Digital Twin–Based Predictive Maintenance in Industry 4.0 and Industry 5.0: An Empirical Study Using Machine Learning

Abstract

The rapid digital transformation of industrial manufacturing has introduced digital twin (DT) technology as a cornerstone for intelligent maintenance systems within Industry 4.0 and Industry 5.0 environments. This study presents an empirical investigation of digital twin–based predictive maintenance (PdM) using machine learning algorithms on real-world industrial sensor data. The research evaluates the performance of Random Forest, Gradient Boosting, Support Vector Machine, and Artificial Neural Networks in predicting equipment failures and optimizing maintenance strategies. A dataset of over 10,000 machine operation records, including temperature, vibration, pressure, and operational cycles, was analyzed to assess predictive accuracy and operational impact. Results indicate that Random Forest achieved the highest predictive accuracy (92.4%), while digital twin integration reduced unplanned machine downtime by approximately 28% compared to reactive maintenance approaches. The study highlights vibration and temperature as the most critical indicators of machine failure, demonstrating the importance of sensor-driven monitoring in predictive maintenance. Findings further show that digital twin–enabled predictive maintenance supports proactive maintenance planning, human-centered decision-making, and operational efficiency, bridging the gap between Industry 4.0 automation and Industry 5.0 human–AI collaboration.