During drilling phase, a huge quantity of data is collected in real-time every day: Logging Wwhile Drilling (LWD), gas data, drilling parameters and directional data. The last challenge of digital transformation is using this data to gain competitive advantages: this means using data both to mitigate risks, to increase operational efficiency, by improving processes, and to create new business models.
We propose a new approach where expert geologists collaborate with data scientists to develop a tool based on Machine Learning (ML) and Artificial Intelligence (AI) that creates repeatable and reliable solutions directed toward what can be called the accelerated "digitization" of the automation system — support operations geologist during drilling phases with reduction of the likelihood of errors related to "human factors".
The Eni propietary AI tools can be applied automatically while drilling and, now, consists of three different use cases:
Litho-Fluid interpretation, a set of AI algorithms used to identify in real-time the lithology and to interpret the formation fluids based on the data collected.
Virtual Logging, algorithms used to reconstruct the sonic curve from gamma ray and resistivity curves together with drilling parameters.
Well-to-well Log Correlation and look ahead, models used to find analogies between intervals of the well being drilled and the reference well, allowing to estimate the distance and time of arrival to a given geological event, such as the casing point.
These three models have been developed and tested in two different areas, characterized by a similar geological setting, to lay the foundations for a replicable standard in other areas. In the case of drilling a well in a different geographic area, but a geological analogue, the same models are re-trained on historical data related to that area.
Different techniques are used within each model: supervised and unsupervised machine learning algorithms, time-series and other advanced analytics algorithms are combined with site-specific business rules.
During the re-training of models for a planned well in different geographic areas, each step is shared between geologists and data scientists, in order to validate the models in both geological and statistical terms.
The results obtained have been remarkable in terms of accuracy.
The first feedback from operations geologists reaffirms the usefulness and expected benefits; the tools allow to improve well control and speed up some repetitive and time-consuming operations.
It therefore demonstrates how AI is not just a "buzzword" for the energy industry but translates into an increase in efficiency and value for the company, whenever it is possible to exploit the data available for business needs.
The AI tools described in this paper have proved to be valid and robust in different geographic areas, characterized by a similar geological setting, with a minimum set of wells already drilled; it is planned to widen the applications in new areas having the same geological background. Finally, the application of models directly on real-time data provides immediate support to geological operations.