Class-Consistent Matching Attention Wavelet Networks for Partial Transfer Intelligent Diagnosis.

Area of Science:

• Machine Learning
• Industrial Fault Diagnosis
• Domain Adaptation

Background:

• Existing domain adaptation (DA) methods achieve high accuracy in fault diagnosis.
• Partial DA (PDA) is crucial for industrial scenarios where target labels are a subset of source labels.
• Current PDA methods struggle to differentiate common and private samples, leading to negative transfer.

Purpose of the Study:

• To propose a novel method for partial domain adaptation fault diagnosis.
• To address the challenge of separating common and private samples in PDA.
• To improve the accuracy and robustness of fault diagnosis in industrial settings.

Main Methods:

• Class-consistency matching using label consensus score to identify common and private samples.
• Parameter-free cosine attention wavelet blocks (PCAWBs) for learning complementary spatial and frequency domain features.
• Enriching domain-invariant features extracted by a shared encoder.

Main Results:

• The proposed method effectively identifies common and private samples.
• PCAWBs enhance feature representation by integrating spatial and frequency information.
• Significant performance improvement over state-of-the-art PDA fault diagnosis approaches demonstrated on a real motor system.

Conclusions:

• The developed class-consistency matching and PCAWBs offer a robust solution for PDA fault diagnosis.
• The method mitigates negative transfer by accurately distinguishing sample types.
• This approach advances the field of intelligent fault diagnosis in industrial applications.