A fault diagnosis method based on Auxiliary Classifier Generative Adversarial Network for rolling bearing.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearing fault diagnosis is critical for rotating machinery but challenged by noise and limited samples.
• Extracting early fault features is difficult due to complex working conditions and signal interference.
• Existing methods often suffer performance decline with insufficient fault data.

Purpose of the Study:

• To propose a novel fault diagnosis method combining ACGAN and SDAE for improved performance.
• To address challenges of high background noise and small fault sample sizes in bearing diagnostics.
• To enhance the diagnostic accuracy and generalization ability of rotating machinery fault diagnosis systems.

Main Methods:

• A hybrid ACGAN-SDAE model was developed for fault diagnosis.
• Adversarial learning mechanism was employed for alternating optimization of generator and discriminator.
• Stacked Denoising Auto Encoder (SDAE) was integrated for feature extraction and noise reduction.

Main Results:

• The proposed ACGAN-SDAE model demonstrated high diagnostic accuracy with limited fault samples.
• The method exhibited superior adaptation performance across different load domains.
• Enhanced anti-noise performance was observed compared to traditional methods.

Conclusions:

• The ACGAN-SDAE method offers a promising solution for challenging rolling bearing fault diagnosis scenarios.
• The model effectively handles noisy vibration signals and small datasets.
• This approach significantly improves the reliability and robustness of condition monitoring systems.