Sucker Rod Pump Failure Analysis using Machine Learning Algorithms

Research output: ThesisMaster's Thesis

Harvard

Hofmaninger, FA 2023, 'Sucker Rod Pump Failure Analysis using Machine Learning Algorithms', Dipl.-Ing., Montanuniversitaet Leoben (000).

APA

Hofmaninger, F. A. (2023). Sucker Rod Pump Failure Analysis using Machine Learning Algorithms. [Master's Thesis, Montanuniversitaet Leoben (000)].

Bibtex - Download

@mastersthesis{08edd2ee02ce46018a02af4a36a22600,
title = "Sucker Rod Pump Failure Analysis using Machine Learning Algorithms",
abstract = "The sucker rod pump system represents the main artificial lift method used by OMV Austria Exploration & Production GmbH in the Vienna Basin for oil production wells requiring artificial lift. Production in the Vienna Basin is characterized by harsh production conditions and high water content due to the high maturity of the oil field. To ensure economical production, it is necessary that the production wells are operated uninterrupted for as long as possible in order to keep maintenance costs and downtime low. This requires a well thought-out design of the system, which should avoid preceding causes of failure and adapt to the production conditions as good as possible in terms of components and operating parameters. This study aims to investigate the influence of design and operational data from the years 2016-2023 on the runlife of OMV's wellstock. Furthermore, the most common causes of failure are examined individually. Various regression methods were tested to select a suitable model and examine the determined regression coefficients. A random forest model was comparatively included in the selection of influencing factors. It was found that classical linear regression models cannot be applied due to the discrete character of the runlife metric. The use of a classical Poisson model was ruled out due to detected overdispersion. The best fit was provided by a regression model based on a negative binomial distribution. However, even with this model, it was not possible to make reliable statements about the influence of the descriptive variables. The complexity of the system and the variability of the individual wells are one possible problem, which results in a high number of unknown influencing factors. Furthermore, the observations are not entirely independent, as several production periods were examined for some of the investigated wells. The limited amount of data, especially for the individual failure causes, posed further problems. In conclusion, possible solutions for the described issues are presented to facilitate further treatment of the problem in the future.",
keywords = "Gest{\"a}ngetiefpumpe, Ausfallanalyse, Designoptimierung, Laufzeit, Regression, Z{\"a}hldaten, Poisson Regression, Random Forest Regression, Sucker rod pump, Failure analysis, Design optimization, Runlife, Regression, Count data, Poisson regression, Random forest regression",
author = "Hofmaninger, {Florian Alexander}",
note = "no embargo",
year = "2023",
language = "English",
school = "Montanuniversitaet Leoben (000)",

}

RIS (suitable for import to EndNote) - Download

TY - THES

T1 - Sucker Rod Pump Failure Analysis using Machine Learning Algorithms

AU - Hofmaninger, Florian Alexander

N1 - no embargo

PY - 2023

Y1 - 2023

N2 - The sucker rod pump system represents the main artificial lift method used by OMV Austria Exploration & Production GmbH in the Vienna Basin for oil production wells requiring artificial lift. Production in the Vienna Basin is characterized by harsh production conditions and high water content due to the high maturity of the oil field. To ensure economical production, it is necessary that the production wells are operated uninterrupted for as long as possible in order to keep maintenance costs and downtime low. This requires a well thought-out design of the system, which should avoid preceding causes of failure and adapt to the production conditions as good as possible in terms of components and operating parameters. This study aims to investigate the influence of design and operational data from the years 2016-2023 on the runlife of OMV's wellstock. Furthermore, the most common causes of failure are examined individually. Various regression methods were tested to select a suitable model and examine the determined regression coefficients. A random forest model was comparatively included in the selection of influencing factors. It was found that classical linear regression models cannot be applied due to the discrete character of the runlife metric. The use of a classical Poisson model was ruled out due to detected overdispersion. The best fit was provided by a regression model based on a negative binomial distribution. However, even with this model, it was not possible to make reliable statements about the influence of the descriptive variables. The complexity of the system and the variability of the individual wells are one possible problem, which results in a high number of unknown influencing factors. Furthermore, the observations are not entirely independent, as several production periods were examined for some of the investigated wells. The limited amount of data, especially for the individual failure causes, posed further problems. In conclusion, possible solutions for the described issues are presented to facilitate further treatment of the problem in the future.

AB - The sucker rod pump system represents the main artificial lift method used by OMV Austria Exploration & Production GmbH in the Vienna Basin for oil production wells requiring artificial lift. Production in the Vienna Basin is characterized by harsh production conditions and high water content due to the high maturity of the oil field. To ensure economical production, it is necessary that the production wells are operated uninterrupted for as long as possible in order to keep maintenance costs and downtime low. This requires a well thought-out design of the system, which should avoid preceding causes of failure and adapt to the production conditions as good as possible in terms of components and operating parameters. This study aims to investigate the influence of design and operational data from the years 2016-2023 on the runlife of OMV's wellstock. Furthermore, the most common causes of failure are examined individually. Various regression methods were tested to select a suitable model and examine the determined regression coefficients. A random forest model was comparatively included in the selection of influencing factors. It was found that classical linear regression models cannot be applied due to the discrete character of the runlife metric. The use of a classical Poisson model was ruled out due to detected overdispersion. The best fit was provided by a regression model based on a negative binomial distribution. However, even with this model, it was not possible to make reliable statements about the influence of the descriptive variables. The complexity of the system and the variability of the individual wells are one possible problem, which results in a high number of unknown influencing factors. Furthermore, the observations are not entirely independent, as several production periods were examined for some of the investigated wells. The limited amount of data, especially for the individual failure causes, posed further problems. In conclusion, possible solutions for the described issues are presented to facilitate further treatment of the problem in the future.

KW - Gestängetiefpumpe

KW - Ausfallanalyse

KW - Designoptimierung

KW - Laufzeit

KW - Regression

KW - Zähldaten

KW - Poisson Regression

KW - Random Forest Regression

KW - Sucker rod pump

KW - Failure analysis

KW - Design optimization

KW - Runlife

KW - Regression

KW - Count data

KW - Poisson regression

KW - Random forest regression

M3 - Master's Thesis

ER -