Enhanced Failure Detection in Induction Motors Using Supervised Learning and Data Smoothing Techniques

Production machinery failures often cause significant financial losses, and traditional maintenance methods may struggle to mitigate these issues effectively. Predictive maintenance offers a modern solution by forecasting failures before they occur. This study applies supervised learning algorithms to pre-collected vibration data from a single-phase induction motor under normal and simulated abnormal working conditions. Data smoothing techniques such as Moving Average Filter, Kalman Filter, and FFT were employed to enhance data quality. Supervised algorithms including Logistic Regression, Decision Tree, Random Forest, SVM, Naive Bayes, and kNN were implemented and evaluated. Performance metrics such as accuracy, precision, recall, and F1 score were used to compare these algorithms. Results indicate that data smoothing significantly improves model performance, with the Moving Average Filter achieving the highest accuracy for Random Forest at $\mathbf{9 9. 7 7 \%}$. These findings highlight the potential of advanced machine learning techniques combined with data preprocessing for predictive maintenance in industrial settings. By reducing downtime and maintenance costs, these models enhance operational reliability and efficiency. This study contributes to the growing knowledge on applying machine learning for predictive maintenance, offering practical insights for improving the sustainability and productivity of manufacturing processes.