Prediksi Kegagalan Aset Primary Air Fan Menggunakan Metode Deep Learning

Mahajana, Agung Wahyu (2026) Prediksi Kegagalan Aset Primary Air Fan Menggunakan Metode Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember (ITS).

Text 6010231086-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Pembangkit listrik memiliki peran krusial sebagai sektor hulu dalam rantai pasok tenaga listrik, di mana keandalan operasional peralatan sangat dipengaruhi oleh efektivitas pemeliharaan peralatan. Pemeliharaan prediktif berbasis machine learning dan deep learning telah banyak dikembangkan untuk mendukung diagnosis dan prognosis kegagalan peralatan. Berdasarkan laporan downtime tahun 2025 pada salah satu Pembangkit Listrik Tenaga Uap (PLTU) PT XYZ, Primary Air Fan (PA Fan) diidentifikasi sebagai aset kritis yang memberikan kontribusi signifikan terhadap MWh Losses. Oleh karenanya, diperlukan suatu metode proaktif untuk memprediksi potensi kegagalan peralatan dan mendukung pengambilan keputusan pemeliharaan. Penelitian ini mengusulkan model prediksi Piecewise Remaining Useful Life (Piecewise RUL) peralatan PA Fan dengan menggunakan Deep Learning. Metode yang diusulkan dibagi ke dalam tiga tahap utama, yaitu pemrosesan awal data, pengembangan model, dan prediksi kegagalan berikutnya. Tahap pemrosesan awal data terdiri atas perhitungan dan normalisasi Piecewise RUL, analisis korelasi serta seleksi fitur menggunakan Pearson Correlation Coefficient (PCC), normalisasi fitur menggunakan MinMax Scaler, pengisian data kosong menggunakan K-Nearest Neighbor (KNN) Imputer, dan pembentukan sliding window untuk data deret waktu. Selain itu, enam jenis fitur statistik diekstraksi dari data sensor pada rentang enam dan 24 jam dan diintegrasikan dengan representasi urutan LSTM melalui late fusion untuk memperkaya informasi kondisi peralatan. Penelitian ini menerapkan Bayesian Optimization untuk mengoptimalkan hyperparameter Deep Learning. Kombinasi hyperparameter terbaik kemudian digunakan untuk membangun model prediksi utama. Model Deep Learning yang diusulkan adalah Long Short-Term Memory (LSTM), dibandingkan dengan Transformer, Gated Recurrent Unit (GRU), Support Vector Regression (SVR), dan Extreme Gradient Boosting (XGBoost). Hasil evaluasi berdasarkan matriks Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), dan coefficient of determination (R2) menunjukkan bahwa model LSTM memiliki kinerja yang lebih baik dibandingkan algoritme lainnya dalam memprediksi Piecewise RUL PA Fan A.
==================================================================================================================================
Power plants play a crucial role as the upstream sector in the electricity supply chain, where the operational reliability of equipment is strongly influenced by the effectiveness of equipment maintenance. Predictive maintenance based on machine learning and deep learning has been widely developed to support fault diagnosis and prognosis. Based on the 2025 downtime report of a coal-fired steam power plant operated by PT XYZ, the Primary Air Fan (PA Fan) was identified as a critical asset that contributed significantly to MWh losses. Therefore, a proactive method is required to predict potential failures and support maintenance decision-making. This study proposes a Piecewise Remaining Useful Life (Piecewise RUL) prediction model for PA Fan using deep learning. The proposed method is divided into three main stages: data preprocessing, model development, and next-failure prediction. The data preprocessing stage consists of Piecewise RUL calculation and normalization, correlation analysis and feature selection using Pearson Correlation Coefficient (PCC), feature normalization using MinMax Scaler, missing data imputation using the K-Nearest Neighbor (KNN) Imputer, and sliding window construction for time-series data. In addition, six types of statistical features were extracted from the sensor data over six- and 24-hour intervals and integrated with the LSTM sequence representation through late fusion to enrich the equipment condition information. This study applies Bayesian Optimization to optimize the hyperparameters of the deep learning models. The best hyperparameter combination is then used to develop the main prediction model. The proposed main model is Long Short-Term Memory (LSTM), which is compared with Transformer, Gated Recurrent Unit (GRU), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost). The evaluation results based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and coefficient of determination (R2) indicate that the LSTM model outperformed the other algorithms in predicting the Piecewise RUL of PA Fan A.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Predictive Maintenance, Primary Air Fan, Deep Learning, Pembangkit Listrik, Deret Waktu Predictive Maintenance, Primary Air Fan, LSTM, Deep Learning, Time Series
Subjects:	T Technology > TS Manufactures > TS174 Maintainability (Engineering) . Reliability (Engineering)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26101-(S2) Master Thesis
Depositing User:	Agung Wahyu Mahajana
Date Deposited:	04 Aug 2026 03:30
Last Modified:	04 Aug 2026 03:30
URI:	http://repository.its.ac.id/id/eprint/143332

Actions (login required)

View Item