TY - THES

T1 - Application of Machine Learning Techniques for Rate of Penetration Optimization Analysis

AU - Tekum, Peter Mbah

N1 - embargoed until null

PY - 2020

Y1 - 2020

N2 - Reducing well construction cost per foot can be executed by several methods, focusing just on drilling performance and efficiency, which can be evaluated mainly by drilling speed, thus, a slight increase in average Rate of Penetration will definitely improve the overall drilling performance and efficiency and consequently, remarkably decrease in effective drilling time and reduction in overall cost can be achieved. Therefore, Rate of penetration (ROP) optimization has been a common research interest in the drilling discipline within the petroleum industry, being a very important topic especially during periods of low oil price, increased steel price and necessity to apply acknowledged expensive technologies. The ROP efficiency enhancement can be achieved through many different efforts, such as good well path design, optimum operational parameters, as well as implementation of novelty and innovative new technologies. In this context, ways of enhancing ROP while drilling has been an understandable practical and cost-effective method, and so has been in evidence for the last decades, covering research resulting in different approaches, models and methods, with reasonable application in the industry, but as evaluated, with rooms for improvements. Conceptually, many different ROP models can be found in the literature, which are mostly relates empirical models with specific coefficients that translate actual specific scenario to a mathematical equation, which defined, starts representing the relation between the drilling parameters and the ROP response that happened, allowing in seeking for set of parameters that may maximize ROP. Due to the many empirical coefficient and functional constraints involved for fitting, the representations have shown to have rooms for improvement, where novelty approaches have been valuable. In this perspective, this thesis details the research developed aiming an alternative method to predict and to maximize ROP based on drilling mechanics data, supported by machine learning techniques, which drove the attention since showed to be more robust, precise, and so a better way to mathematically model ROP response from drilling operations in combination to specific key operational parameters. This improvement when compared to previous models used was achieved by applying the Artificial Neural Network (ANN) and Random Forest techniques, followed by appliance of the Genetic Algorithm (GA) technique, all supported by the software and developed script in Matlab. Two models have been developed and were validated using real data. During the validation phase several cases were studied and noticeable conclusions have been reached, one of the most important is that the performance of the predictive models are function of several hyper parameters which have to be adjusted carefully to improve the efficiency of the models. The other important observation was that, the flow rate has significant impact on ROP optimization and must be considered when performing Drill Off Test.

AB - Reducing well construction cost per foot can be executed by several methods, focusing just on drilling performance and efficiency, which can be evaluated mainly by drilling speed, thus, a slight increase in average Rate of Penetration will definitely improve the overall drilling performance and efficiency and consequently, remarkably decrease in effective drilling time and reduction in overall cost can be achieved. Therefore, Rate of penetration (ROP) optimization has been a common research interest in the drilling discipline within the petroleum industry, being a very important topic especially during periods of low oil price, increased steel price and necessity to apply acknowledged expensive technologies. The ROP efficiency enhancement can be achieved through many different efforts, such as good well path design, optimum operational parameters, as well as implementation of novelty and innovative new technologies. In this context, ways of enhancing ROP while drilling has been an understandable practical and cost-effective method, and so has been in evidence for the last decades, covering research resulting in different approaches, models and methods, with reasonable application in the industry, but as evaluated, with rooms for improvements. Conceptually, many different ROP models can be found in the literature, which are mostly relates empirical models with specific coefficients that translate actual specific scenario to a mathematical equation, which defined, starts representing the relation between the drilling parameters and the ROP response that happened, allowing in seeking for set of parameters that may maximize ROP. Due to the many empirical coefficient and functional constraints involved for fitting, the representations have shown to have rooms for improvement, where novelty approaches have been valuable. In this perspective, this thesis details the research developed aiming an alternative method to predict and to maximize ROP based on drilling mechanics data, supported by machine learning techniques, which drove the attention since showed to be more robust, precise, and so a better way to mathematically model ROP response from drilling operations in combination to specific key operational parameters. This improvement when compared to previous models used was achieved by applying the Artificial Neural Network (ANN) and Random Forest techniques, followed by appliance of the Genetic Algorithm (GA) technique, all supported by the software and developed script in Matlab. Two models have been developed and were validated using real data. During the validation phase several cases were studied and noticeable conclusions have been reached, one of the most important is that the performance of the predictive models are function of several hyper parameters which have to be adjusted carefully to improve the efficiency of the models. The other important observation was that, the flow rate has significant impact on ROP optimization and must be considered when performing Drill Off Test.

KW - Drillability

KW - Optimization

KW - Real-time

KW - Machine Learning

KW - Drill-rate test

KW - Bohrbarkeit

KW - Optimierung

KW - Echtzeit

KW - Maschinelles lernen

KW - Verstärkungslernen

KW - Drill-Rate-Test

M3 - Master's Thesis

ER -