Deteksi Kondisi Abnormal Level Minyak Pada Sistem Lube Oil Tank Small Scale Menggunakan Support Vector Regression (SVR)

Heriyani, Fransiska Septi (2026) Deteksi Kondisi Abnormal Level Minyak Pada Sistem Lube Oil Tank Small Scale Menggunakan Support Vector Regression (SVR). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221023-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Energi panas bumi merupakan salah satu sumber energi terbarukan yang berperan penting dalam penyediaan listrik di Indonesia. Pada Pembangkit Listrik Tenaga Panas Bumi (PLTP), sistem pelumasan turbin berfungsi menjaga keandalan operasi dengan meminimalkan gesekan antar komponen. Salah satu permasalahan yang terjadi di PT Geo Dipa Energi Unit Dieng adalah kontaminasi air pada Lube Oil Tank yang menyebabkan kenaikan level lube Oil dan berpotensi meningkatkan vibrasi turbin. Kondisi tersebut dapat menurunkan performa sistem apabila tidak terdeteksi sejak dini. Oleh karena itu, Proyek Akhir ini bertujuan mengembangkan sistem deteksi kondisi Abnormal menggunakan metode Support Vector Regression (SVR) sebagai pembentuk baseline kondisi operasi normal. Data yang digunakan berupa data Time Series hasil pemantauan sensor Level Lube Oil, Average Turbine Vibration, dan tekanan Inlet Oil Conditioner. Sebelum pemodelan dilakukan tahapan Preprocessing meliputi sinkronisasi data, pemeriksaan missing value, duplikasi, validasi rentang data, standardisasi fitur, serta pembagian data secara time-based. Model SVR dioptimasi menggunakan GridSearchCV dengan kernel Radial Basis Function (RBF). Hasil prediksi digunakan untuk menghitung residual yang kemudian dianalisis menggunakan metode Z-Score untuk mengidentifikasi kondisi Abnormal. Kinerja model dievaluasi menggunakan metrik regresi berupa RMSE, MAE, dan R², sedangkan performa deteksi kondisi Abnormal dievaluasi menggunakan Confusion Matrix. Hasil Proyek Akhir menunjukkan bahwa model SVR mampu merepresentasikan pola operasi normal dengan baik, ditunjukkan oleh nilai R² sebesar 0,814 pada Level Lube Oil dan 0,373 pada Average Turbine Vibration. Sistem berhasil mendeteksi 12 titik kondisi Abnormal pada grafik Level Lube Oil dan 23 titik pada grafik kondisi Abnormal pada vibrasi turbin yang terkonsentrasi pada periode lonjakan akibat kontaminasi air. Selain itu, perbandingan dengan metode Linear Regression menunjukkan bahwa SVR unggul pada kemampuan deteksi kondisi Abnormal, meskipun Linear Regression lebih baik pada metrik regresi Level Lube Oil. Hasil Proyek Akhir menunjukkan bahwa metode SVR layak diterapkan sebagai pendukung condition monitoring dan early warning system untuk meningkatkan keandalan sistem pelumasan turbin.
==================================================================================================================================
Geothermal energy is one of Indonesia's major renewable energy resources and plays an important role in electricity generation. In a Geothermal Power Plant (GPP), the turbine lubrication system is essential for maintaining operational reliability by minimizing friction between mechanical components. One of the operational problems encountered at PT Geo Dipa Energi Dieng Unit is water contamination in the Lube Oil Tank, which increases the lube Oil level and potentially causes excessive turbine vibration. If not detected at an early stage, this condition may reduce system performance and increase the risk of equipment failure. Therefore, this study aims to develop an anomaly detection system using Support Vector Regression (SVR) to establish a baseline of normal operating conditions. The study employed time-series data collected from Lube Oil Level, Average Turbine Vibration, and Inlet Oil Conditioner Pressure sensors. Data Preprocessing included data synchronization, missing value inspection, duplicate removal, range validation, feature standardization, and time-based data splitting. The SVR model was optimized using GridSearchCV with a Radial Basis Function (RBF) kernel. Prediction residuals were subsequently analyzed using the Z-score method to identify anomalies. Model performance was evaluated using regression metrics, including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination (R²), while anomaly detection performance was assessed using a Confusion Matrix. The results indicate that the proposed SVR model effectively represents normal operating conditions, achieving an R² value of 0.814 for Lube Oil Level and 0.373 for Turbine Vibration under normal testing data. The anomaly detection evaluation achieved an accuracy of 87.4% for Lube Oil Level and 86.0% for Turbine Vibration. Furthermore, a comparison with Linear Regression demonstrates that SVR excels in anomaly detection, although Linear Regression yields better regression metrics for Lube Oil Level. These findings suggest that the proposed approach is suitable for supporting condition monitoring and early warning systems in geothermal turbine lubrication systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	PLTP, Deteksi Kondisi Abnormal, Level Oil tank, Monitoring, Support Vector Regression, Geothermal Power Plant, Anomaly Detection, Oil level, Monitoring, Support Vector Regression
Subjects:	A General Works > AI Indexes (General) A General Works > AI Indexes (General) T Technology > T Technology (General) > T57.5 Data Processing T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1318 Geothermal Power Plants T Technology > TS Manufactures > TS174 Maintainability (Engineering) . Reliability (Engineering)
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Fransiska Septi Heriyani
Date Deposited:	11 Aug 2026 09:26
Last Modified:	11 Aug 2026 09:26
URI:	http://repository.its.ac.id/id/eprint/142323

Actions (login required)

View Item