Please login first
Predicting Corrosion Rate in Oil Wells Using First Principles and Machine Learning
* 1, 2 , 3 , 3
1  PDRA, University of Manchester, UK
2  Pungo Tech, Ecuador
3  Corporacion para la Investigacion de la Corrosion, Piedecuesta 681011, Santander, Colombia
Academic Editor: Chuang Deng

Abstract:

Corrosion in oil wells results from the interaction of factors such as chemical composition, temperature, pressure, and fluid flow. Accurately predicting corrosion rates is crucial for operational efficiency and risk mitigation. Traditional models often have limitations in adapting to new operating conditions and capturing complex interactions between variables, while machine learning models can uncover these interactions. This study aims to develop a hybrid model that combines physical and chemical understanding of corrosion with machine learning approaches. Integrating fundamental corrosion principles with machine learning adaptivity, the model seeks to improve accuracy in predicting corrosion rates, thereby reducing operational costs, downtime, and safety risks. The methodology we applied began with the collection of historical data from a representative set of oil wells (more than 30,000 logs), including corrosion rates, well operating parameters, tubing parameters, and fluid compositions. The dataset was carefully cleaned and transformed, addressing missing values and outliers. To integrate first principles, the NORSOK corrosion model, based on fundamental chemical and physical concepts, was applied to calculate theoretical corrosion rates. Subsequently, machine learning models—such as Linear Models, Bootstrap Forest, and Boosted Tree—were trained using metrics like RSquare and Mean Average Error (MAE) to evaluate performance. Among the tested models, a Boosted Tree model (XGBoost) achieved the best performance, achieving an RSquare of 0.163 and an MAE of 1.39, capturing complex parameter relationships. Incorporating the NORSOK-based first principles ensured the model remained robust under varying conditions. Finally, a hybrid approach combining predictions from both physics- and chemistry-based and machine learning models was deployed in the cloud for real-time corrosion rate predictions. This deployment offers significant cost savings—up to a 30% reduction in chemical treatment costs—and enhances operational efficiency by minimizing downtime due to corrosion events.

Keywords: corrosion rate; artificial intelligence; hybrid model; machine learning; oil production
Comments on this paper
Currently there are no comments available.



 
 
Top