Feature Space Transformation for Fault Diagnosis of Rotating Machinery under Different Working Conditions

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning models are used for rotating machinery fault diagnosis.
• Domain shift issues hinder practical implementation due to differing data distributions.
• Collecting diverse failure data is costly and time-consuming.

Purpose of the Study:

• To develop a new transformation method for latent spaces to minimize domain discrepancies.
• To preserve fault attributes during domain adaptation.
• To improve the generalization performance of fault diagnosis models.

Main Methods:

• A novel latent space transformation method using source domain and target normal data.
• Application of spatial attention to learn latent feature spaces.
• Implementation of a 1D Convolutional Neural Network (CNN) Long Short-Term Memory (LSTM) architecture.

Main Results:

• The proposed method demonstrated higher cross-domain diagnostic accuracy compared to existing methods.
• Validated on rolling bearing (CWRU) and heavy machinery gearbox datasets.
• Showcased reliable generalization performance across various operating conditions.

Conclusions:

• The developed method effectively addresses domain shift challenges in rotating machinery fault diagnosis.
• It offers a reliable approach for cross-domain fault diagnosis with easily collectible data.
• The technique shows significant potential for real-world industrial applications.