Analisis Perbandingan Metode Univariat Time Series, Control Chart, Dan Machine Learning Pada Deteksi Dini Kegagalan Komponen Pembangkit Listrik Tenaga Uap (Studi Kasus Pada Salah Satu Perusahaan Pengelola Pembangkit)

Manik, Jacob Pratama (2025) Analisis Perbandingan Metode Univariat Time Series, Control Chart, Dan Machine Learning Pada Deteksi Dini Kegagalan Komponen Pembangkit Listrik Tenaga Uap (Studi Kasus Pada Salah Satu Perusahaan Pengelola Pembangkit). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5010211145_Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Deteksi dini kegagalan merupakan elemen krusial dalam strategi pemeliharaan prediktif, terutama pada sistem pembangkit listrik tenaga uap (PLTU) yang kompleks. Penelitian ini bertujuan untuk mengembangkan dan membandingkan tiga pendekatan utama dalam deteksi anomali berbasis data sensor, yaitu model univariat Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Multivariate Exponentially Weighted Moving Average (M-EWMA), dan deep learning berbasis Long Short-Term Memory (LSTM). Studi dilakukan pada komponen coal pulverizer B di salah satu PLTU di Indonesia, yang memiliki riwayat kegagalan kebocoran (leakage) serta data sensor historis yang memadai. Masing-masing metode diuji dalam skenario nyata menggunakan data yang dibagi menjadi periode pelatihan dan pengujian, dan dievaluasi dari sisi indikator teknis yang meliputi precision, recall, F1-score, dan lead time serta indikator ekonomi berupa potensi total penghematan jika model diimplementasikan. Hasil analisis menunjukkan bahwa LSTM memiliki performa paling unggul secara teknis dan finansial, dengan F1-score 91% dan total penghematan sebesar 248.5 juta rupiah. Namun, setiap metode memiliki keunggulan dan karakteristik tersendiri: SARIMAX unggul dalam lead time, M-EWMA sensitif terhadap perubahan kolektif, dan LSTM menunjukkan presisi tinggi. Namun demikian, ada kemungkinan bahwa pemilihan metode yang digunakan dalam implementasi nyata perlu disesuaikan dengan kebutuhan operasional, sumber daya, serta karakteristik sistem yang dimonitor.
===================================================================================================================================
Early failure detection is a crucial element in predictive maintenance strategies, especially for complex coal-fired power plant systems. This study aims to develop and compare three main approaches to sensor-based anomaly detection: the univariate Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX) model, Multivariate Exponentially Weighted Moving Average (M-EWMA), and deep learning based on Long Short-Term Memory (LSTM). The study focuses on Coal Pulverizer B in one of Indonesia’s coal-fired power plants, which has a known history of leakage failures and sufficient historical sensor data. Each method was tested in real-world scenarios using data split into training and testing periods and evaluated based on technical indicators, including precision, recall, F1-score, and lead time, as well as economic indicators, namely the potential total cost savings if the model were implemented. The analysis results show that LSTM outperforms the other methods both technically and financially, achieving an F1-score of 91% and a total saving of IDR 248.5 million. However, each method offers its own strengths and characteristics: SARIMAX excels
in lead time, M-EWMA is sensitive to collective changes, and LSTM demonstrates high precision. Nonetheless, the choice of method for real-world implementation may need to be adjusted based on operational needs, available resources, and the characteristics of the monitored system.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	pemeliharaan prediktif, univariat, multivariat, machine learning, deret waktu, deteksi anomali, predictive maintenance, univariate, multivariate, machine learning, time series, anomaly detection
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA169.5 Failure analysis T Technology > TJ Mechanical engineering and machinery > TJ174 Maintenance and repair of machinery T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1322.6 Electric power-plants
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26201-(S1) Undergraduate Thesis
Depositing User:	Jacob Pratama M
Date Deposited:	30 Jul 2025 07:57
Last Modified:	30 Jul 2025 07:57
URI:	http://repository.its.ac.id/id/eprint/124415

Actions (login required)

View Item