Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Detection of Subsea Gas Leakages via Computational Fluid Dynamics and Convolutional Neural Networks
Submission Author:
Gustavo Luís Rodrigues Caldas , RJ
Co-Authors:
Gustavo Luís Rodrigues Caldas, Thiago Bento, Roger Matsumoto Moreira, Maurício Bezerra de Souza Júnior
Presenter: Gustavo Luís Rodrigues Caldas
doi://10.26678/ABCM.COBEM2021.COB2021-2152
Abstract
Gas leakage can generate a considerable number of losses in environmental and economic terms. Constant monitoring of deep-water wells is an important safety tool to enable fast and proper responses, prevent the conditions' aggravation, and avoid unnecessary costs. A monitoring approach is to install video cameras in the subsea environment to track possible leaks. Automation of this process is possible by using image analyses algorithms. In this context, convolutional neural network (CNN) is a methodology part of the Artificial Intelligence framework, capable of extracting relevant features from images. The objective of this work is to develop a tool to detect fault in subsea pipelines using the presence of bubbles in the images as an indicator. Computational Fluid Dynamics (CFD) simulations were developed aiming to reproduce subsea gas leakages. The volume of fluid (VOF) method is employed to model the two-phase gas-liquid flow, in which bubbles are released into stagnant water with several velocities and from different pipelines orifice diameters. The frames obtained from the simulation images served as input to the CNN. The methodology is intended to distinguish between a scenario of normality (no leakage/bubble) and abnormality (leakage/bubble). The classification task of bubble presence or absence performed by the CNN reported high accuracy (99 %). No false alarms (fall-outs) in all sets, high specificity (true negative rate) and precision of the classifier were found. Good accuracy supports its potential as fault detector. It could be extended to other applications in the field of Fault Detection and Diagnosis, not limited to the present scope.
Keywords
gas leakage, Convolutional neural networks, Subsea Pipelines, CFD, Fault Detection
DOWNLOAD PDF VIEW PRESENTATION
