Digital Twin-Enabled Predictive Maintenance of Industrial Machinery Using Machine Learning and Real-Time Sensor Analytics

Predictive maintenance has become an important component of intelligent industrial systems by enabling machinery condition assessment before critical failures occur. This study developed a data-driven digital-twin framework integrating machine learning with temporal sensor analytics for industrial machinery health monitoring. The analysis used 1,000 chronologically ordered observations comprising vibration, temperature, pressure, and machinery fault states. The dataset included normal operation (60.9%), bearing faults (30.3%), and overheating (8.8%). Sensor measurements were standardized and transformed using 128-observation sliding windows, producing 44 multivariate temporal samples across three sensor channels. Shapelet-based temporal feature extraction was incorporated to capture condition-related patterns, while Logistic Regression and Random Forest were considered for machine-state classification. The results demonstrated substantial variability in the dynamic sensor measurements but weak individual relationships with machinery fault status. Vibration exhibited the highest absolute sensor–fault correlation (r = −0.032), followed by pressure (r = 0.009) and temperature (r = −0.008), indicating that individual sensor measurements alone provided limited linear discrimination between operating states. The proposed framework therefore emphasizes multivariate temporal behavior for updating digital-twin machinery states and supporting condition-based maintenance decisions. The findings demonstrate the feasibility of combining temporal sensor processing, machine learning, and data-driven digital twins for industrial condition monitoring, while further validation using larger run-to-failure datasets is required.