediction. It shows that moving from Estimate Class 3 (affordable for desalination companies) to Class 1 (needed for FID) increases the workload and complexity by a factor of 10. Normally, the Class 3 FEED preparation takes 4 – 9 months and costs US$ 1 – 5 million.

The difference between Classes 1 and 3 adds room for uncontrollable project scope creep.

The FEED budget and schedule are estimated based on the quotations valid for less than 60 days. As the span between requests for quotation (RFQ) and FID is substantially longer than the mentioned validity period, it adds to the FEED inaccuracy.

Being a historical golden standard for over 50 years, FEED has remained immune to new emerging challenges that plague the project-plant life cycle. They are described below.

Plant systems mismatch due to over- or under-design. The capacity variation between systems is mostly within 20%. It is a substantial reserve for decreasing the production costs as both Capex and Opex are subject to optimization.

Plant annual production prediction. Its rigid procedure is computationally intensive as it requires an hourly tracking of local conditions’ variation over the time of the year. Trying to implement it manually adds substantial risk to the PPP projects of missing the contract product price.

"What-if" scenario simulations. They shall address situations leading to capital items' replacement during O&M, prolonged off-design conditions coverage (like algae bloom or jellyfish inrush), and potentially unsafe operations identified by the HAZOP study.

Plant control algorithms validation and tuning. Today, it is executed during the plant operation. As this practice is inherently unsafe, it negatively affects the plant's operation and economics.

Plant alarms rationalization. It deals with the alarm settings re-validation and fine-tuning, and alarms prioritization and suppression. It is needed to stabilize the plant operation and make it safe. This process takes 2-3 years to accomplish, and a lot of trial and error.

Plant maintenance costs prediction. Maintenance cost is the biggest after the energy cost and capital cost. Despite its criticality for the PPP project's success, its rigorous prediction is beyond the engineering companies' expertise and computational capabilities.

Plant operator training. Today, operators are trained on a live plant, which is not acceptable at all. Equipment damage as a result of the operator's wrong action is not rare.

AI model training. The AI progress in desalination hinges on the availability of contextualized structured data covering all aspects of the plant design and operation. Building a perfect data source is a task beyond engineering companies' expertise. Under the circumstances, a digital twin may be successfully used to engineer, train, and tune AI models.

Crenger.com solves all the mentioned challenges by replacing the FEED methodology with the Digital-Twin one (DT). DT is defined as a digital emulation of the project-plant life cycle.

DT changes everything by removing the budget-time constraints. For instance, it takes about 50 man-hours to generate the FEED Class 1 data.

DT represents a fundamental shift from "static snapshots" to "living data" continuously updated till the end of the project-plant life cycle. DT eradicates the disconnect between the "as-designed" intent and the "as-built" reality. The "handover" problem vanishes. There is no massive pile of manuals to digitize later because the "manual" was built alongside the asset.