Development of Machine Learning Models forPredictive Maintenance in ManufacturingEquipment to Reduce Downtime

Abstract

The Predictive maintenance has become an essential component of modern manufacturing systems, enabling industries to anticipate equipment failures, prevent costly downtime, and improve overall equipment efficiency. This research develops and evaluates machine learning models for predictive maintenance using multi-sensor datasets that capture vibration, temperature, acoustic signals, pressure variations, and power consumption data. A combination of supervised and unsupervised algorithms, including Random Forest, Gradient Boosting, Support Vector Machines, and Autoencoders, is implemented to detect anomalies, classify fault conditions, and estimate degradation trends. Data preprocessing includes noise reduction, normalization, feature extraction, and correlation analysis. Model performance is evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. Experimental results indicate that ensemble learning models deliver excellent predictive accuracy, whereas deep learning–based Autoencoders efficiently capture hidden changes in signal distribution associated with early-stage failures. The integration of these models into manufacturing environments significantly improves maintenance scheduling accuracy, reduces machine downtime, and enhances operational productivity. This study confirms that machine learning–driven predictive maintenance is a scalable and reliable solution that aligns with Industry 4.0 requirements and supports the digital transformation of industrial systems