Integrasi Machine Learning Catboost dan API 581 untuk Evaluasi Probability of Failure Berbasis Kondisi Operasi Dinamis pada Fasilitas Pengolahan Gas

Salsabila, Salwa (2026) Integrasi Machine Learning Catboost dan API 581 untuk Evaluasi Probability of Failure Berbasis Kondisi Operasi Dinamis pada Fasilitas Pengolahan Gas. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5011221169-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Industri minyak dan gas bumi beroperasi dengan risiko tinggi, sehingga penerapan strategi Risk-Based Inspection (RBI) berdasarkan standar API 581 sangat krusial untuk memprioritaskan jadwal inspeksi peralatan kritis. Namun, pendekatan konvensional sering mengandalkan asumsi laju korosi statis, sedangkan algoritma machine learning ensemble lain rentan mengalami overfitting pada data kategorikal. Penelitian ini bertujuan memprediksi laju korosi secara dinamis menggunakan algoritma Categorical Boosting (CatBoost) dan mengintegrasikannya ke dalam penilaian risiko bejana tekan (Feed Gas K.O. Drum dan Feed Gas Coalescer) di fasilitas gas bumi Jambaran Tiung Biru. Hasil pengujian blind test menghasilkan tingkat kesalahan sebesar 1.3–2%. Model CatBoost terbukti sangat akurat dengan nilai MAE 0.00632 mm/tahun, RMSE 0.01213 mm/tahun, dan R^2 sebesar 0.85. Model memprediksi laju korosi sebesar 0.1027 mm/tahun (shell) dan 0.1049 mm/tahun (head) pada K.O. Drum, serta 0.1038 mm/tahun (shell) dan 0.1052 mm/tahun (head) pada Coalescer. Spesifikasi metalurgi material dan kecepatan aliran fluida merupakan fitur paling dominan yang memengaruhi laju korosi. Integrasi prediksi ke dalam standar API 581 menempatkan kedua bejana pada Probability of Failure (PoF) Kategori 2 (Low). Namun, evaluasi Consequence of Failure (CoF) berada pada Kategori D (High) dengan potensi kerugian ekonomi mencapai lebih dari USD 9.4 juta yang didominasi biaya kehilangan produksi. Pemetaan matriks risiko menyimpulkan kedua peralatan berada pada level 2D (Medium Risk). Sebagai langkah mitigasi, direkomendasikan kombinasi inspeksi IVI, UTT, dan NDT lanjutan, dengan interval maksimum 7 tahun untuk K.O. Drum dan 3 tahun untuk Coalescer.
=========================================================================================================================================
The oil and gas industry operates under high-risk conditions, making the implementation of Risk-Based Inspection (RBI) based on the API 581 standard crucial for prioritizing the inspection schedules of critical equipment. However, conventional approaches often rely on static corrosion rate assumptions, while other machine learning ensemble algorithms remain susceptible to overfitting on categorical data. This study aims to dynamically predict corrosion rates using the Categorical Boosting (CatBoost) algorithm and integrate the results into the risk assessment of pressure vessels (Feed Gas K.O. Drum and Feed Gas Coalescer) at the Jambaran Tiung Biru gas facility. Blind test results yielded an error rate of 1.3–2%. The CatBoost model proved highly accurate with a MAE of 0.00632 mm/year, an RMSE of 0.01213 mm/year, and an R^2 of 0.85. The model predicted corrosion rates of 0.1027 mm/year (shell) and 0.1049 mm/year (head) for the K.O. Drum, and 0.1038 mm/year (shell) and 0.1052 mm/year (head) for the Coalescer. Material metallurgy specifications and fluid velocity were identified as the most dominant features influencing the corrosion rate. Integrating these predictions into the API 581 standard placed both vessels in Probability of Failure (PoF) Category 2 (Low). Conversely, the Consequence of Failure (CoF) evaluation fell into Category D (High), with potential economic losses exceeding USD 9.4 million, dominated by production loss costs. Risk matrix mapping concluded that both equipment pieces are at level 2D (Medium Risk). As a mitigation step, a combination of IVI, UTT, and advanced NDT inspections is recommended, with a maximum interval of 7 years for the K.O. Drum and 3 years for the Coalescer.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Categorical Boosting (CatBoost), Laju Korosi, Probability of Failure, Consequence of Failure, API 581, Categorical Boosting (CatBoost), Corrosion Rate, Probability of Failure, Consequence of Failure, API 581
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TA Engineering (General). Civil engineering (General) > TA169.5 Failure analysis T Technology > TA Engineering (General). Civil engineering (General) > TA462 Metal Corrosion and protection against corrosion
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Material & Metallurgical Engineering > 28201-(S1) Undergraduate Thesis
Depositing User:	Salwa Salsabila
Date Deposited:	24 Jul 2026 01:20
Last Modified:	24 Jul 2026 01:20
URI:	http://repository.its.ac.id/id/eprint/136830

Actions (login required)

View Item