Regression-based finite element machines for reliability modeling of downhole safety valves

Colombo, Danilo; Lima, Gilson Brito Alves; Pereira, Danillo Roberto; Papa, João P. [UNESP]

Regression-based finite element machines for reliability modeling of downhole safety valves

Data

2020-06-01

Autores

Colombo, Danilo

Lima, Gilson Brito Alves

Pereira, Danillo Roberto

Papa, João P.

Tipo

Artigo

Resumo

Downhole Safety Valve (DHSV) stands for a device widely used in offshore wells to ensure the integrity and avoid uncontrolled leaks of oil and gas to the environment, known as blowouts. The reliability estimation of such valves can be used to predict the blowout occurrence and to evaluate the workover demand, as well as to assist decision-making actions. In this paper, we introduce FEMaR, a Finite Element Machine for regression problems, which figures no training step, besides being parameterless. Another main contribution of this work is to evaluate several machine learning models to estimate the reliability of DHSVs for further comparison against traditional statistical methods. The experimental evaluation over a dataset collected from a Brazilian oil and gas company showed that machine learning techniques are capable of obtaining promising results, even in the presence of censored information, and they can outperform the statistical approaches considered in this work. Such findings also investigated using uncertainty analysis, evidenced that we can save economic resources and increase the safety at the offshore well operations.