Contrastive Learning Framework With Cross-Sensor Adaptive Signal Representation for Fault Diagnosis

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Multisource sensor (MS) signal-based mechanical fault diagnosis (MFD) offers improved performance.
• Existing MFD methods struggle with adaptability and generalization when using fewer sensor signals.

Purpose of the Study:

• To propose a general two-stage signal representation contrastive learning fault diagnosis framework (T-SCF).
• To enhance model robustness and data fusion for MFD with varying numbers of sensor signals.

Main Methods:

• An adaptive contrastive algorithm generates contrastive samples and labels for MS signals.
• Supervised contrastive loss (SCL) is employed to differentiate between various fault signals.
• A parallel encoder architecture merges contrasting features from different sensor signals.

Main Results:

• The T-SCF framework demonstrates improved adaptability to different sensor signal configurations.
• The method effectively preserves time-domain properties of sensor signals.
• Validation on multiple datasets confirms the framework's effectiveness.

Conclusions:

• The proposed T-SCF framework offers a robust solution for MFD with adaptable sensor signal usage.
• This approach advances information fusion, fault detection, and classification in MFD.