DrillX Kick Prediction: Machine Learning to prevent hydrocarbons kicks
1st Prize - Best Innovators 2020
A project presented by Vincent Milville, Nicolas Baudouin, Xavier Pujalte, Thibault Forget and Ridha Zegdane
The Kick Prediction solution from DrillX uses artificial intelligence (AI) to predict the risk of kicks in real-time up to six hours in advance, via a web application directly accessible by the operational teams on the rigs. This impressive high-tech tool is unique in the Oil & Gas industry.
Artificial Intelligence (AI) in Drilling, A First
Uncontrolled hydrocarbon kicks that occur during drilling can end in tragedy. Predicting the risk of kicks is a core concern for the industry and the drilling metier and has always been difficult to achieve until today. Here comes DrillX, the digital transformation program of Total Drilling & Wells metier, and the possibilities offered by AI models.
In April 2019, the DrillX team started to develop a machine-learning based predictive solution: Kick Prediction. This tool helps our supervisors on platforms to anticipate and mitigate in real-time the risk of kick during drilling operations, assisting them in their decision-making process. This multi-disciplinary project combines our internal know-how in data science, IT, drilling and geoscience.
The AI engine consists of five different and independent machine learning models. Each of these sends its calculations to a meta-model that delivers a final kick risk level. This risk level is accompanied by indicators that give users a real-time understanding as to which parameters contribute the most to the calculated risk.
79% accurate kick prediction
It took only four months to develop the solution in Agile mode, code the user interface and deploy the pilot in the North Sea in July 2019. The model has been trained over a long period using a set of data covering 160 drilling sections, with 25% of these including a kick, and enriched by data transformations that recreate the different signals usually monitored by supervisors during operations. This phase has enabled the solution to learn to recognize weak signals that can indicate the probable future occurrence of a kick, in the Group’s various drilling environments (onshore, offshore, deep water etc.). Once deployed and connected to real-time data streams coming from the drilling rigs, the solution detects anomalies and assigns them levels of risk in accordance with the acquired learning model based on the training data.
The system is able to successfully predict 79% of kicks, six hours in advance, a first in the industry for these complex events.
With the Kick Prediction solution, TotalEnergies has a tool that is unique among its competitors and will help to improve the safety of its operations and reduce their costs by minimizing the number of kicks. There is no doubt this clear competitive advantage will benefit TotalEnergies and strengthen its image as an innovative major.
A promising deployment and a future generalization on platforms
In August 2019, the Kick Prediction solution was connected via innovative IT architecture to real-time data streams from drilling rigs connected to the Real-Time Support Center (RTSC), our remote operations monitoring center. Collected by sensors already in place on the drilling rigs, the data is sent in real-time to the RTSC and then retrieved within DrillX cloud-based environment, where it is processed by the AI model. The resulting predictions are then sent within a few minutes to our on-site supervisors on a web application accessible securely from anywhere using only a standard web browser.
The Kick Prediction solution has already been used by four affiliates since its deployment and has accurately predicted a kick offshore Congo. By the end of 2020, the solution will be rolled-out to all our operated drilling rigs connected to the RTSC.
Drilling & Wells
Artificial intelligence to shape the future of drilling
Drilling & Wells
Drilling Wells Safely & Cost-Effectively
Drilling & Wells
Enhance Well Risk Management