Banca de DEFESA: ANNY VIRGINIA SOUZA DE LIMA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : ANNY VIRGINIA SOUZA DE LIMA
DATE: 15/02/2023
TIME: 09:00
LOCAL: Programa de Pós-Graduação em Engenharia Civil
TITLE:

Application of Machine Learning Techniques and Sensitivity Analysis of Injection Pressure in Reactivation of Geological Fault Zones Scenar

KEY WORDS:

Fault reactivation, Sensitivity analysis, Injection Pressure, Linear Discriminant Analysis, Artificial Neural Networks

PAGES: 86
BIG AREA: Engenharias
AREA: Engenharia Civil
SUMMARY:

Geological faults are common structures in oil reservoirs that can act as channels facilitating flow or as sealing barriers. During hydrocarbon exploration, the pressure within the reservoir varies, causing a mechanical response in the medium that can lead to the phenomenon of fault reactivation. This occurs when rock deformations re-open the fault, increasing its permeability and allowing fluid flow. In this work, a sensitivity analysis was carried out to study the influence of different injection scenarios on deformation, shear stress, permeability, liquid pressure and fluid flow in geological fault zones. To do this, the CODE_BRIGHT software was used, a simulator based on the finite element method, where the hydromechanical coupling is done in an implicit way and the elastoplastic model used to represent the mechanical behavior of the fault was the Drucker Prager model. In addition, it was possible to implement machine learning algorithms such as Linear Discriminant Analysis (LDA) and Artificial Neural Network (ANN) using the Mohr Coulomb analytical model to classify scenarios of reactivation of geological faults. These models were implemented in Python programming language with libraries already established in the literature for the mentioned methods. The results of the sensitivity analysis showed that increases in injection pressure are responsible for fault reactivation, associated with increasing plastic deformations and the development of pore pressure and fluid flow in the length of the fault. A limit was found for this injection pressure, which would not reactivate the fault. Higher injection pressures can turn the fault into a conduit that can lead to fluid loss and reservoir depressurization, among other problems related to fault reactivation. The fault reactivation classification models had excellent performance for both LDA and ANN and can be a preliminary approach to evaluate reactivation scenarios. They can also incorporate numerical solutions and laboratory data on geomechanical characterization to increase the complexity and generality of these techniques.

COMMITTEE MEMBERS:
Presidente - 1130915 - BERNARDO HOROWITZ
Externo à Instituição - JONATHAN DA CUNHA TEIXEIRA - UFAL
Externo à Instituição - MANOEL PORFIRIO CORDAO NETO - UnB