Optimizing Predictive Maintenance in Industrial Machinery with Data Smoothing and Machine Learning

Abstract

Unplanned machinery breakdowns result in significant financial losses, making predictive maintenance essential in industrial operations. This study focuses on fault detection in a single-phase induction motor used in jewelry manufacturing by analyzing vibration data under normal and abnormal conditions. Data was collected using an accelerometer, and three preprocessing techniques—Kalman Filter, Moving Average Filter, and Fast Fourier Transform (FFT)—were applied to reduce noise and improve data quality. Six supervised classification algorithms were evaluated on both raw and preprocessed data. Results demonstrate that preprocessing significantly enhances model accuracy, with the Moving Average Filter enabling Random Forest to achieve the highest accuracy of 99.77%. Kalman Filter also improved model performance, while FFT was particularly beneficial for Logistic Regression. This research highlights the importance of combining machine learning with effective preprocessing to optimize predictive maintenance strategies, reduce downtime, and minimize maintenance costs in industrial environments. By demonstrating the practical implications of these methods, this study contributes to the advancement of reliable fault detection systems for critical machinery components.