Widiarna, Gede Bagus Nugratama (2025) Analisis Komparatif Kinerja Ekstraksi Fitur Frequency Domain Terhadap Time Domain Dalam Pembangunan Health Index Rolling Element Bearing Untuk Pengukuran Remaining Useful Life Berbasis LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5007211170-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5007211170-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy |
Abstract
Rolling Element Bearing (REB) merupakan komponen kritis pada mesin industri yang rentan mengalami degradasi dan kegagalan, sehingga memerlukan sistem pemeliharaan prediktif yang efektif untuk meminimalkan downtime. Penelitian ini bertujuan untuk melakukan analisis komparatif terhadap kinerja ekstraksi fitur dari domain frekuensi (frequency-domain) dan domain waktu (time-domain) dalam pembangunan Health Index (HI) untuk memprediksi Remaining Useful Life (RUL) pada REB. Metodologi penelitian dimulai dengan akuisisi data getaran dari dataset publik, yaitu Intelligent Maintenance System (IMS) dan PRONOSTIA, yang mencakup data dari kondisi sehat hingga terjadi kegagalan. Data mentah tersebut kemudian diproses melalui tahap ekstraksi fitur untuk mendapatkan sembilan fitur statistik pada domain waktu dan delapan fitur pada domain frekuensi. Fitur-fitur yang dihasilkan kemudian diseleksi berdasarkan nilai trendability menggunakan korelasi Spearman. Health Index (HI) kemudian dibangun dari fitur-fitur terpilih untuk melatih model prediksi berbasis Long Short Term Memory (LSTM), dengan seluruh proses komputasi didukung oleh Google Colab. Hasil pada dataset IMS dengan domain frekuensi menunjukkan performa model yang memuaskan dengan melebihi nilai batas evaluasi, dimana nilai korelasi Spearman rata-rata 0.9023, RMSE 0.0383, dan MAE 0.0113 untuk prediksi HI. Dalam perbandingan, fitur domain frekuensi secara konsisten terbukti lebih unggul daripada domain waktu dengan metrik error yang lebih rendah pada prediksi HI maupun RUL, meskipun membutuhkan waktu komputasi yang sedikit lebih lama. Untuk menguji generalisasi, metode ini divalidasi pada set data PRONOSTIA dan menunjukkan performa yang robust dengan nilai RMSE 0.0559 dan MAE 0.0339 yang mengonfirmasi efektivitas pendekatan yang diusulkan untuk prognosis kondisi mesin.
========================================================================================================================================
Rolling Element Bearing (REB) is a critical component in industrial machinery that is susceptible to degradation and failure, thus requiring an effective predictive maintenance system to minimize downtime. This study aims to conduct a comparative analysis of the performance of feature extraction from the frequency-domain and the time-domain in the development of a Health Index (HI) to predict the Remaining Useful Life (RUL) of REBs. The research methodology begins with the acquisition of vibration data from public datasets, namely the Intelligent Maintenance System (IMS) and PRONOSTIA, which cover data from a healthy condition to failure. The raw data is then processed through a feature extraction stage to obtain nine statistical features in the time-domain and eight features in the frequency-domain. The resulting features are then selected based on their trendability value using Spearman correlation. A Health Index (HI) is subsequently constructed from the selected features to train a prediction model based on Long Short-Term Memory (LSTM), with the entire computational process supported by Google Colab. The results on the IMS dataset with the frequency-domain show satisfactory model performance, exceeding the evaluation thresholds, with an average Spearman correlation of 0.9023, an RMSE of 0.0383, and an MAE of 0.0113 for HI prediction. In comparison, frequency-domain features consistently proved to be superior to time-domain features with lower error metrics in both HI and RUL predictions, despite requiring slightly longer computation time. To test for generalization, the method was validated on the PRONOSTIA dataset and demonstrated robust performance with an RMSE of 0.0559 and an MAE of 0.0339, confirming the effectiveness of the proposed approach for machine condition prognosis.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Frekuensi, Health Index, RUL, Frequency, Health Index, RUL |
| Subjects: | Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis |
| Depositing User: | Gede Bagus Nugratama Widiarna |
| Date Deposited: | 02 Aug 2025 08:22 |
| Last Modified: | 02 Aug 2025 08:22 |
| URI: | http://repository.its.ac.id/id/eprint/125802 |
Actions (login required)
 |
View Item |