Deteksi Dini Anomali berbasis Deep Learning pada Sistem Pembacaan Sensor Sales Gas Compressor untuk Predictive Maintenance

Zulkarnain, Achmad Doddy (2026) Deteksi Dini Anomali berbasis Deep Learning pada Sistem Pembacaan Sensor Sales Gas Compressor untuk Predictive Maintenance. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032242119-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Lapangan Gas Jambaran Tiung Biru (JTB) merupakan aset strategis nasional yang vital bagi pemenuhan pasokan gas di wilayah Jawa Timur dan Jawa Tengah. Untuk menjaga kontinuitas dan kualitas pasokan sesuai dengan Perjanjian Jual Beli Gas (PJBG), keandalan fasilitas produksi menjadi prioritas utama. Salah satu peralatan paling krusial dalam fasilitas ini adalah Sales Gas Compressor (CP02-260-CG9101). Kegagalan pada sistem compressor turbine tersebut berdampak signifikan terhadap potensi kerugian finansial yang besar akibat unplanned shutdown. Saat ini, sistem pemantauan operasional yang berjalan masih mengandalkan operator surveillance konvensional dengan batas ambang statis (high and low thresholds). Pendekatan ini dinilai kurang efektif dalam mengidentifikasi gejala anomali dini, sehingga meningkatkan risiko kegagalan tak terduga. Penelitian ini bertujuan untuk mengembangkan dan menguji kinerja sistem deteksi anomali pada pembacaan sensor lube oil system pada Sales Gas Compressor (CP02-260-CG9101) dengan menerapkan empat arsitektur Deep Learning, yaitu Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), dan Bidirectional Long Short-Term Memory (BiLSTM). Evaluasi kinerja prediksi model diukur menggunakan metrik Mean Absolute Error (MAE) dan Root Mean Squared Error (RMSE), sedangkan penentuan batas ambang dinamis (dynamic threshold) dihitung menggunakan metode statistik persentil. Hasil penelitian menunjukkan bahwa model GRU memberikan performa terbaik dengan konfigurasi pembagian data latih dan data uji sebesar 70%:30% serta ukuran timestep 10. Konfigurasi tersebut menghasilkan nilai MAE terkecil sebesar 0,40640, RMSE sebesar 1,13866 dan R2 sebesar 0,43256. Melalui penerapan perhitungan persentil 98 %, sistem yang dikembangkan berhasil membentuk threshold secara dinamis dan mampu memberikan peringatan dini (early warning) dalam durasi 8 menit 9 detik sebelum terjadinya aktual alarm. Model terbukti mampu meningkatkan kemampuan deteksi anomali secara proaktif dan dapat mendukung implementasi predictive maintenance pada fasilitas pengolahan gas di JTB.
========================================================================================================================================
The Jambaran Tiung Biru (JTB) Gas Field is a strategic asset that plays a vital role in ensuring gas supply for East Java and Central Java. The availability of JTB gas must be maintained in accordance with the quality standards specified in the Gas Sales and Purchase Agreement (GSPA). One of the most critical pieces of equipment is the Sales Gas Compressor (CP02-260-CG9101). Failure of this compressor turbine system can result in substantial financial losses due to unplanned shutdowns.The primary challenge is that the current operational monitoring system relies heavily on operator surveillance, while the existing high and low threshold alarms are not sufficiently effective in identifying anomalies at an early stage. This limitation increases the risk of unexpected equipment failures. Therefore, this study aims to develop and evaluate an anomaly detection system for the lube oil sensor readings of the Sales Gas Compressor (CP02-260-CG9101) using Deep Learning methods, including Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Long Short-Term Memory (BiLSTM). To assess the predictive performance of all models, Mean Absolute Error (MAE) is employed as the evaluation metric, while anomaly threshold values are determined using a percentile-based statistical approach. The results indicate that the GRU model outperformed the other architectures when using a 70%:30% training-testing data split and a timestep of 10, achieving an MAE of 0.40640, an RMSE of 1.13866 and R 2 of 0,43256. In addition, the proposed system successfully established a dynamic threshold based on the 98th percentile and provided an early warning lead time of 8 minutes and 9 seconds prior to an unplanned shutdown. These results demonstrate the effectiveness of the proposed approach in enabling proactive anomaly detection and supporting the implementation of predictive maintenance strategies at the Jambaran-Tiung Biru (JTB) gas processing facility.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deteksi Dini Anomali, Sales Gas Compressor, Deep Learning, Mean Absolute Error, Predictive Maintenance, Jambaran Tiung Biru.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Achmad Doddy Zulkarnain
Date Deposited:	27 Jul 2026 15:26
Last Modified:	27 Jul 2026 15:26
URI:	http://repository.its.ac.id/id/eprint/139159

Actions (login required)

View Item