Analisis Remaining Useful Life Komponen Kritis Booster Compressor Menggunakan Deep Learning dan Bayesian Optimization Berbasis Data Temperatur

Al Amin, Muhamad Nabil (2026) Analisis Remaining Useful Life Komponen Kritis Booster Compressor Menggunakan Deep Learning dan Bayesian Optimization Berbasis Data Temperatur. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009221131-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Industri gas production plant seperti PT Pertamina EP sangat bergantung pada keandalan unit booster compressor, di mana kegagalan pada sub-sistem krusial dapat menyebabkan downtime yang signifikan dan kerugian biaya yang besar akibat automatic shutdown. Overheating merupakan salah satu penyebab utama kerusakan mesin, sehingga diperlukan analisis Remaining Useful Life (RUL) berbasis data temperatur untuk memprediksi sisa umur pakai komponen kritis secara proaktif. Penelitian ini bertujuan untuk mengembangkan dan membandingkan performa dari model prediksi RUL melalui temperature forecasting menggunakan arsitektur deep learning 1D LSTM, Bi-LSTM, dan GRU yang dioptimasi dengan Bayesian Optimization demi mendukung poin SDG 9 dan SDG 12 terkait infrastruktur, inovasi, serta produksi dan konsumsi. Metode penelitian diawali dengan identifikasi komponen kritis melalui Failure Mode Effect and Criticality Analysis (FMECA), yang menetapkan auxiliary gearbox sebagai komponen kritis. Data temperatur dari sensor engine jacket water digunakan sebagai input untuk pelatihan dan pengujian model. Hasil penelitian menunjukkan bahwa Bayesian Optimization memiliki performa yang baik dalam pemilihan hiperparameter arsitektur deep learning dengan nilai training loss (MSE) rata-rata 0,0037 serta nilai validation loss (MSE) rata-rata 0,008, dengan model arsitektur Gated Recurrent Unit (GRU) yang merupakan model dengan performa temperature forecasting terbaik dengan rata-rata persentase error RUL sebesar 10,55%, rata-rata MAE sebesar 2,01°F, serta rata-rata MAPE sebesar 1,05%.
=======================================================================================================================================
Gas production plants such as PT Pertamina EP rely heavily on the reliability of booster compressor units, where failures in crucial subsystems can cause significant downtime and substantial financial losses due to automatic shutdowns. Overheating is one of the primary causes of engine failure; thus, temperature-based Remaining Useful Life (RUL) analysis is needed to proactively predict the remaining service life of critical components. This study aims to develop and compare the performance of RUL prediction models through temperature forecasting using 1D LSTM, Bi-LSTM, and GRU deep learning architectures optimized with Bayesian Optimization to support SDG 9 and SDG 12 related to infrastructure, innovation, and sustainable production and consumption. The research methodology begins with the identification of critical components through Failure Mode Effect and Criticality Analysis (FMECA), which identifies the auxiliary gearbox as a critical component. Temperature data from the engine jacket water sensor is used as input for model training and testing. The results demonstrate that Bayesian Optimization effectively selects hyperparameters for deep learning architectures, achieving mean training loss (MSE) values of 0.0037 and mean validation loss (MSE) values of 0.008. The Gated Recurrent Unit (GRU) architecture model yields the best temperature forecasting performance, with a mean RUL error percentage of 10.55%, a mean MAE of 2.01°F, and a mean MAPE of 1.05%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Booster Compressor, Remaining Useful Life, FMECA, LSTM, GRU, Data Temperatur, SDG 9, SDG 12, Booster Compressor Units, Remaining Useful Life, FMECA, LSTM, GRU, Temperature Data, SDG 9, SDG 12
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Muhamad Nabil Al Amin
Date Deposited:	30 Jul 2026 03:24
Last Modified:	30 Jul 2026 03:24
URI:	http://repository.its.ac.id/id/eprint/139938

Actions (login required)

View Item