A Generative Adversarial Network Based a Rolling Bearing Data Generation Method Towards Fault Diagnosis.

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Machine learning for bearing fault diagnosis requires substantial data.
• Insufficient data hinders the effectiveness of traditional diagnostic models.
• Generative models offer a potential solution for data augmentation.

Purpose of the Study:

• To propose a novel generative adversarial network (GAN) based method for generating rolling bearing data.
• To address the challenge of insufficient data in bearing fault diagnosis.
• To enhance the accuracy and efficiency of fault diagnosis systems.

Main Methods:

• A generative adversarial network (GAN) model was developed and trained adversarially.
• Rolling bearing operational data was analyzed and segmented into three periods based on time-domain characteristics.
• The generated data was compared with real data in both time and frequency domains to validate similarity.

Main Results:

• The proposed GAN effectively generated pseudoreal rolling bearing data.
• The generated data demonstrated high similarity to actual operational data in time and frequency domains.
• The method shows promise for improving fault diagnosis with limited datasets.

Conclusions:

• Generative adversarial networks (GANs) are effective for augmenting limited rolling bearing datasets.
• This approach can significantly improve the performance of machine learning-based fault diagnosis systems.
• The proposed method offers a viable solution for real-world applications with data scarcity.