Estimasi Remaining Useful Life Sistem Traksi Pada LRT Menggunakan Similarity-Based Eksponensial dengan Segmentasi K-Means dan Isolation Forest

Al Husaini, Muhammad Rifqi (2026) Estimasi Remaining Useful Life Sistem Traksi Pada LRT Menggunakan Similarity-Based Eksponensial dengan Segmentasi K-Means dan Isolation Forest. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009211125-Undergraduates_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Sistem traksi merupakan jantung LRT yang mengubah energi listrik menjadi gerak, sehingga kegagalannya dapat menyebabkan downtime dan menurunkan tingkat keselamatan operasi. Penelitian ini mengusulkan metode estimasi Remaining Useful Life (RUL) sistem traksi LRT berbasis similarity-based eksponensial dengan memanfaatkan segmentasi operasi k-means dan deteksi outlier Isolation Forest (IF). Health Indicator (HI) komposit dibentuk melalui kombinasi linear 3 variabel operasi dan 12 sensor yang telah dinormalisasi per-rejim, kemudian trajektori HI dimodelkan sebagai kurva degradasi eksponensial dan direpresentasikan oleh parameter tren. Estimasi RUL dilakukan dengan mencocokan parameter tren antara unit yang dipantau dan pustaka run-to-failure historis menggunakan algoritma K-Nearest Neighbors pada klaster operasi yang sama, sementara IF digunakan untuk membersihkan outlier sebelum konstruksi HI dan pemodelan RUL. Kinerja model dibandingkan antara skenario baseline tanpa IF dan dengan IF pada tiga breakpoint umur pakai, yaitu 50%, 70%, dan 90% life. Evaluasi dilakukan menggunakan metrik Mean Absolute Error (MAE), Root Mean Square Error (RMSE), MAEfrac, dan RMSEfrac dalam fraksi umur komponen, serta analisis kurva Probability Density Function (PDF) untuk menilai karakteristik sebaran kesalahan prediksi. Secara umum, pembersihan outlier dengan IF mampu mereduksi noise HI dan menghasilkan estimasi RUL yang lebih baik pada breakpoint 70% – 90% life, meskipun tidak selalu memberikan kinerja terbaik ketika prediksi dilakukan pada awal umur pakai pada breakpoint 50 %.
==========================================================================================================================
Traction system is the heart of the LRT, converting electrical energy into motion, so its failure can cause downtime and reduce the level of operational safety. This study proposes a Remaining Useful Life (RUL) estimation method for the LRT traction system based on an exponential similarity-based approach by utilizing k-means operational segmentation and Isolation Forest (IF) outlier detection. A composite Health Indicator (HI) is formed through a linear combination of 3 operational variables and 12 sensors that have been normalized per regime, then the HI trajectory is modeled as an exponential degradation curve and represented by trend parameters. RUL estimation is carried out by matching the trend parameters between the monitored unit and the historical run-to-failure library using the K-Nearest Neighbors algorithm within the same operational cluster, while IF is used to clean outliers before HI construction and RUL modeling. The model performance is compared between the baseline scenario without IF and with IF at three life breakpoints, namely 50%, 70%, and 90% life. The evaluation is conducted using the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), MAEfrac, and RMSEfrac metrics in terms of component life fraction, as well as analysis of the Probability Density Function (PDF) curves to assess the characteristics of the prediction error distribution. In general, outlier cleaning with IF can reduce HI noise and produce more stable RUL estimates at the 70 % and 90% life breakpoint, However, it does not consistently achieve the best performance under early-life prediction conditions, especially at the 50% life breakpoint.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Health Indicator, Isolation Forest, k-means, Remaining Useful Life, Similarity-based eksponensial, Sistem Traksi, Health Indicator, Isolation Forest, k-means, Remaining Useful Life, Similarity-based exponential, Traction System.
Subjects:	T Technology > TF Railroad engineering and operation > TF193 Estimates, costs, etc.
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Rifqi Al Husaini
Date Deposited:	03 Feb 2026 08:53
Last Modified:	03 Feb 2026 08:53
URI:	http://repository.its.ac.id/id/eprint/131859

Actions (login required)

View Item