Machine learning delivers instant benefits to Elgin/Franklin wells

Dec. 1, 2021
Machine learning is helping TotalEnergies improve drilling returns at the Elgin/Franklin fields in the UK North Sea.

Machine learning is helping TotalEnergies improve drilling returns at the Elgin/Franklin fields in the UK North Sea. The development is one of the earliest results of the company’s group-wide digital transformation campaign, which started in 2018. Elgin/Franklin, an HP/HT gas-condensate field which came onstream in 2000, was one of the first in the company’s portfolio to be put forward for digital initiatives.

Daniel O’Leary of TotalEnergies E&P UK, presenting at the Digital Drilling session at the SPE’s Offshore Europe virtual event in September, said that in recent years, drilling through the field’s Late Cretaceous interval was close to the technical limit in the 12 1/2 and 8-in. section. The mud weight window was so tight, “it was a tightrope between taking gains and going onto losses.” Managed pressure drilling was critical for providing over-balance, as was managing losses with cement reinforcement, with on-bottom ROP accounting for 20% of the time spent drilling these sections.

Team leader Todd Brian had identified ROP optimization as a key digital opportunity for the Central North Sea drilling/wells department, along with an online reporting tool. There was no consistently applied method in place. Typically, the team would read old reports on prior ROP experience; post the results on large offset cheat sheets on a wall to show what worked in each formation; seek advice from service companies and vendors; combine and paste the input into a drilling program for supervisors to take offshore; and during drilling, start light, increase slowly, then go up to maximum. Ultimately effective, but highly labor-intensive and time-consuming.

There might be a scenario, O’Leary added, where the bit is dying in a formation, with mud logs on screen clearly showing that ROP has decreased, with 200 m (656 ft) still to go to T/D. “You can’t set shallow, so you might then ask what benefits a new bit could bring, and how can these be qualified/visualized?” But vendors have a commercial bias. So, TotalEnergies would conduct its own cost benefit analysis on whether to proceed to T/D or trip out, based on Excel calculations, creation of data sheets and new files, with the data input manually: all of which takes up time that might have been spent starting the trip out of hole.

This latter issue was the first to be addressed via the new digital well activity reporting portal (WARP), launched in March 2020. ROP optimization was next on the agenda. A team of data scientists, led by Brian, were tasked with devising a digital solution: they selected machine learning, with its capability to capture complex relationships and the fact that it does not require specific knowledge of drilling parameters. They selected a boosting tree algorithm, XGBoost, as this can be used for classification or regression, to deliver a numeric target variable, supported by previous mud logging well data supplied by TotalEnergies’ real-time support center. The resulting optimizer model was trained on data from 10 Central North Sea wells and evaluated on a further two. It is designed to predict ROP from the controllable inputs (drilling parameters) with the bit and formation fixed. Each input passes through the model, with the combination retained that indicates the highest ROP.

To ensure optimized ROP at any time, a grid search is conducted with new drilling parameters applied, again with the bit fixed in the formation, prior to applying the ROP model. For Elgin/Franklin, O’Leary explained, the currently controllable parameters are used as a reference with a grid 10-100% above and below the value created, with around 8,000 combinations of variables passing through the model to determine which combination could deliver the highest ROP.

The initial trial in 2020 was on Franklin well F13 in the Late Cretaceous interval, with PDF cheat sheets drawn up for the formation. Ranges derived from the model for each of the parameters were much tighter than would previously have been specified. Targeted ROP values were also drawn up along with performance indicators – what level of ROP would be considered good in this interval – and recommendations on bit type and operational advice, all based on real data. The sheets were issued to the offshore team to incorporate into their work instructions.

Results were positive. In F13’s 12½-in. section there was an improvement in drilling performance compared with both the previous well drilled and with average performance, with evident potential to optimize performance further, using data from the real-time support center. Optimized parameters were selected at each moment, and there were no losses in this section. But the main achievement, according to O’Leary, was the successful adoption of a new method by the offshore and onshore teams, with none of the ambiguity of the previous trial and error approach.

At the same time, using guideline pdfs was not exactly digital. So, earlier this year, a Cloud-based solution was adopted for the Elgin well E19’s 17½-in. and 12½-in. sections. This solution created an ROP optimization module in the WARP reporting tool; a dashboard showing recommendations based on real-time data from the mud logging unit, with target ranges displayed on a live progress chart; a forecaster that refreshes every 15 minutes, showing times to T/D and the predicted ROP, the optimal ROP and the historic means; and a bit trip module based on known tripping speeds. The capability also exists to add a BOP test to determine whether a bit trip would be beneficial or not. To work online, the real-time data-fetching, optimizer and forecaster are all implemented on the Cloud Store line with prepared data stored on the SQL database, live log data being retrieved every 15 minutes.

Now the wells team spends less time collating data upfront, performing analyses, or experimenting with parameters offshore. And TotalEnergies E&P UK has taken ownership of ROP optimization, O’Leary stressed, making it less reliant on third-party input and advice. At any time, users can provide feedback that can lead to continuous improvement. TotalEnergies now plans to extend the technology to other UK offshore operations and globally, starting with the Alwyn complex in the northern North Sea, followed by Qatar. In addition, WARP has been integrated into the company’s DrillX platform, a new Cloud-based center for smart operations, with further capabilities such as kick prediction/detection.