Deep Learning Enhanced Predictive Maintenance Framework Using Industrial Internet of Things Sensors for Smart Manufacturing Systems

Manufacturing systems generate massive sensor data, yet transforming this information into actionable maintenance insights remains challenging due to traditional threshold-based approaches suffering from high false positive rates and insufficient advance warning. This study developed and validated a hybrid deep learning framework combining convolutional neural networks for spatial feature extraction with long short-term memory networks for temporal pattern recognition in smart manufacturing environments. The methodology involved collecting 18 months of operational data from 127 industrial machines across three Saudi Arabian facilities, encompassing 1.2 million sensor readings and 3,452 maintenance events from vibration, temperature, current, pressure, and acoustic sensors. The hybrid CNN-LSTM framework achieved 94.3% accuracy in predicting equipment failures 48 hours in advance with a 2.1% false positive rate, demonstrating statistically significant superiority over Random Forest (15.4 percentage point improvement), Support Vector Machines (15.1 percentage points), and threshold-based monitoring (25.9 percentage points). Significance was assessed on paired predictions using McNemar’s test (two-sided, alpha = 0.05) with Bonferroni correction across model comparisons; improvements were significant (p < 0.001). Cross-facility validation confirmed robust generalization capabilities. Economic analysis revealed 28% maintenance cost reduction, 37% unplanned downtime decrease, and 15% overall equipment effectiveness improvement, yielding 183% return on investment with a 6.5-month payback period. These findings demonstrate the practical viability and substantial economic benefits of hybrid deep learning approaches for industrial predictive maintenance, establishing a foundation for enhanced operational efficiency in Industry 4.0 manufacturing systems.

Hali tarjima qilinmagan