Banca de QUALIFICAÇÃO: ANTONIO PAULO AMANCIO FERRO

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : ANTONIO PAULO AMANCIO FERRO
DATE: 30/11/2022
TIME: 14:00
LOCAL: Sala de aula do Laboratório de Computação Científica - LCCV
TITLE:

Predictive models for ROP as support for real-time optimization of operational drilling parameters in oil well drilling

KEY WORDS:

Drilling. Rate of Penetration. Machine Learning. Real-Time. Optimization.

PAGES: 68
BIG AREA: Engenharias
AREA: Engenharia Civil
SUBÁREA: Estruturas
SPECIALTY: Mecânica das Estruturas
SUMMARY:

The rate of penetration (ROP) is a parameter of great interest in real-time drilling optimization. A higher ROP reduces drilling time and can lead to an overall reduction in operational costs. ROP models correlate factors that influence the drilling process and thereby predict the resulting ROP. These models enable to find out optimal operating parameters such as RPM, WOB, and fluid flow rate in real-time throughout the drilling process. Obtaining more accurate ROP models is a difficult task, due to the large number of factors interacting non-linearly with each other. Furthermore, given the complexity of the drilling process, the influence of each parameter is not fully known. Analytical expressions for ROP and analysis in offset wells are traditional strategies for predicting ROP, however, they are limited to specific formations or the type of bit used. These difficulties led to a necessity to develop more robust and data-based models using machine learning. This work involves the study of different predictive models of ROP. Traditional models from the literature: Bourgoyne & Young and specific mechanical energy, and their adaptations for real-time analysis, are compared with machine learning models: artificial neural networks, support vector machines and random forests. Machine learning models are
used to capture complex patterns in data. It is proposed to compare the predictive capacity of the methods in an open dataset with 7 wells in the same region. Three benchmarks are used for comparison so traditional strategies that use offset well data are compared against a scenario similar to real-time analysis. It is expected to verify which models best capture the nonlinear pattern of ROP. It is expected to determine the best strategy with the used benchmarks, verifying possible increases or decreases on prediction accuracy in real-time. With a more accurate ROP model, it is also possible to determine variables that most affect ROP in a retro analysis, and so it is proposed to evaluate the impact of the model variables and analyze the consistency of the responses between the models 

BANKING MEMBERS:
Interno(a) - 1121075 - ALINE DA SILVA RAMOS BARBOZA
Interno(a) - 1121260 - EDUARDO NOBRE LAGES
Externo(a) à Instituição - CHARLTON OKAMA DE SOUZA - PETROBRAS