A Probabilistic Bayesian Parallel Deep Learning Framework for Wind Turbine Bearing Fault Diagnosis.

Area of Science:

• Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Wind turbine reliability is crucial for renewable energy.
• Current deep learning methods struggle with feature extraction and diagnostic confidence for wind turbine bearing faults.

Purpose of the Study:

• To propose a novel probabilistic Bayesian parallel deep learning (BayesianPDL) framework.
• To enhance the accuracy and confidence of fault diagnosis in wind turbine bearings.

Main Methods:

• Developed a parallel deep learning (PDL) framework to address feature extraction challenges.
• Transformed deterministic weights and biases into probability distributions for an uncertainty-aware approach.
• Applied the BayesianPDL framework to gearbox output shaft bearing fault data.

Main Results:

• Achieved a diagnostic accuracy of 99.14% for wind turbine bearing faults.
• Demonstrated higher confidence in diagnostic results compared to existing methods.
• Validated the framework's effectiveness using real-world wind farm data.

Conclusions:

• The BayesianPDL framework offers a robust solution for reliable wind turbine bearing fault diagnosis.
• The uncertainty-aware approach significantly improves diagnostic confidence.
• This technology presents unique advantages for maintaining wind turbine operational integrity.