Prediksi Fitur Health Index Multivariat Bearing Boiler Feedwater Pump Turbine pada Sistem Boiler PLTU Menggunakan Fitur Nonlinier dan Metode Attention Mechanism-LSTM

Khasyyatullah, Kautzar Randra Noor (2026) Prediksi Fitur Health Index Multivariat Bearing Boiler Feedwater Pump Turbine pada Sistem Boiler PLTU Menggunakan Fitur Nonlinier dan Metode Attention Mechanism-LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221038-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

PLTU menyumbang lebih dari 65% kebutuhan listrik nasional Indonesia, sehingga kegagalan komponen kritisnya berdampak langsung terhadap keandalan pasokan. Bearing Boiler Feedwater Pump Turbine (BFPT) rentan terhadap kegagalan tidak terduga akibat pola degradasi multivariat dan nonlinier yang belum dapat ditangkap oleh pendekatan pemeliharaan konvensional. Penelitian ini mengusulkan prediksi health index bearing BFPT menggunakan arsitektur Multi-Channel Attention LSTM (MCA-LSTM), yang memadukan channel attention untuk seleksi sensor, LSTM untuk dependensi temporal, dan hidden state attention untuk seleksi posisi historis. Health index dibangun dari eksponen Detrended Fluctuation Analysis (DFA) enam sensor getaran dan suhu pada bearing 1 dan 2 BFPT A, menghasilkan dataset 570 hari (Mei 2023 - Mei 2025), dengan model dilatih memprediksi tujuh hari ke depan menggunakan riwayat 60 hari. Model terbaik pasca-tuning menghasilkan MAPE = 6.44%, MAE = 0.0849, dan RMSE = 0.1110. Dengan cakupan tuning yang disetarakan (18 kombinasi untuk kedua model), MCA-LSTM tetap mengungguli VAR-Alpha terbaik (MAPE = 6.85%). Channel attention mengidentifikasi sensor getaran vertikal BRG 2 (VIB Y) sebagai paling informatif dengan bobot 0.4225, konsisten dengan fisika zona beban vertikal pada mesin horizontal. Prediksi tujuh hari ke depan menunjukkan health index stabil di kisaran 1.3192 - 1.3292 dengan tren meningkat tipis, mengindikasikan kondisi bearing yang perlu mendapat pemantauan lebih ketat. Penelitian ini menyimpulkan bahwa MCA-LSTM berbasis health index DFA efektif untuk memantau tren kondisi bearing multivariat sekaligus menghasilkan interpretabilitas yang dapat dijadikan dasar perencanaan pemeliharaan prediktif pada PLTU.
======================================================================================================================================
Steam power plants supply over 65% of Indonesia's national electricity demand, making the reliability of critical components essential. Boiler Feedwater Pump Turbine (BFPT) bearings are susceptible to unexpected failure due to multivariate and nonlinear degradation patterns that conventional maintenance approaches have been unable to adequately capture. This study proposes health index prediction for BFPT bearings using a Multi-Channel Attention LSTM (MCA-LSTM) architecture, which integrates channel attention for sensor selection, LSTM for temporal dependencies, and hidden state attention for historical position selection. The health index is constructed from DFA (Detrended Fluctuation Analysis) exponents of six vibration and temperature sensors on bearings 1 and 2 of BFPT A, producing a 570-day dataset (May 2023 - May 2025), with the model trained to predict seven days ahead from a 60-day history. The best post-tuning model achieves MAPE = 6.44%, MAE = 0.0849, and RMSE = 0.1110. With an equalized tuning budget (18 combinations for both models), MCA-LSTM still outperforms the best VAR-Alpha (MAPE = 6.85%). Channel attention identifies the vertical vibration sensor of BRG 2 (VIB Y) as the most informative, with a weight of 0.4225, consistent with the vertical load-zone physics of horizontal machinery. The seven-day forecast shows a stable health index ranging from 1.3192 to 1.3292 with a slight upward trend, indicating a bearing condition that warrants closer monitoring. This study concludes that MCA-LSTM based on a DFA health index is effective for monitoring multivariate bearing condition trends while producing interpretability that can serve as a basis for predictive maintenance planning in steam power plants.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	pembangkit listrik tenaga uap, pemeliharaan prediktif, health index, detrended fluctuation analysis, long short-term memory, attention mechanism, steam power plant, predictive maintenance, health index, detrended fluctuation analysis, long short-term memory, attention mechanism
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Kautzar Randra Noor Khasyyatullah
Date Deposited:	22 Jul 2026 06:20
Last Modified:	22 Jul 2026 06:20
URI:	http://repository.its.ac.id/id/eprint/136238

Actions (login required)

View Item