A universal multi-source domain adaptation method with unsupervised clustering for mechanical fault diagnosis under...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Collecting comprehensive mechanical fault data is challenging.
• Existing methods struggle with incomplete datasets and unknown fault types.
• Domain adaptation (DA) methods often misclassify unknown faults.

Purpose of the Study:

• To develop a universal DA method for mechanical fault diagnosis with incomplete data.
• To address the limitations of existing methods in handling unknown fault classes.
• To leverage multi-source information for improved fault diagnosis.

Main Methods:

• A universal DA approach incorporating unsupervised clustering.
• A composite clustering metric for recognizing shared and unknown classes.
• A class-wise DA algorithm using maximum mean discrepancy.
• Entropy regularization for enhanced clustering.

Main Results:

• The proposed method effectively explores the intrinsic structure of target data.
• It improves the separation of known and unknown fault classes.
• Demonstrated efficacy on three rotating machinery datasets with inadequate monitoring data.

Conclusions:

• The developed universal DA method enhances mechanical fault diagnosis under data scarcity.
• It offers a robust solution for identifying unknown fault types.
• The approach provides a valuable tool for industrial condition monitoring.