A novel semisupervised classification framework for coupling faults in hot rolling mill process.

Area of Science:

• Industrial Process Control
• Machine Learning Applications
• Fault Diagnosis Systems

Background:

• Coupling faults present significant challenges in industrial settings due to their inherent uncertainties and diversity.
• Accurate classification of these faults is crucial for timely intervention and maintaining operational stability.
• Existing fault classification methods often struggle with the complexity and variability of coupling faults.

Purpose of the Study:

• To develop a novel semi-supervised classification scheme for addressing coupling fault diagnosis.
• To provide operators with reliable diagnostic decisions for immediate remedial actions.
• To enhance the robustness and accuracy of fault classification in industrial applications.

Main Methods:

• Proposed a semi-supervised classification scheme integrating adaptive classification and multi-task feature selection.
• Developed an approach for adaptive and automatic learning of the number of classifications.
• Enabled the capture of common and specific features between single and associated coupling faults.

Main Results:

• Demonstrated the effectiveness of the proposed scheme through a case study on a hot rolling mill process.
• Achieved more successful classification results compared to traditional methods in two distinct cases.
• Validated the scheme's ability to improve classification performance by capturing relevant fault features.

Conclusions:

• The novel semi-supervised classification scheme offers a promising solution for complex coupling fault diagnosis.
• The adaptive and feature-centric approach enhances diagnostic decision-making capabilities for online operators.
• This research contributes to advancing fault classification techniques in industrial automation and control.