Early Prediction of Deterioration and Predictive Modelling in Tapered Roller Bearings Using Support Vector Regression Based on Bearing Health State

Kusumawardhani, Rania Ayu (2025) Early Prediction of Deterioration and Predictive Modelling in Tapered Roller Bearings Using Support Vector Regression Based on Bearing Health State. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5007211001-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Bearing, atau bantalan, merupakan komponen krusial dalam mesin berputar yang digunakan untuk mengurangi gesekan, menopang beban, memastikan gerakan mesin yang mulus, serta mencegah keausan pada berbagai sistem mekanis. Namun seiring berjalannya waktu, kinerja bantalan akan mengalami penurunan akibat faktor-faktor seperti beban mekanis, pelumasan yang tidak tepat, kontaminasi, dan perubahan temperatur yang drastis. Oleh karena itu, deteksi dan prediksi dini kerusakan awal bantalan sangat penting untuk mencegah kegagalan yang tidak terduga, meminimalkan waktu henti, memperpanjang usia pakai peralatan, dan mengoptimalkan jadwal perawatan mesin. Penelitian ini bertujuan untuk mengembangkan model prediktif untuk prediksi kerusakan awal bantalan berdasarkan kondisi kesehatan bantalan dengan menggunakan model Support Vector Regression (SVR) yang dioptimalkan. Data yang digunakan adalah data eksperimental yang diperoleh dari mahasiswa ITS, di mana alat Electrical Discharge Machining (EDM) digunakan untuk membuat kerusakan bantalan dalam empat tipe dengan tujuan menunjukkan tahap kegagalan bantalan dari kondisi sehat hingga gagal. Data getaran yang diperoleh menjalani proses ekstraksi fitur statistik, reduksi dimensi menggunakan Principal Component Analysis (PCA), konstruksi kondisi kesehatan bantalan, optimasi hiperparameter SVR menggunakan optimasi Bayesian, dan evaluasi model. Proses optimasi Bayesian menghasilkan hiperparameter optimal dengan C = 1000, epsilon (ε) = 0.001, dan gamma (γ) = 0.000394. Model SVR yang dioptimalkan kemudian digunakan untuk memprediksi degradasi dini bantalan dengan menggunakan panjang data masukan yang bervariasi, yaitu 150, 160, 170, 180, dan 190 data, guna menyelidiki pengaruh jumlah data terhadap kinerja prediksi. Hasil menunjukkan bahwa peningkatan jumlah data pelatihan memungkinkan model untuk menangkap pola degradasi dengan lebih baik, sehingga menghasilkan prediksi waktu kegagalan yang lebih akurat.
==========================================================================================================================
Bearings are critical components in rotating machinery that are used to reduce friction, support loads, facilitating smooth movement, and preventing wear and tears in a variety of mechanical systems. However, over time, bearings can degrade due to factors such as mechanical stress, improper lubrication, contamination, and thermal expansion. Therefore, early prediction of initial deterioration is essential to prevent unexpected failures, reduce downtime, extend equipment life, and optimize maintenance schedules. This research aims to develop a predictive model for bearing initial deterioration prediction based on bearing health state using optimized Support Vector Regression (SVR) model. The dataset used is an experimental data obtained from ITS student, where Electrical Discharge Machining (EDM) apparatus was used to fabricate bearing damage in four types aims to show the failure stage of the bearing from health to failure. The acquired vibration data underwent a pre-processing process, dimensionality reduction using Principal Component Analysis (PCA), health state construction, SVR hyperparameter optimization using Bayesian optimization, and model evaluation. The Bayesian optimization process yielded optimal hyperparameters with C = 1000, epsilon (ε) = 0.001, and gamma (γ) = 0.000394. The resulting optimized SVR model was then employed to predict early bearing deterioration using varying input data lengths, namely 150, 160, 170, 180, and 190 data points, in order to investigate the influence of data quantity on predictive performance. The results demonstrated that increasing the amount of training data enabled the model to better capture degradation patterns, leading to more precise predictions of failure timing.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bearing Health State, Early Deterioration, Predictive Modelling, Tapered Roller Bearings, Support Vector Regression.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T174 Technological forecasting T Technology > TJ Mechanical engineering and machinery > TJ174 Maintenance and repair of machinery
Divisions:	Faculty of Industrial Technology > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Rania Ayu Kusumawardhani
Date Deposited:	04 Aug 2025 03:43
Last Modified:	04 Aug 2025 03:43
URI:	http://repository.its.ac.id/id/eprint/125118

Actions (login required)

View Item