A novel fault diagnosis method based on EEWT-EWLTSA and improved deep ELM

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Gearbox fault diagnosis is critical for industrial equipment reliability.
• Traditional methods often struggle with complex vibration signals and noise.
• Advanced signal processing and machine learning are needed for accurate fault detection.

Purpose of the Study:

• To develop a novel and highly accurate gearbox fault diagnosis method.
• To enhance feature extraction and dimensionality reduction for vibration signals.
• To improve the generalization and anti-interference capabilities of diagnostic models.

Main Methods:

• Extensive Empirical Wavelet Transform (EEWT) for signal decomposition based on amplitude modulation.
• Entropy-Weighted Local Tangent Space Alignment (EWLTSA) for feature dimensionality reduction.
• Improved Deep Extreme Learning Machine (IDELM) with regularization and optimized training parameters.

Main Results:

• The proposed EEWTEWLTSA-IDELM method achieved 99.5% accuracy in Case 1 and 98.5% in Case 2.
• Demonstrated superior performance compared to multiple alternative methods (e.g., EEWTLTSA-IDELM, MDCNN-LSTM, VMD-DNN).
• Exhibited strong anti-interference capability under various noise levels and signal-to-noise ratios.

Conclusions:

• The EEWTEWLTSA-IDELM method offers a highly effective solution for gearbox fault diagnosis.
• The combination of EEWT, EWLTSA, and IDELM significantly improves diagnostic accuracy and robustness.
• The proposed method is well-suited for real-world applications, even in noisy environments.