AI-Driven Predictive Maintenance for Smart Manufacturing Systems (Zenodo 2025)

Overview

The AI-Driven Predictive Maintenance for Smart Manufacturing Systems dataset published on Zenodo in November 2025 addresses the critical need for predictive maintenance in Industry 4.0 environments, moving beyond reactive and preventive strategies.

Research Context

• Industry 4.0 smart manufacturing demands high equipment reliability, minimized downtime, and cost-effective operations.
• Traditional maintenance approaches (reactive: fix-when-broken, preventive: scheduled regardless of condition) lead to inefficiencies and unexpected failures.
• AI-driven predictive maintenance leverages real-time sensor data and historical records to forecast failures before they occur.

Dataset Characteristics

• Real-time sensor data: Continuous measurements from industrial equipment including vibration, temperature, pressure, current, voltage, and operational parameters.
• Historical equipment records: Maintenance logs, failure events, repair histories, and component replacement data.
• Equipment health indicators: Derived features representing degradation states and remaining useful life estimations.
• Failure labels: Binary (failure/no-failure) and multi-class (failure type) labels for supervised learning.
• Temporal context: Time-series structure enabling modeling of degradation progression over time.

Machine Learning Framework

• Supervised learning models: Decision trees, random forests, gradient boosting, and neural networks for failure prediction.
• Feature engineering: Statistical features (mean, variance, trends), frequency domain features (FFT), and time-domain features (peaks, RMS) extracted from raw sensor data.
• Prediction horizons: Models trained to forecast failures at different lead times (hours to weeks ahead).
• Performance metrics: Accuracy, precision, recall, F1-score, and ROC-AUC for evaluating prediction quality.

Reported Outcomes

• Significant improvement in operational uptime through timely intervention scheduling.
• Reduction in maintenance costs by avoiding unnecessary preventive maintenance and emergency repairs.
• Enhanced Overall Equipment Effectiveness (OEE) through optimized production planning.
• Transition from reactive/preventive to condition-based predictive maintenance strategies.

Use Cases

• Failure prediction: Training models to forecast when equipment will fail based on current sensor readings and historical patterns.
• Maintenance scheduling: Optimizing maintenance intervals to balance cost, downtime, and failure risk.
• Root cause analysis: Identifying which sensors and conditions are most predictive of specific failure modes.
• Digital twin development: Creating virtual replicas of physical equipment for simulation and what-if analysis.