A Dual-Source Evidence-Driven Semi-Supervised Belief Rule Base for Fault Diagnosis.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Industrial Engineering

Background:

• Complex industrial systems require accurate fault diagnosis.
• Labeled data for training belief rule base (BRB) models is scarce and costly.
• Unlabeled data presents challenges due to uncertain pseudo-label quality.

Purpose of the Study:

• To propose a semi-supervised BRB method (SS-BRB) for fault diagnosis.
• To effectively utilize abundant unlabeled data while maintaining model interpretability.
• To enhance parameter optimization for BRB models in data-scarce scenarios.

Main Methods:

• Developed a dual-source evidence-driven pseudo-labeling mechanism.
• Integrated local similarity with global BRB inference for pseudo-label reliability.
• Introduced entropy and feature distance factors for adaptive pseudo-label confidence adjustment.

Main Results:

• The SS-BRB method demonstrated strong diagnostic performance on gearbox and WD615 diesel engine datasets.
• Achieved high accuracy, robustness, and generalization with limited labeled data.
• Effectively improved pseudo-label quality and reduced the impact of noisy samples.

Conclusions:

• The proposed SS-BRB method offers an effective solution for fault diagnosis with limited labeled data.
• The dual-source evidence-driven approach enhances the reliability of semi-supervised learning in BRB models.
• This method preserves the interpretability and transparency of BRB models in complex industrial systems.