Optimizing Svm For Misalignment And Unbalance Detection In Tapered Roller Bearings With Hybrid Kernel And Bayesian Optimization

Sita, Nina Palupi (2025) Optimizing Svm For Misalignment And Unbalance Detection In Tapered Roller Bearings With Hybrid Kernel And Bayesian Optimization. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5007211194-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Pada mesin industri, komponen elemen rol seperti bearing merupakan komponen penting yang mempengaruhi perfoma dari mesin tersebut. Namun, sering kali ditemukan kegagalan mesin yang disebabkan oleh kegagalan pada bearing, seperti unbalanced dan misaligned Supaya tidak mempengaruhi performa mesin industri dan meningkatkan keselamatan pekerja, maka diperlukan pemeliharaan seperti pengujian vibrasi tidak merusak sebagai predictive maintenance. Permodelan yang dibangun dengan SVM, sebuah algoritma Machine Learning dan memiliki peran penting dalam dunia predictive maintenance sebagai algoritma klasifikasi kegagalan pada bearing. Penelitian ini bermaksud untuk menguji model algoritma klasifikasi melalui pendekatan simulasi yang memiliki nilai akurasi yang tinggi. Algoritma seperti PCA untuk mereduksi dimensi, LDA untuk mereduksi dimensi dan separasi kelas, Bayesian Optimization yang memiliki peran untuk mengoptimalkan parameter, hybrid kernel sebagai pembantu SVM untuk melakukan pemisahan kelas non-linear, dan Stratified K-Fold Cross Validation sebagai validasi hasil model, meminimalkan bias, dan mencegah overfitting. Dataset yang digunakan adalah data eksperimen yang diambil oleh mahasiswa dari Lab. Rekayasa Vibrasi dan Sistem Otomotif, Institut Teknologi Sepuluh Nopember pada tapered roller bearing dengan insiasi kecacatan dan diambil data hingga dengan 100 sample untuk setiap kondisi (sehat, misalignment, dan unbalanced) pada bearing merk Timken seri X30304. Dari penelitian ini, didapatkan optimasi parameter C = 0.54, γ = 1.04, ρ = 0.680, dan d = 3, sehingga kernel yang dipilih adalah kernel dengan dominasi polynomial. Model ini mampu mendeteksi jenis kegagalan 3 kondisi dengan akurasi senilai 100% tanpa melakukan misklasifikasi dan matriks parameter lainnya seperti, precision = 100%, recall = 100%, dan F-1 score = 100%. Oleh karena itu, permodelan dapat melakukan klasifikasi dengan sangat baik.
=======================================================================================================================================
In industrial machinery, rolling elements like bearings are crucial components that impact the performance of the machine. However, machine failures caused by bearing issues, such as unbalance, misalignment, or wear, are often encountered. To prevent these from affecting machine performance and to enhance worker safety, maintenance is required, such as non- destructive vibration testing for predictive maintenance. The model built using SVM, a machine learning algorithm, plays an important role in predictive maintenance as a classification algorithm for bearing failures. This study aims to test the classification algorithm model through a simulation approach that yields high accuracy. Algorithms such as PCA for dimensionality reduction, LDA for dimensionality reduction and class separation, Bayesian Optimization to optimize parameters, hybrid kernel to assist SVM in performing non-linear class separation, and Stratified K-Fold Cross Validation for model validation, minimizing bias, and preventing overfitting, are employed. The dataset used is experimental data on tapered roller bearings with defect initiation conducted by student from Vibration Engineering and Automotive System Laboratory, Institut Teknologi Sepuluh Nopember, with up to 100 samples collected for each condition (healthy, misalignment, and unbalanced) on Timken brand bearings from the X30304 series. From this study, the optimized parameters were determined as follows: C = 0.54, γ = 1.04, ρ = 0.680, and d = 3. As a result, the dominance of the kernel is polynomial kernel. The resulting model successfully detected all three failure conditions with an accuracy of 100%, achieving perfect classification performance without any misclassification. Furthermore, the evaluation metrics demonstrated outstanding results, with a precision of 100%, recall of 100%, and F1-score of 100%. Therefore, the proposed model is capable of performing classification tasks with exceptionally high effectiveness.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Machine Learning, SVM, Diagnostic Bearing Failure, Hybrid Kernel
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA169 Reliability (Engineering)
Divisions:	Faculty of Industrial Technology > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Nina Palupi Sita
Date Deposited:	31 Jul 2025 06:12
Last Modified:	31 Jul 2025 06:12
URI:	http://repository.its.ac.id/id/eprint/124610

Actions (login required)

View Item