Rotating Machine Fault Diagnosis Using Neural Network With Utocorrelation And Synthetic Minority Over-Sampling Technique For Imbalanced Data Training

Nur, Muhammad Athillah (2025) Rotating Machine Fault Diagnosis Using Neural Network With Utocorrelation And Synthetic Minority Over-Sampling Technique For Imbalanced Data Training. Other thesis, INSTITUT TEKNOLOGI SEPULUH NOPEMBER.

Text 5007211014-UNDERGRADUATE_THESIS.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

In industrial sectors such as petroleum and manufacturing, unplanned machine failures primarily due to misalignment, unbalance, and bearing faults can cause major operational and financial disruptions. This study proposes a vibration based machine fault diagnosis model integrating autocorrelation, Synthetic Minority Oversampling Technique (SMOTE), and a neural network. Vibration signals were collected from four machine conditions (normal, unbalance, misalignment, bearing fault), but class imbalance was significant, with the normal condition comprising only 3.51% of the dataset. SMOTE was applied to balance the data, while FFT-autocorrelation reduced noise and highlighted periodic patterns. The preprocessed data were divided into training and validation sets (75:25) and used to train a neural network with two hidden layers (128 neurons each). The enhanced model achieved 99.79% accuracy, precision, recall, and F1-score significantly outperforming the baseline model without SMOTE and autocorrelation 78.37% accuracy. Cross-validation 10-fold confirmed model stability with average result 99,76%, and UMAP visualization showed clear class separation. The improved model also demonstrated aligned training and validation curves, indicating no overfitting. This study confirms that combining SMOTE, FFT-autocorrelation, and neural networks effectively improves diagnostic accuracy and robustness in detecting rotating machinery faults
=================================================================================================================================
In industrial sectors such as petroleum and manufacturing, unplanned machine failures primarily due to misalignment, unbalance, and bearing faults can cause major operational and financial disruptions. This study proposes a vibration based machine fault diagnosis model integrating autocorrelation, Synthetic Minority Oversampling Technique (SMOTE), and a neural network. Vibration signals were collected from four machine conditions (normal, unbalance, misalignment, bearing fault), but class imbalance was significant, with the normal condition comprising only 3.51% of the dataset. SMOTE was applied to balance the data, while FFT-autocorrelation reduced noise and highlighted periodic patterns. The preprocessed data were divided into training and validation sets (75:25) and used to train a neural network with two hidden layers (128 neurons each). The enhanced model achieved 99.79% accuracy, precision, recall, and F1-score significantly outperforming the baseline model without SMOTE and autocorrelation 78.37% accuracy. Cross-validation 10-fold confirmed model stability with average result 99,76%, and UMAP visualization showed clear class separation. The improved model also demonstrated aligned training and validation curves, indicating no overfitting. This study confirms that combining SMOTE, FFT- autocorrelation, and neural networks effectively improves diagnostic accuracy and robustness in detecting rotating machinery faults

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autocorrelation, Bearing Fault, Diagnostic, Misalignment, Neural Network, SMOTE, Unbalance.
Subjects:	Q Science > QA Mathematics > QA76.76.S64 Software maintenance. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Athillah Nur
Date Deposited:	28 Jul 2025 06:14
Last Modified:	28 Jul 2025 06:14
URI:	http://repository.its.ac.id/id/eprint/122049

Actions (login required)

View Item