Field Proven AI/ML Solutions, Next Step in the Automation Space?

The amount of data which is generated and available during drilling operations has been constantly increasing over the last years. With this trend, Data Driven solutions and especially AI/ML solutions have become a more important and valuable tool for a rig crew to use in their operations. There are a few reasons for that. Firstly, these solutions manage to work with the data which is already readily available, of its quality, latency and frequency. Secondly, AI/ML solutions can predict outputs based on the trends in the data and solve the problems where physical models are limited or hard to implement. Thirdly, with the problem of Big Data still being not always properly structured, defined or available, targeted AI/ML solutions (solutions aimed to solve a particular problem in the well, e.g. stuck pipe, hole cleaning) can be more suitable for operations than complex models. Finally, targeted data driven models can benefit over physical models in automation systems as they are self-contained within the rig control system, not requiring well engineering or configuration data from outside.

Real time data has been used to create risk awareness of hole problems during operations and provide drilling optimization recommendations to rig teams over a hundred wells. Implementation of AI/ML solutions on these wells has demonstrated the advantages of using Data Driven solutions and their relevancy for drilling automation. The recommendations and warnings are given in real time in a timely manner for the team (and for the rig control system in case of drilling automation) to be able to respond and take corrective action. The use of available standards (such as WITSML) and cloud applications simplify their use and support, making the software available globally to various teams, including remote operating centers. Targeted AI/ML solutions can deliver outputs with a low number of false positives which is an important pre-requisite for automation.

To achieve drilling automation AI/ML solution must be consistent in its recommendations and proven in a field. A rig team should be able to trust the solution to provide the best results (and the same applies to automation systems) even without a full overview of its functionality. These steps are already achievable today. Self-corrective AI/ML solutions, such as ROP optimization, are more qualified for automation than classification models which can provide false classifications. Such classification models where decision mechanism is built in would require some time to refine and verify imperfect human decision rules into safe guidelines based on operational experience. All content contained within this webinar is copyrighted by Serafima Schaefer and its use and/or reproduction outside the portal requires express permission from Serafima Schaefer.

SPE Webinars are FREE to members courtesy of the