Physics-Guided Machine Learning (Shallow and Deep ML) as Surrogate Model for Investigation of CO2 Top of Line Corrosion (TLC) in 22inch Pipeline Containing of Multiphase Fluids

Mitraningsih, Farih (2026) Physics-Guided Machine Learning (Shallow and Deep ML) as Surrogate Model for Investigation of CO2 Top of Line Corrosion (TLC) in 22inch Pipeline Containing of Multiphase Fluids. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text NRP6011232003-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

CO2 top-of-line corrosion (TLC) continues to pose a significant integrity threat to oil and gas pipelines, traditionally linked to the water condensation rate (WCR). However, field data from 22-inch pipeline with minimal WCR indicated severe TLC, suggesting that liquid hold-up and gas void dynamics also contribute to this phenomenon. To address the lack of predictive tools, this study introduces a physics-guided machine learning (ML) framework incorporating both shallow and deep neural networks (ANN, DNN). The physics guidance is based on 35 multiscale features, with key factors including pressure, temperature, WCR, natural pH, superficial gas velocity, and liquid hold-up fraction. Comprehensive preprocessing steps, such as cleaning, scaling, and stratified splitting, are conducted prior to model development. Model performance is evaluated using R2, MAE, MSE, RMSE, and rRMSE metrics. While the interpretability is maintained through Shapley additive explanations (SHAP) and validation on unseen datasets. The findings reveal that the physics-guided DNN consistently outperforms the ANN, achieving over 90% accuracy and strong allignent with baseline TLC rates. By surpassing mechanistic models and in-line inspection (ILI), the proposed surrogate offers a scalable, interpretable, and industry-ready approach for proactive CO2 TLC prediction and pipeline integrity management.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	CO2 top of line corrosion (TLC), corrosion modelling, physics-guided ML, ANN, DNN
Subjects:	T Technology > TJ Mechanical engineering and machinery > TJ930 Pipelines (General). Underwater pipelines.
Divisions:	Faculty of Industrial Technology > Material & Metallurgical Engineering > 27101-(S2) Master Thesis
Depositing User:	Mitraningsih Farih
Date Deposited:	28 Jul 2026 02:54
Last Modified:	28 Jul 2026 02:54
URI:	http://repository.its.ac.id/id/eprint/139239

Actions (login required)

View Item