Estimasi Remaining Useful Life (Rul) Sistem Traksi Lrt Jabodebek Menggunakan Similarity Based Polynomial

Uwais, Muhammad Jamil (2026) Estimasi Remaining Useful Life (Rul) Sistem Traksi Lrt Jabodebek Menggunakan Similarity Based Polynomial. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009211123-Undergraduate_Thesis.pdf Download (3MB)

Abstract

Sistem traksi merupakan salah satu subsistem utama pada LRT Jabodebek karena berperan langsung dalam mengubah energi listrik menjadi energi mekanik untuk menggerakkan kereta. Penurunan kondisi pada sistem traksi dapat memengaruhi keandalan operasi, sehingga diperlukan pendekatan predictive maintenance melalui estimasi Remaining Useful Life (RUL). Penelitian ini bertujuan untuk menentukan formulasi Health Indicator (HI) dan merancang estimator RUL pada sistem traksi LRT Jabodebek menggunakan pendekatan similarity-based polynomial. Data yang digunakan merupakan data historis run-to-failure sistem traksi dengan 32.845 baris data, 17 kolom variabel, 133 ID pengamatan, tiga variabel operasi, dan dua belas sensor utama. Tahapan penelitian meliputi pembagian data train dan test, pengujian variabel operasi menggunakan mean silhouette, clustering kondisi operasional dengan K-Means, normalisasi sensor per cluster, pemilihan sensor trendable, pembentukan HI, pemodelan degradasi polinomial, serta estimasi RUL menggunakan residual-based weighted KNN. Hasil penelitian menunjukkan bahwa Op3 atau torsi motor menjadi variabel operasi terbaik untuk clustering. Sensor yang digunakan dalam pembentukan HI adalah Sensor3, Sensor2, Sensor6, Sensor5, dan Sensor8, dengan target health condition ideal berbentuk HC(τ)=1−τ³. Evaluasi estimator RUL menunjukkan bahwa nilai MAPE pada breakpoint 50%, 70%, dan 90% masing-masing sebesar 27,1163%, 24,7927%, dan 20,6347%. Nilai MAEfrac pada ketiga breakpoint tersebut sebesar 13,5278%, 7,4326%, dan 2,0686%, sedangkan RMSEfrac sebesar 16,8350%, 9,7830%, dan 2,6400%. Seluruh nilai MAPE berada di bawah target performansi 30%, sehingga estimator RUL yang dirancang memenuhi kriteria performansi penelitian. Selain itu, dashboard prediksi RUL dikembangkan sebagai media visualisasi hasil model untuk mendukung pemantauan kondisi sistem traksi.
=====================================================================================================================================
The traction system is one of the main subsystems of the LRT Jabodebek because it directly converts electrical energy into mechanical energy to drive the train. Degradation in the traction system may affect operational reliability; therefore, a predictive maintenance approach through Remaining Useful Life (RUL) estimation is required. This study aims to formulate a Health Indicator (HI) and design an RUL estimator for the LRT Jabodebek traction system using a similarity-based polynomial approach. The dataset used in this study is historical run-to-failure traction system data consisting of 32,845 rows, 17 variables, 133 observation IDs, three operational variables, and twelve main sensors. The research stages include train-test data splitting, operational variable evaluation using mean silhouette, operational condition clustering using K-Means, sensor normalization per cluster, trendable sensor selection, HI construction, polynomial degradation modeling, and RUL estimation using residual-based weighted KNN. The results show that Op3, representing motor torque, is selected as the best operational variable for clustering. The selected sensors for HI construction are Sensor3, Sensor2, Sensor6, Sensor5, and Sensor8, with a polynomial ideal health condition target defined as HC(τ)=1−τ³. The model evaluation shows that the MAPE values at the 50%, 70%, and 90% breakpoints are 27.1163%, 24.7927%, and 20.6347%, respectively. The MAEfrac values at the same breakpoints are 13.5278%, 7.4326%, and 2.0686%, while the RMSEfrac values are 16.8350%, 9.7830%, and 2.6400%. All MAPE values are below the 30% performance target, indicating that the proposed RUL estimator satisfies the performance criteria of this study. The MAE, RMSE, MAEfrac, and RMSEfrac results also indicate that the proposed model improves RUL prediction performance compared with the previous study, especially when a longer degradation trajectory is available. In addition, an RUL prediction dashboard is developed as a visualization medium for the model results to support traction system condition monitoring.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Health Indicator, LRT Jabodebek, Remaining Useful Life, Residual-Based Weighted KNN, Similarity-Based Polynomial. Health Indicator, LRT Jabodebek, Remaining Useful Life, Residual-Based Weighted KNN, Similarity-Based Polynomial.
Subjects:	H Social Sciences > HA Statistics > HA31.7 Estimation
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Jamil Uwais
Date Deposited:	04 Aug 2026 09:32
Last Modified:	04 Aug 2026 09:32
URI:	http://repository.its.ac.id/id/eprint/143132

Actions (login required)

View Item