Comparison of deep learning models for predictive maintenance in industrial manufacturing systems using sensor data.

Area of Science:

• Industrial Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Predictive maintenance (PdM) is crucial for optimizing industrial manufacturing operations.
• Traditional methods often struggle with the complexity and volume of sensor data.
• Deep learning offers advanced capabilities for analyzing complex industrial data.

Purpose of the Study:

• To compare deep learning models for predictive maintenance in industrial manufacturing.
• To evaluate the performance of Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and hybrid variants.
• To identify optimal deep learning architectures for equipment failure prediction and remaining useful life estimation.

Main Methods:

• A framework for data acquisition, preprocessing, and model construction was developed.
• Multiple deep learning architectures, including CNN, LSTM, and CNN-LSTM hybrids, were implemented.
• Experiments were conducted on three distinct industrial datasets.

Main Results:

• The CNN-LSTM hybrid model achieved the highest performance, with 96.1% accuracy and 95.2% F1-score.
• Hybrid models outperformed standalone CNN and LSTM architectures in predicting equipment failures.
• Ablation studies identified critical components influencing model performance.

Conclusions:

• Deep learning, particularly CNN-LSTM hybrids, shows significant potential for revolutionizing PdM in industrial manufacturing.
• Accurate fault prediction and remaining useful life estimation can be achieved using data-driven strategies.
• The findings offer valuable insights for implementing advanced PdM in real-world industrial applications.