A Novel Study on a Generalized Model Based on Self-Supervised Learning and Sparse Filtering for Intelligent Bearing...

Area of Science:

• Engineering
• Computer Science

Background:

• Deep learning methods are widely used in fault diagnosis.
• These methods often require large labeled datasets, limiting their generalization ability across different scenarios.

Purpose of the Study:

• To propose a novel generalized model for fault diagnosis that overcomes data limitations.
• To enhance diagnostic performance and generalization ability using self-supervised learning and sparse filtering.

Main Methods:

• Developed a two-stage generalized model based on self-supervised learning and sparse filtering (GSLSF).
• Stage 1: Designed self-supervised learning pretext tasks and pseudo-labels for pre-training a sparse filtering model.
• Stage 2: Implemented knowledge transfer, extracted deep fault features using sparse filtering, and applied softmax regression for failure classification.

Main Results:

• The proposed GSLSF method significantly enhances the model's diagnostic performance.
• Demonstrated improved generalization ability even with limited training data.
• Validated the method's effectiveness through fault diagnosis on two bearing datasets.

Conclusions:

• The GSLSF model offers a robust solution for fault diagnosis with limited labeled data.
• Self-supervised learning combined with sparse filtering improves diagnostic accuracy and adaptability.
• The approach is effective for bearing fault diagnosis applications.