A Multi-Sensor Fusion and CWT-CNN-BiLSTM-Based Approach for Small-Sample Fault Diagnosis in Rotating Machinery

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rotating machinery is vital in industry but prone to failures under harsh conditions.
• Accurate fault diagnosis is crucial for safety and reliability.
• Challenges include limited training data and sensor noise interference.

Purpose of the Study:

• To enhance fault recognition accuracy for rotating machinery.
• To address challenges of small sample sizes and noisy sensor data.
• To develop a robust and accurate fault diagnosis method.

Main Methods:

• Multi-sensor data fusion and concatenation.
• Continuous Wavelet Transform (CWT) for feature image generation.
• Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) for feature extraction.

Main Results:

• Achieved an average diagnostic accuracy of 99.90%.
• Demonstrated superior performance in fault classification under small-sample conditions.
• Verified method feasibility using a public rotating machinery fault diagnosis database.

Conclusions:

• The proposed CWT-CNN-BiLSTM method offers accurate and robust fault classification.
• Effective for rotating machinery fault diagnosis with limited data and noisy environments.
• Highlights the potential of multi-sensor fusion and deep learning for industrial diagnostics.