Synthesizing Rolling Bearing Fault Samples in New Conditions: A Framework Based on a Modified CGAN.

Area of Science:

• Mechanical Engineering
• Machine Learning
• Condition Monitoring

Background:

• Bearings are critical in rotating machinery, and their faults cause significant operational costs and downtime.
• Diagnosing bearing faults is challenging due to the rarity of fault data under diverse operating conditions (loads and speeds).
• Existing data-driven diagnostic tools struggle with insufficient fault data, impacting performance and safety.

Purpose of the Study:

• To develop a novel algorithm for generating synthetic bearing fault data.
• To address the scarcity of fault data in real-world bearing datasets.
• To improve the performance and safety of rotating machinery through enhanced fault diagnosis.

Main Methods:

• A conditional generative adversarial network (CGAN) algorithm was developed.
• The CGAN was trained on normal and actual fault data.
• The algorithm generates synthetic fault data from normal data under target conditions.

Main Results:

• The proposed CGAN method successfully generated fault data for various conditions on a real-world bearing dataset.
• Evaluation using state-of-the-art classifiers and visualization models confirmed the quality of the synthesized data.
• The results demonstrate the algorithm's effectiveness in creating realistic fault data.

Conclusions:

• The developed CGAN algorithm is effective for generating synthetic bearing fault data.
• This approach can significantly enhance data-driven fault diagnosis systems for rotating machinery.
• The method offers a viable solution for overcoming data scarcity in bearing condition monitoring.