Prediksi Remaining Useful Life (RUL) Ball Bearing Berbasis Getaran Menggunakan MBCNN-BiLSTM Untuk Penentuan Kondisi Dan Rekomendasi Penjadwalan Maintenance

Naimah, Nur Lailatun (2026) Prediksi Remaining Useful Life (RUL) Ball Bearing Berbasis Getaran Menggunakan MBCNN-BiLSTM Untuk Penentuan Kondisi Dan Rekomendasi Penjadwalan Maintenance. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5007221030-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Ball bearing merupakan komponen penting pada mesin berputar yang rentan mengalami degradasi selama masa operasi. Kegagalan bearing dapat menyebabkan penurunan performa mesin hingga penghentian operasi yang tidak terencana. Oleh karena itu, prediksi Remaining Useful Life (RUL) diperlukan untuk mendukung penerapan predictive maintenance. Penelitian ini bertujuan memprediksi RUL ball bearing berbasis sinyal getaran menggunakan model MBCNN-BiLSTM, serta menerjemahkan hasil prediksi tersebut menjadi informasi kondisi bearing dan rekomendasi penjadwalan maintenance. Penelitian ini menggunakan dataset PRONOSTIA dari IEEE PHM 2012 Data Challenge yang berisi data getaran bearing hingga kondisi gagal. Tahap preprocessing meliputi transformasi sinyal menggunakan Fast Fourier Transform (FFT), normalisasi data, pembentukan label RUL, dan segmentasi menggunakan time window. Model dilatih menggunakan Asymmetric Loss Function dan dievaluasi menggunakan metrik MAE, MSE, dan RMSE. Hasil prediksi RUL kemudian diterjemahkan ke dalam fase Normal, Warning, dan Critical melalui proses health-stage mapping. Fase tersebut digunakan untuk menentukan maintenance window, yaitu rentang antara awal fase Warning dan sebelum fase Critical. Validasi kondisi dilakukan menggunakan tren RMS dan spectrogram untuk meninjau keterkaitan antara fase kondisi dengan perkembangan degradasi pada sinyal getaran.
Hasil penelitian menunjukkan bahwa model MBCNN-BiLSTM mampu menghasilkan prediksi RUL dengan nilai evaluasi global MAE sebesar 0,0473, MSE sebesar 0,0034, dan RMSE sebesar 0,0585 pada gabungan seluruh window testing. Hasil prediksi RUL juga dapat diterjemahkan menjadi fase kondisi bearing dan rekomendasi maintenance window pada masing-masing bearing pengujian. Validasi menggunakan RMS dan spectrogram menunjukkan bahwa fase kondisi hasil health-stage mapping memiliki hubungan yang logis dengan perkembangan degradasi pada sinyal getaran, meskipun batas transisi fase tidak selalu berimpit langsung dengan lonjakan RMS maupun perubahan spektral yang diskrit. Dengan demikian, pendekatan yang digunakan pada penelitian ini dapat mendukung prediksi RUL sekaligus membantu interpretasi kondisi bearing untuk rekomendasi maintenance berbasis kondisi.
===========================================================================================================================
Ball bearing is an important component in rotating machinery and is prone to degradation during operation. Bearing failure can reduce machine performance and may lead to unplanned downtime. Therefore, RUL prediction is needed to support the implementation of predictive maintenance. This research aims to predict the RUL of ball bearings based on vibration signals using the MBCNN-BiLSTM model and to translate the prediction results into bearing condition information and maintenance scheduling recommendations.
This research used the PRONOSTIA dataset from the IEEE PHM 2012 Data Challenge, which contains run-to-failure vibration data of bearings. The preprocessing stage consisted of signal transformation using Fast Fourier Transform (FFT), data normalization, RUL label generation, and time-window segmentation. The model was trained using an Asymmetric Loss Function and evaluated using MAE, MSE, and RMSE metrics. The RUL prediction results were then translated into three condition phases, namely Normal, Warning, and Critical, through a health-stage mapping process. These phases were used to determine the maintenance window, defined as the interval between the beginning of the Warning phase and before the Critical phase. Condition validation was carried out using RMS trends and spectrogram analysis to examine the relationship between the mapped condition phases and the degradation behavior reflected in the vibration signals.
The results show that the MBCNN-BiLSTM model produced RUL predictions with a global MAE of 0.0473, MSE of 0.0034, and RMSE of 0.0585 across all testing windows. The predicted RUL could also be translated into bearing condition phases and maintenance window recommendations for each tested bearing. Validation using RMS and spectrogram analysis showed that the condition phases obtained from health-stage mapping had a logical relationship with the degradation behavior observed in the vibration signals, although the phase transition boundaries did not always coincide directly with sharp RMS increases or discrete spectral changes. Thus, the proposed approach can support RUL prediction while also improving the interpretability of bearing condition assessment for condition-based maintenance recommendations.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Remaining Useful Life, MBCNN-BILSTM, Maintenance, Ball Bearing, Predictive Maintenance
Subjects:	T Technology > TJ Mechanical engineering and machinery > TJ174 Maintenance and repair of machinery
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Nur Lailatun Naimah
Date Deposited:	03 Aug 2026 08:47
Last Modified:	03 Aug 2026 08:47
URI:	http://repository.its.ac.id/id/eprint/142358

Actions (login required)

View Item