Explainable AI for Bearing Fault Prognosis Using Deep Learning Techniques.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing failures are critical in rotary machine health monitoring, impacting operational efficiency and maintenance planning.
• Automated detection of bearing defects using vibration signals offers a promising alternative to manual inspection.
• Deep learning techniques have shown significant potential in advancing automatic defect identification.

Purpose of the Study:

• To propose a novel approach for identifying bearing defects and their degradation levels under variable shaft speeds.
• To enhance machine health monitoring systems with accurate and robust fault prediction capabilities.
• To integrate explainable AI for better understanding of the deep learning model's decision-making process.

Main Methods:

• Vibration signals are pre-processed using the Short-Time Fourier Transform (STFT) to generate spectrograms.
• A Convolutional Neural Network (CNN) model, specifically VGG16, is employed for feature extraction and health status classification.
• Remaining Useful Life (RUL) prediction is performed using regression techniques.
• Explainable AI (LIME) is utilized to interpret the CNN's output and identify critical image regions.

Main Results:

• The proposed method demonstrates high accuracy and robustness in detecting bearing faults and quantifying degradation.
• Spectrogram representation combined with VGG16 CNN effectively captures bearing defect characteristics.
• The integration of RUL prediction provides valuable insights for proactive maintenance scheduling.
• LIME successfully visualizes the regions within spectrograms that are most influential for fault classification.

Conclusions:

• The developed deep learning approach offers a reliable solution for automated bearing fault diagnosis and prognosis.
• This method contributes to improved machine health monitoring by enabling early detection and accurate assessment of bearing conditions.
• The study highlights the effectiveness of combining STFT, VGG16 CNN, regression, and LIME for comprehensive bearing health analysis.