Pengembangan Model Diagnosis Untuk Deteksi Kegagalan Mesin Menggunakan Algoritma Random Forest Dalam Klasifikasi Pembelajaran Mesin Berdasarkan Uji Non-Destructive Vibrasi

Nabil, Muhammad Dzakwan (2024) Pengembangan Model Diagnosis Untuk Deteksi Kegagalan Mesin Menggunakan Algoritma Random Forest Dalam Klasifikasi Pembelajaran Mesin Berdasarkan Uji Non-Destructive Vibrasi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5007201020-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (4MB) | Request a copy

Abstract

Bearing memiliki peran penting dalam industri karena memberikan dukungan mesin yang efektif dan tanpa suara. Bearing mengurangi gesekan di berbagai industri dan peralatan, seperti gearbox, rotating shaft, dan dryer sehingga memudahkan gerakan. Pengujian getaran non-destructive digunakan untuk melakukan predictive maintenance berdasarkan data vibrasi untuk mengetahui jenis kegagalan. Model Random Forest, jenis pembelajaran mesin, semakin populer karena lebih akurat dalam menemukan kegagalan pada sistem bearing karena cocok untuk kasus multiklasifikasi dengan dataset yang banyak. Penelitian ini bertujuan untuk memvalidasi model ini dengan data melalui simulasi dan eksperimental, serta untuk membentuk sistem yang dapat diandalkan untuk mendeteksi kegagalan pada sistem bearing. Pada penelitian ini, penulis membangun model diagnosis kegagalan menggunakan algoritma Random Forest dengan tambahan metode Recursive Feature Elimination – Cross Validation pada feature selection dan K-Folds Cross Validations untuk memvalidasi hasil model yang dibangun serta mengurangi bias dan mencegah adanya overfitting. Proses utama menggunakan data eksperimental secara langsung pada test rig melalui vibration meter tools Wilcoxon MAC800 type dengan mengambil sampel sebanyak 100 kali untuk setiap variasi kegagalan yaitu imbalancing, healthy, misalignment, dan outer ring pada objek bearing merk Timken seri X30304. Selanjutnya, algoritma yang telah dibangun akan diuji dengan subset data k=10 sebagai proses cross validation. Dari penelitian ini, model mampu mendeteksi jenis kegagalan dengan nilai akurasi diatas 95%. Dimana skor precision, recall, dan F1-score menunjukkan nilai rata-rata di 98.6%,98.5%, dan 98.4%, lebih besar dari penelitian terdahulu di 94.07%.
========================================================================================================================
Bearings are essential components in the industrial sector as they offer efficient and noiseless support for machines. Bearings mitigate friction in several sectors and equipment, including gearboxes, rotating shafts, and dryers, thereby facilitating motion. Vibration testing that does not cause damage is employed for predictive maintenance by analyzing vibration data to identify different sorts of failures. The Random Forest model, a machine learning algorithm, is gaining popularity due to its high accuracy in detecting problems in bearing systems. This is attributed to its effectiveness in handling multiclassification scenarios with big datasets. The objective of this study is to test the proposed model by using simulation and experimental approaches with data. The purpose is to build a dependable system for detecting faults in bearing systems. The author of this study constructs a failure diagnosis model by employing the Random Forest algorithm. Additionally, the Recursive Feature Elimination – Cross Validation technique is utilized for feature selection, and K-Folds Cross Validation is employed to validate the model outcomes, minimize bias, and prevent overfitting. The primary procedure utilizes empirical data obtained from a test rig using Wilcoxon MAC800 type vibration meter equipment. The approach involves collecting 100 samples for each failure variant, namely imbalance, healthy condition, misalignment, and outer ring failure, specifically on Timken brand bearings series X30304. Afterwards, the constructed method will undergo testing using subsets of data with k=10, as part of a cross-validation. From this research, the model is capable of detecting failure types with an accuracy rate above 95%. The precision, recall, and F1-score metrics show average values of 98.6%, 98.5%, and 98.4%, respectively, which are higher than the previous research at 94.07%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Random Forest, RFECV, K-Folds CV, Bearing Fault Mode, Diagnostic, Diagnosis, Kegagalan Bearing
Subjects:	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:	Muhammad Dzakwan Nabil
Date Deposited:	12 Aug 2024 05:16
Last Modified:	12 Aug 2024 05:16
URI:	http://repository.its.ac.id/id/eprint/113812

Actions (login required)

View Item