DEVELOPMENT OF SUPERVISED MACHINE LEARNING MODEL TO CLASSIFY PIPELINES USING QUANTITATIVE RISK ASSESSMENT IN IMPROVING PEATRICE SOFTWARE

• Main Author: Rebecca Panjaitan, Michelle
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/85081
• Institution: Institut Teknologi Bandung

Pipelines are mainly used in the oil and gas industry to transmit fluids due to their cost-effectiveness, reliability, and safety. However, failures can occur during installation and ongoing operations, making risk assessment important to be done. Risk assessment involves evaluating the probability of failure (PoF) and consequence of failure (CoF). PEATRICE software provides calculations regarding this assessment, but it requires a large amount of data, complex analysis, and a long time. Addressing these challenges, this final project develops machine learning to provide risk level prediction as the risk assessment result. The risk assessment method used in this final project is a quantitative risk assessment to evaluate 361 pipeline segments. Then, five machine learning algorithms, logistic regression, random forest, support vector machine, k-nearest neighbor, and decision tree, are utilized to predict risk level based on several features. The assessment results show that 58% of total segments have the PoF category 4, 71% have the consequence area category D, and 45% have the financial consequence category D. There is also a relation between CO2 content to the probability of failure, which has a positive trend. Additionally, more than 78% of segments have medium-high risk levels. Then, it is proved that machine learning can make predictions based on selected features. The support vector machines achieve the highest average accuracy of 78.24% for area-based and logistic regression achieves the best of 84.8% for financial-based risk levels.